Space Launch and Satellites: A Practical Reference

A small launch vehicle is a dedicated rocket sized to put a few hundred kilograms into low Earth orbit for one customer. Rocket Lab's Electron carries about 300 kg to LEO and roughly 200 kg to a 500 km sun-synchronous orbit; Firefly Alpha is quoted at about 1,030 kg to LEO and 630 kg to SSO; Vega C sits above both at around 2,300 kg to a polar orbit. What the customer is buying is not lift, since a rideshare seat can carry the same satellite. It is control: the orbit, the inclination, the local time of the ascending node, and the launch date are all yours, and nothing else on the vehicle can delay you. That control costs roughly four times as much per kilogram as sharing a ride, which is the whole economics of the category.

The strength is schedule and orbit control, and for a handful of customers that is worth a large premium. A dedicated vehicle also flies payloads that a rideshare provider will not take, whether for classification reasons, hazardous propellants, or an orbit nobody else is going to. The weakness is price per kilogram, at roughly $25,000–30,000/kg against $6,000–7,000/kg on a Transporter seat. The failure mode is not technical, it is commercial: the addressable market for customers who will pay 4× is far smaller than the number of vehicles built to serve it. Dozens of small-launch companies were funded between 2015 and 2022, and two or three now fly regularly. Several reached orbit, could not fill a manifest at a price that covered fixed costs, and shut down or pivoted to larger vehicles.

Buy a dedicated small launch when the mission needs a specific orbit plane or a specific LTAN that no rideshare is flying to, when the payload cannot share a vehicle for security or safety reasons, or when a schedule slip costs more than the ticket. Responsive call-up is the clearest case: the Space Force's Victus Nox mission launched 27 hours after the order to go, and no shared mission can do that. If your satellite can accept a standard SSO drop-off and a date that may move by months, take a rideshare seat instead and put the savings into the spacecraft. If you need a specific orbit but not a specific date, the usual winner is a rideshare seat plus an orbital transfer vehicle, which reaches most LEO destinations for a fraction of dedicated launch and is now the default answer for payloads under a few hundred kilograms.

Electron carries about 300 kg to LEO and roughly 200 kg to a 500 km SSO · Firefly Alpha about 1,030 kg to LEO and 630 kg to SSO · list price $7.5–8.5M per Electron launch · roughly $25,000–30,000/kg against $6,000–7,000/kg on rideshare · Victus Nox launched 27 hours after the call-up order · dozens of small-launch companies funded since 2015, two or three flying regularly.

US launches need an FAA Part 450 vehicle operator license, which covers the vehicle, the trajectory, and the public-safety analysis, plus range access at a federal site (Cape Canaveral, Vandenberg, Wallops) or a licensed commercial spaceport. Vehicle telemetry and flight termination run on range-coordinated S-band and are licensed separately from anything the payload transmits, so a launch license buys you nothing toward an FCC or NOAA payload authorization. The vehicle itself is ITAR-controlled, which is why Rocket Lab operates its New Zealand pad under a bilateral Technology Safeguards Agreement layered on top of New Zealand's own Outer Space and High-altitude Activities Act. If you are flying a foreign payload on a US vehicle, or a US payload from foreign soil, budget months for the export licensing rather than weeks.

Rocket Lab Electron (the only Western small launcher with a real flight rate, and the vehicle that flew Victus Nox for the Space Force), Firefly Alpha, Avio's Vega C from Kourou, ISRO's SSLV, and China's Kuaizhou and Ceres-1. The consolidated side of the ledger is just as instructive: Virgin Orbit reached orbit and then went bankrupt in 2023, Astra retired Rocket 3.3 after repeated failures and went private, and Rocket Lab itself is moving up-market to the medium-lift Neutron.

The fixed costs of a launch company (pad, range, integration, a standing engineering team) barely change with vehicle size, so a small launcher spreads the same overhead across a much smaller ticket. That is why the category needs high flight rate to work at all and why almost nobody achieved it. Rocket Lab is the exception, and its own strategy tells you how the economics run: it is building Neutron, a much larger vehicle, because the margin on a $7.5M launch is thin even at a good cadence. If you are underwriting a small-launch business, the question is not whether the rocket works. It is how many customers per year genuinely need a dedicated orbit, what they will pay, and whether a rideshare-plus-tug combination is going to take that customer away at a third of the price.

6 minWhy The Electron Rocket May Be Cheapest Way To Get To SpaceScott Manley · 500k+ views

3 minDedicated Launch with ElectronRocket Lab · 10k+ views

3 minRocket Lab | Constellation Deployment with ElectronRocket Lab · 10k+ views

A Small Launch per Month? - 2022 Edition of the Annual Industry Survey (Small Satellite Conference) · Commercial Space Launch and Reentry Regulations: Overview and Select Issues (Congressional Research Service)

A rideshare launch sells you a seat on a rocket whose mission somebody else defined. SpaceX's Transporter missions fly a Falcon 9 to a sun-synchronous orbit around four times a year carrying a hundred or more payloads on a stack of ESPA-class adapters and CubeSat deployers, and the Bandwagon missions do the same thing to a mid-inclination orbit. Pricing is published rather than negotiated: roughly $350,000 buys the first 50 kg, and mass above that runs about $6,000–7,000/kg as of 2026. Every payload rides to the same drop-off, deploys inside the same short sequence, and slips whenever the mission slips. That standardization is what took launch off the critical path for small satellites, and it is the reason a smallsat industry exists at the scale it does.

The strength is price and availability. At $6,000–7,000/kg a rideshare seat is roughly a quarter the cost of a dedicated small launcher, the price list is public, and there is another mission within a few months if you miss one. The weakness is that you control nothing except your own hardware. You take the mission's orbit, its inclination, its local time of the ascending node, and its schedule, and a delay caused by the primary payload is your delay too. The failure mode worth planning for is not the rocket. It is arriving in an orbit that is close to what you wanted but not close enough, with no propulsion to fix it, which turns a working satellite into a partly useful one for its whole life. The second failure mode is identification: when 143 objects separate within an hour, as they did on Transporter-1, operators can spend days to weeks working out which catalog object is theirs, and you cannot command a satellite you cannot find.

Take a rideshare seat as the default for anything under a few hundred kilograms going to LEO or SSO. It is the right answer whenever the mission tolerates a standard drop-off orbit and a launch date that can move by months, which covers most technology demonstrations, most CubeSat science, and most constellation build-out where the satellites have their own propulsion and can spread themselves out afterward. Do not buy a seat if you need a specific plane or a specific LTAN that no Transporter flies to, if the payload cannot legally or physically share a vehicle, or if being late costs more than the launch does; buy a dedicated small launch instead. If you need the orbit but not the date, the usual winner is a rideshare seat plus an orbital transfer vehicle, which adds roughly $0.5–3M and still lands well under a dedicated ticket.

Roughly $350,000 for the first 50 kg on a Transporter mission · about $6,000–7,000/kg above that as of 2026 · 100+ payloads on a single mission, and 143 on Transporter-1 · about four Transporter missions a year to SSO, plus Bandwagon flights to mid-inclination · roughly $25,000–30,000/kg on a dedicated small launcher for comparison · $0.5–3M for a last-mile tug on top of the seat.

The launch provider holds the FAA Part 450 license and the range agreement; none of that covers your payload. You need your own transmit authorization before integration, and the provider will ask for it in writing. For a US commercial operator that means an FCC license, either a full Part 25 filing or the streamlined small-satellite process for spacecraft under 180 kg on missions under six years that deorbit within five, which is faster and much cheaper but comes with those eligibility limits baked in. Anything that images the Earth also needs a NOAA remote-sensing license, and anything carrying controlled technology needs export authorization, which is the item that most often makes a foreign payload on a US vehicle late. Budget six to twelve months for spectrum authorization and treat it as running in parallel with the build rather than after it, because on a rideshare the rocket is rarely the long pole.

SpaceX Transporter (Transporter-1 in January 2021 carried 143 spacecraft, a record at the time) and Bandwagon for mid-inclination orbits. Integrators sit between the customer and the rocket and sell most of the seats: Exolaunch, Maverick Space Systems, Spaceflight Inc., D-Orbit, and Impulse Space's Mira for payloads that want to move after deployment. Non-SpaceX options include Arianespace's Vega C rideshare service using the SSMS dispenser and ISRO's PSLV, and Rocket Lab sells excess Electron capacity when a dedicated mission has margin.

Rideshare works because the marginal cost of an additional 200 kg on a Falcon 9 already flying to SSO is close to zero, so the operator can price near the floor and still make money on a mission it wanted to fly anyway. That prices the whole small-launch sector off a number it cannot match, and it is why so much of the value in this part of the market moved to the integrators and tug operators who resell and reposition the seats. If you are building a business on rideshare, the risk to underwrite is dependence: a single provider sets the price, the cadence, and the destination orbits, and it also runs the largest satellite constellation in the world. Watch what happens if Neutron and New Glenn reach real cadence, since a second high-volume rideshare provider is the only thing that would change the pricing dynamic.

12 minSpaceX's Record Breaking* Rideshare Mission Launches 143 SatellitesScott Manley · 100k+ views

8 minWatch SpaceX deploy Transporter-5 rideshare satellites in spaceVideoFromSpace · 10k+ views

Rideshare on NASA Missions: Realities and Triumphs (NASA) · Streamlining Licensing Procedures for Small Satellites: Report and Order (Federal Communications Commission)

A reusable medium-lift launcher flies a two-stage vehicle in the 15–25 tonne class to LEO and brings the first stage back to a pad or a droneship for reflight. Falcon 9 is the only one operating at scale: 22.8 t to LEO fully expendable, roughly 17 t with the booster recovered, a 5.2 m fairing, and a $74M list price in 2026, which works out to about $3,000/kg against maximum payload and closer to $4,000–5,000/kg on a real mission that recovers the booster. The first stage is the expensive part of a rocket, so recovering it turns most of the vehicle from a consumable into a capital asset that gets amortized over dozens of flights. Fleet-leading boosters have flown more than 25 times each. The vehicle set the reference price every other option in this sheet is quoted against.

The list price is the least interesting strength. What reuse actually bought is cadence and schedule certainty: 165 Falcon 9 flights in 2025 out of 324 orbital launch attempts worldwide, with a standing fleet of flight-proven boosters large enough that one customer's slip does not consume a year of someone else's manifest. If your spacecraft is four weeks late, there is another slot. The weakness is that recovery costs performance, so a mission that needs full expendable capacity pays for it in a discarded booster and a higher price. The structural weakness is concentration: one company flies roughly half the world's orbital launches, which means one grounding event, one anomaly investigation, or one commercial decision moves the entire market. Customers who need a second source have to pay considerably more to keep one alive.

This is the default for anything in the 1–17 tonne class going anywhere from LEO to GTO. Pick it when cost per kilogram and schedule confidence matter more than a specialized upper stage, which covers most commercial constellations, most NASA science missions, and a growing share of national-security launch. Do not pick it when the payload needs a long-coast, multi-restart upper stage for direct GEO insertion or a high-energy escape trajectory, when the mass exceeds what the vehicle can lift, or when the customer's procurement requires a second certified supplier; go to a heavy-lift launcher for those. If your payload is under a tonne, do not buy a whole vehicle at all. Buy a rideshare seat on one.

22.8 t to LEO fully expendable and roughly 17 t with the booster recovered · $74M list price in 2026 · about $3,000/kg against maximum payload, and $4,000–5,000/kg on a typical recovered mission · 165 Falcon 9 flights in 2025 out of 324 orbital launch attempts worldwide · fleet-leading boosters past 25 flights · 5.2 m fairing.

A US launch and its booster landing run under a single FAA Part 450 vehicle operator license, which is written per vehicle and per operation rather than per flight, so raising cadence does not mean relicensing each launch. The binding constraint is environmental and range capacity, not the license itself: annual launch caps at Cape Canaveral and Vandenberg are set through federal environmental reviews, and SpaceX's requested increase at Vandenberg drew a vote against it from the California Coastal Commission in 2024 before the Space Force proceeded on federal authority. National-security payloads need separate vehicle certification through the NSSL program, which is a multi-year process independent of the FAA license. Vehicle telemetry and flight-termination links are range-coordinated and licensed apart from anything the payload transmits, and the vehicle remains ITAR-controlled, so foreign payloads need export authorization well before they arrive at the integration facility.

SpaceX Falcon 9 Block 5, flying from LC-39A and SLC-40 at the Cape and SLC-4E at Vandenberg, recovering boosters at Landing Zone 1 and on the droneships "Of Course I Still Love You", "Just Read the Instructions", and "A Shortfall of Gravitas". The competition is close but not yet flying at rate: Rocket Lab's Neutron targets a first flight in late 2026 with a reusable first stage, Stoke Space's Nova is designed for full reuse of both stages, and Blue Origin's New Glenn is larger and recovers its booster at sea. China has several reusable medium-lift programs in flight test, including Landspace's Zhuque-3 and Space Pioneer's Tianlong-3.

The cost structure of a reused vehicle is closer to an airline's than a rocket builder's: high fixed cost in pads, droneships, refurbishment, and recovery crews, plus a per-flight cost dominated by the expendable upper stage, the fairing recovery cycle, propellant, and range fees. That structure rewards volume brutally, which is why the operator with the highest flight rate has the lowest cost and why the price gap widens rather than narrows. The published list price is not the marginal cost, and there is no reason to think it is close to it, since the operator has no competitive pressure to price lower and uses most of the capacity for its own constellation. If you are underwriting a competitor, the number that matters is not whether the booster comes back. It is flights per booster per year and how many customers exist at that cadence, because a reusable vehicle flying six times a year has worse economics than an expendable one flying thirty.

8 minHow Does SpaceX Recover and Reuse Falcon 9?NASASpaceflight · 50k+ views

6 minHow Many Times Can SpaceX Reuse The Falcon 9?Primal Space · 500k+ views

15 minHow Reusable Rockets WorkThe Space Race · 500k+ views

Performance Efficient Launch Vehicle Recovery and Reuse (NASA) · National Security Space Launch: Increased Commercial Use of Ranges Underscores Need for Improved Cost Recovery (GAO)

A heavy-lift launcher puts 20 tonnes or more into low Earth orbit and, more importantly, carries an upper stage built for long coasts and multiple restarts. Vulcan Centaur, Ariane 6, New Glenn at about 45 t to LEO, and Falcon Heavy at up to 63.8 t expendable are the Western vehicles in service or entering it, and prices typically run $110–250M. Raw lift is rarely what the customer is buying. A direct-GEO insertion, a six-hour coast to drop two national-security payloads into different planes, or a departure burn onto an escape trajectory all depend on an upper stage that can restart hours after launch and hold cryogenic propellant that long, which the cheaper medium-lift vehicles do not do as well. Governments also pay a deliberate premium here to keep more than one certified supplier alive, and that assurance is a real part of the product.

The strength is capability at the top end: mass, fairing volume, and high-energy performance that no medium-lift vehicle matches, plus mission designs that need the upper stage to behave like a spacecraft for most of a day. The weakness is price per kilogram, roughly two to four times a reusable medium-lift launcher, and cadence, which for most of these vehicles is a handful of flights a year rather than one a week. The failure mode is schedule. When a program has a single certified vehicle and that vehicle has an anomaly or a supply problem, payloads wait years, and Ariane 6's late arrival left European institutional missions buying Falcon 9 seats in the interim. Low flight rate also keeps unit cost high, because fixed factory and pad costs spread over few launches, which makes the price gap self-reinforcing.

Choose heavy lift when the payload genuinely exceeds medium-lift capability, when the mission needs direct GEO insertion or a high-energy departure that saves the spacecraft months of orbit raising and hundreds of kilograms of propellant, or when the buyer requires a second certified provider. National-security space launch is the clearest case: the US buys across two providers by policy, and Europe buys Ariane for institutional payloads as a matter of sovereignty. Do not pay the premium when a medium-lift launcher plus electric orbit raising gets the same spacecraft to the same place for less, which is the usual answer for a commercial GEO satellite that can accept four to six months of transfer. If the payload is under 10 tonnes and the schedule is flexible, medium lift almost always wins on cost.

Falcon Heavy up to 63.8 t to LEO expendable · New Glenn about 45 t to LEO · typical prices $110–250M per launch · roughly two to four times the per-kilogram cost of a reusable medium-lift launcher · a few flights a year per vehicle rather than one a week · upper stages designed for multi-hour coasts and several restarts.

US heavy-lift launches fly under an FAA Part 450 license like any other, but the binding approval for the main customer is separate: National Security Space Launch certification, which requires a defined number of successful flights plus a deep design and process review, and which took Vulcan into 2025 to complete. Ariane 6 is not under FAA jurisdiction at all. It launches from the Guiana Space Centre under French national space law, with ESA and the European Commission setting a preference for European launchers on institutional payloads. Export control is the recurring friction on transatlantic missions: US commercial satellites moved to the US Munitions List in 1999 and most returned to the Commerce Control List in 2014, but flying a US payload on a foreign vehicle still means a technical assistance agreement, monitored access, and a licensing timeline measured in months. Assume export paperwork, not the rocket, sets the earliest date a cross-border mission can integrate.

ULA's Vulcan Centaur with its Centaur V upper stage, certified for national-security missions in 2025; Arianespace's Ariane 6 in A62 and A64 configurations from Kourou; Blue Origin's New Glenn with a 7 m fairing and a reusable first stage; and SpaceX's Falcon Heavy, which flies rarely but carries the high-energy missions Falcon 9 cannot, including Psyche and the Europa Clipper. Non-Western equivalents include China's Long March 5 and India's LVM3.

Heavy lift is a low-volume, high-fixed-cost business kept alive by government demand. Commercial GEO orders, which used to fill these manifests, collapsed to roughly a dozen satellites a year worldwide, so the remaining revenue is national-security launch, civil science, and a small number of very large commercial payloads. That is why the US Space Force buys on multi-year block contracts across two providers and why Europe funds Ariane 6 with an explicit annual support payment. For anyone building here, the market is not price-elastic and it is not growing on the commercial side, so the realistic path is winning a share of an assured-access program. The interesting wildcard is that Amazon's Leo constellation bought heavy-lift capacity in bulk across Vulcan, Ariane 6, and New Glenn, which is the largest commercial order this class has ever seen and is not a repeatable event.

20 minHow does ULA's Vulcan rocket compare to the competition?Everyday Astronaut · 100k+ views

5 minEurope’s next big rocket in a nutshellEuropean Space Agency, ESA · 50k+ views

15 minThe Real Reason SpaceX Developed The Falcon Heavy Rocket!The Space Race · 500k+ views

Ariane 6 overview (ESA) · Defense Primer: National Security Space Launch Program (Congressional Research Service)

A super-heavy reusable launcher aims to put roughly 100 tonnes into low Earth orbit and recover both stages. Starship is the only vehicle in this class flying: a stainless-steel two-stage design where the booster returns to the launch tower and the upper stage is meant to come back too, with Block 3 targeting about 100 t to LEO in the fully reusable configuration. About a dozen integrated flights had flown by mid-2026, and as of that point it had not delivered an operational payload for a customer. The vehicle only reaches high-energy destinations by refueling in orbit, so the Artemis lunar lander architecture needs something on the order of 8–16 tanker flights per Moon mission. That dependency, not the launch itself, is the part of the plan with the least flight data behind it.

The interesting number is not price. It is that a roughly 9 m diameter payload bay and 100 t of mass would remove the two constraints that have shaped satellite design since the 1960s. Fifty years of spacecraft engineering exists to make things light and to fold them, and if mass and volume stop binding, you build satellites more like ships than like watches: thicker structure, off-the-shelf parts, real radiators, and no deployable that has to work on the first try. The weakness today is that none of this is available to buy. Cost per flight is far above the $10–30M target, which is why this entry carries the $100M–1B cost band rather than a low one. The failure mode is schedule and reliability, not concept: every anomaly triggers an FAA mishap investigation that grounds the vehicle, and the whole architecture rests on cryogenic propellant transfer that nobody has yet demonstrated at scale.

For almost every mission being planned today, do not design around it. Use a medium-lift or heavy-lift launcher, take the mass and volume constraints, and treat super-heavy as an upgrade you can adopt later. The exceptions are missions that cannot exist otherwise: a lunar surface element in the tens of tonnes, a large-aperture telescope that would rather fly a rigid 8 m mirror than fold a segmented one, or a constellation whose satellites are too big to launch economically any other way, which is the first operational job Starship is actually being built for. If you are a commercial buyer, the useful posture is to design a payload that flies on today's vehicle but has a growth path that uses the volume, and to watch the tanker demonstrations rather than the launch statistics, because refueling is the gate on everything past LEO.

About 100 t to LEO targeted for Starship Block 3 in the fully reusable configuration · roughly 9 m payload bay diameter · about a dozen integrated flights by mid-2026 · 8–16 tanker flights per crewed lunar mission in the Artemis architecture · a stated long-run target of $10–30M per flight, far below present cost · FAA approval to raise the Starbase launch cap from 5 to 25 flights a year.

Starship flies under an FAA Part 450 license, and the license terms have been the practical throttle on the program. The FAA raised the annual cap at Starbase from 5 to 25 flights in 2025 after an environmental assessment, and a separate review covers Starship operations from Kennedy Space Center. Every flight anomaly triggers an FAA-supervised mishap investigation that grounds the vehicle until closed, which is why the flight rate is driven as much by paperwork as by hardware. The debris footprint is a live issue: upper-stage breakups in early 2025 forced airspace closures over the Caribbean and diverted commercial aviation, and the resulting hazard areas now cover a large stretch of ocean on every flight. Add ITAR on top, plus a maritime and airspace coordination burden that grows with cadence, and the regulatory work scales faster than the vehicle count does.

SpaceX Starship and its Super Heavy booster, flying from Starbase in Texas with a Kennedy Space Center pad in work, contracted by NASA for the Artemis III and IV Human Landing System and intended to deploy the Starlink v3 satellites that Falcon 9 cannot carry economically. Blue Origin's New Glenn is smaller and only partly reusable, so it does not compete directly. China's Long March 9 is the only other announced vehicle in this class and is years from flight.

The cost case rests entirely on flight rate and on how much refurbishment a returned stage actually needs, and neither number is public or settled. The program's own economics today are those of a development program: a very large fixed investment, a production line in Texas building vehicles faster than they fly, and revenue that comes from NASA contracts and from launching the company's own constellation rather than from third-party customers. If cost per flight lands anywhere near the target, the second-order effects matter more than the launch market, because cheap mass in orbit changes what spacecraft, propellant depots, and servicing businesses are worth building. If it lands at a few hundred million per flight instead, it is a specialized heavy-lift vehicle with a lunar contract, which is still a business but a much smaller one. Do not build a plan that requires the low number before tanker flights have been demonstrated.

11 minStarship changes the rules for commercial payloadsEager Space · 10k+ views

4 minSpaceX Starship V3 megarocket explained - How it differs from V2VideoFromSpace · 100k+ views

15 minStarship Orbital RefuelingEager Space · 50k+ views

NASA Artemis Programs: Crewed Moon Landing Faces Multiple Challenges (GAO) · NASA’s Developments in Cryogenic Fluid Management Technology (NASA)

Very low Earth orbit is roughly 250–350 km, below where satellites normally operate, and it is a straight trade of lifetime for performance. Two things improve as you get closer to the ground. Ground sample distance scales with range for a fixed aperture, so the same telescope resolves roughly 1.5 to 2 times finer detail at 300 km than at 500 km, and received power scales with the inverse square of range, which is about 4–5 dB of extra link margin over that same step. One thing gets much worse: atmospheric drag, which rises steeply as altitude falls and swings by roughly an order of magnitude over the 11-year solar cycle. An unpropelled satellite at 300 km reenters in months rather than years, so propulsion stops being optional and becomes the thing the spacecraft is designed around. Everything else about VLEO follows from that one fact.

You get a smaller telescope for the same resolution, a stronger link for the same antenna, cheaper launch because the vehicle lifts more to a lower altitude, and free disposal, since the atmosphere removes the satellite whether you want it to or not. The costs are a propellant budget that dominates the design, atomic oxygen that erodes exposed polymers and coatings, and drag torques that fight the attitude control system continuously. The failure mode is running out of propellant early, and it is easy to hit because drag is not predictable: a solar maximum stronger than the design case can consume a mission's entire margin in a year. Conjunction screening is also harder down here, because drag makes position prediction worse just when the traffic is densest with newly launched and decaying objects.

Go to VLEO when the aperture you can afford is the constraint on the mission and the extra factor of 1.5 to 2 in resolution is worth rebuilding the spacecraft around propulsion, which is the case for sub-10 cm imaging and for direct-to-device links where every decibel of margin matters. It also suits any mission where a deliberately short life is acceptable or desirable, including tactical demonstrations and classified payloads you would rather see reenter on schedule. Do not go there for anything that needs a five-year life on a small propellant budget, for constellations where replacement capex is already the dominant cost, or for science that needs a stable, repeatable ground track. If your aperture is comfortable at 500 km, fly at 500 km and spend the saved propellant mass on the payload.

Roughly 250–350 km altitude · about 1.5 to 2 times finer ground resolution than 500 km for the same aperture · roughly 4–5 dB more received power over that same step · months to natural reentry without propulsion · atmospheric density swinging by about an order of magnitude across the 11-year solar cycle · GOCE operated near 255 km for about four years with continuous drag compensation.

The debris rules that constrain higher orbits are essentially free here. The FCC's five-year post-mission disposal expectation is met automatically at these altitudes, and the orbital debris showing in a license application is the easiest part of the filing. What still binds is spectrum: you need the same FCC or national authorization and the same ITU filing as any other satellite, with the same bringing-into-use milestones, and altitude buys you nothing there. Any imaging payload needs a NOAA remote-sensing license regardless of how low it flies. The one item VLEO operators do have to take seriously is reentry casualty risk, since these satellites come down by design and often more than once a year across a fleet; US government practice sets a limit of 1 in 10,000 for human casualty from surviving debris, which pushes designers toward demisable materials rather than dense titanium and steel components.

ESA's GOCE gravity mission is the best-documented case, flying near 255 km from 2009 to 2013 with electric propulsion running continuously to cancel drag. JAXA's SLATS, also called Tsubame, deliberately walked down from 271 km to 167 km between 2017 and 2019 to characterize the environment. Starlink satellites are inserted around 280–300 km and spend weeks there before raising to their operational shells, which is the closest thing to fleet operating experience anyone has. The commercial VLEO companies are still pre-fleet: Albedo's Clarity satellites target 10 cm-class imagery from around 275 km, Redwire's SabreSat and the Thales Alenia SkimSat platform are aimed at the same regime, and the EU's DISCOVERER program funded much of the underlying materials and aerodynamics work.

VLEO moves cost from the payload to the bus. A 10 cm-class image from 275 km needs a telescope perhaps a third the diameter of one collecting the same detail from 500 km, and optics cost scales steeply with aperture, so the payload gets much cheaper. What you pay back is shorter life, a propulsion system sized for continuous thrust, and a replacement cadence that never stops. The economics work if the satellite is cheap and mass-produced and the imagery commands a premium for resolution; they do not work for a bespoke $100M spacecraft, which is why nobody flies one of those down there. The number to underwrite is fleet capex per year, not per satellite: if a VLEO satellite costs 40% of a conventional one but lasts a third as long, you are worse off, and the whole case rests on the payload savings being larger than that.

13 minAir Breathing Satellites in Very Low Earth OrbitEager Space · 10k+ views

9 minAir Breathing Ion Thrusters & Low Orbit SatellitesScott Manley · 100k+ views

GOCE (ESA) · The Benefits of Very Low Earth Orbit for Earth Observation Missions (arXiv)

Low Earth orbit runs from about 160 km to 2,000 km, and in practice most of the traffic sits between 500 and 600 km. A satellite at 550 km travels at roughly 7.6 km/s and completes an orbit every 95 minutes. It is the cheapest place to put mass, at around $3,000/kg on a reusable medium-lift launcher, the closest to the ground for both resolution and link budget, and the mildest radiation environment outside the atmosphere, since it sits below the inner proton belt except where the South Atlantic Anomaly dips down. It is also, at 550 km, self-cleaning: a dead satellite reenters in roughly five years without help. Everything in this sheet gets compared against LEO, and the comparison usually comes down to one thing, which is that a satellite this close can only see a small piece of the Earth at a time.

The strengths compound. Cheap launch means cheap satellites are worth flying, short range means small apertures and small antennas work, benign radiation means commercial-grade electronics survive a five-year mission behind a couple of millimeters of aluminum, and 20–50 ms round-trip latency in service is good enough for interactive applications that GEO cannot support. The weakness is geometry. From 550 km only about 4% of the Earth's surface is above the horizon, and only about half a percent of it sits above a usable 25° elevation angle, so one satellite gives you a handful of eight-to-twelve-minute passes a day over a given site. Continuous global coverage takes several hundred spacecraft. The failure mode for a business is underestimating that number: a constellation sized for coverage on paper turns out to need two or three times as many satellites once you add elevation-angle margin, capacity, and spares.

Use LEO as the default and make the other orbits justify themselves. It is right for imaging, for any interactive communications service, for technology demonstration, for science that wants to look down, and for anything where per-satellite cost matters more than per-satellite coverage. Go higher only for specific reasons: to GEO when you need a fixed point in the sky and can live with half a second of latency, to MEO when 20–30 satellites covering the globe beats several hundred, to SSO when you need identical lighting on every pass. Go lower, to VLEO, only when aperture is the binding constraint. If you are sizing a constellation, start from the elevation angle your user terminal actually needs and work backwards, because that number sets the satellite count and the satellite count sets the entire capital plan.

160–2,000 km by definition, with most traffic at 500–600 km · about 7.6 km/s orbital velocity and a 95-minute period at 550 km · roughly $3,000/kg to reach · about 4% of the Earth's surface above the horizon from 550 km, and around half a percent above 25° elevation · 20–50 ms round-trip latency in service · roughly five years to natural reentry from 550 km, against the FCC's five-year post-mission disposal expectation.

LEO is where the regulatory pressure has concentrated, because it is where the satellites are. A commercial US operator files with the FCC, which in turn files with the ITU, and non-geostationary licenses carry deployment milestones: half the constellation within six years of grant and all of it within nine, which is what forces operators to launch on a schedule set by paperwork rather than by demand. ITU rules add their own bringing-into-use deadline and a staged milestone schedule after that, and filings that miss them lapse. Sharing rules bind at both ends: equivalent power flux density limits under ITU Article 22 exist to protect the geostationary arc from non-geostationary transmissions, and separate coordination applies between LEO systems, where the FCC's approach has generally been to require the later applicant to accept the interference. Disposal is now a licensing condition rather than a guideline, since the FCC's 2022 rule replaced the old 25-year guidance with a five-year expectation for LEO.

The International Space Station near 400 km, Starlink's main shells around 550 km, Amazon Leo near 590–630 km, Iridium NEXT at 780 km, and the Space Development Agency's Transport and Tracking layers near 1,000 km. Almost all commercial Earth observation flies here too, from Planet's Doves to WorldView Legion, though most of it uses the sun-synchronous subset rather than a general LEO inclination.

LEO turns a space system from a capital asset into a manufacturing problem. Launch is cheap enough and the satellites small enough that unit cost, build rate, and replacement cadence dominate the business case, rather than the exquisite reliability that a single expensive spacecraft demands. That has two consequences worth planning for. First, the capital requirement scales with satellite count, so a coverage-driven system is expensive long before there is any revenue, which only arrives after several hundred units are up. Second, replacement never stops: a fleet with a five-year design life needs a fifth of itself rebuilt and relaunched every year, forever, which is an operating expense that constellation business plans routinely treat as a one-time build. If you are evaluating one, ask what the steady-state annual replacement bill is and whether the revenue covers it.

8 minThe Only Video Needed to Understand Orbital MechanicsAnimations Xplaned · 1m+ views

18 minLEO Satellite Networks: Brief Introduction to Communications ChallengesIain Explains Signals, Systems, and Digital Comms · 10k+ views

4 minHow Do Satellites Get & Stay in Orbit?SciShow Space · 1m+ views

Basics of Space Flight, Chapter 5: Planetary Orbits (NASA) · Mitigation of Orbital Debris in the New Space Age: Second Report and Order (Federal Communications Commission)

A sun-synchronous orbit is a near-polar LEO, typically 500–800 km at a retrograde inclination of 97.4–98.6°, chosen so that the orbit plane rotates at exactly the rate the Earth moves around the Sun. The Earth is not a sphere, and its equatorial bulge tugs on an inclined orbit hard enough to make the orbit plane precess; picking the right inclination for a given altitude tunes that precession to 0.9856° per day, which is 360° in a year. The result is that the satellite crosses the equator at the same local solar time on every pass, forever, without spending any propellant to maintain it. That fixed lighting is the product. It is why nearly all Earth observation flies here, and it is the clearest example in this sheet of an orbital-mechanics side effect becoming a commercial requirement.

Identical illumination on every pass is what makes change detection work: two images of the same field taken three weeks apart differ because the field changed, not because the shadows moved. The dawn-dusk variant, flying along the day-night terminator, keeps the solar arrays in near-continuous sunlight and avoids most eclipse cycling, which suits power-hungry radar satellites well. The costs are real. Launching retrograde gives up roughly 400 m/s of free velocity from the Earth's rotation, so a vehicle's quoted SSO payload typically lands 20–40% below its LEO figure, part from the retrograde launch and part from the higher altitude. The failure mode is congestion rather than physics: the mid-morning shell around 10:30 local time is the most crowded piece of sky there is, because every optical imaging program wants the same lighting, and that concentration drives both conjunction rates and downlink contention.

Choose SSO when consistent lighting matters, which covers optical imaging, most multispectral and hyperspectral work, and any time series where a human or an algorithm compares images taken weeks apart. Choose the dawn-dusk variant for SAR and for any payload whose power budget benefits from continuous sun. Do not pay the SSO launch penalty if you do not need the lighting: a communications constellation gains nothing from it, and a mid-inclination LEO covers populated latitudes with fewer satellites and better launch performance. If revisit rate matters more than lighting consistency, a mix of inclinations usually beats piling more satellites into the same sun-synchronous plane, because satellites in one plane all pass over a site at nearly the same time of day.

Typically 500–800 km altitude · 97.4–98.6° retrograde inclination, rising with altitude · plane precession of 0.9856° per day, matching the Earth's motion around the Sun · roughly 400 m/s of Earth-rotation assist given up at launch · quoted SSO payload 20–40% below the same vehicle's LEO figure · the 10:30 local-time shell as the most congested band · X-band downlinks at 8,025–8,400 MHz.

Two things bind here that do not bind elsewhere. The first is downlink spectrum: Earth observation satellites share a narrow X-band allocation at 8,025–8,400 MHz, and with hundreds of imaging satellites in nearly the same plane wanting the same polar ground stations at the same moments, contention is a real scheduling constraint rather than a paper one. Ka-band at 25.5–27.0 GHz is the growth path and is where high-rate missions are moving. The second is imaging licensing: any US commercial remote-sensing system needs a NOAA license under 15 CFR Part 960, which since the 2020 rewrite sorts systems into three tiers by whether comparable capability is already available elsewhere, and which is what permitted sub-25 cm imagery to be sold commercially. The license carries conditions the government can invoke, including limits on imaging specific areas, and foreign customers add export-control review on top. Everything else, the FCC filing, the ITU coordination, the five-year disposal expectation, is the same as any LEO mission.

Landsat 8 and 9 and the Sentinel-2 pair at mid-morning local time, Sentinel-1 in a dawn-dusk orbit for radar, NOAA's JPSS weather satellites at 13:30, and essentially the whole commercial imaging fleet: Planet's SuperDoves and SkySats, Maxar's WorldView Legion, Airbus's Pléiades Neo, and the ICEYE and Capella SAR constellations. SpaceX's Transporter rideshare missions fly to SSO for exactly this reason, and the polar ground station sites at Svalbard, Inuvik, Punta Arenas, and Troll exist because a sun-synchronous satellite passes over them on every single orbit.

The launch penalty is a fixed tax on the orbit and shows up directly in the price per kilogram, which is part of why rideshare took over this segment: aggregating a hundred small payloads onto one SSO mission spreads the penalty across all of them. The more interesting cost is downstream. Because everyone wants the same local time, imaging companies compete for the same lighting, the same ground contacts, and increasingly the same customers, and image prices have fallen faster than the capital cost of collecting them. If you are building an SSO imaging business, the differentiator is rarely the orbit or the sensor. It is revisit rate, tasking latency, and whether you own the analytics that turn pixels into something a customer will pay a subscription for.

15 minGeostationary, Molniya, Tundra, Polar & Sun Synchronous Orbits ExplainedScott Manley · 500k+ views

6 minSpace: Prograde, Retrograde, and Sun-Synchronous Orbitsastronow · 50k+ views

14 minSun Synchronous Orbits | Orbital Mechanics with Python 34 V2.0Alfonso Gonzalez - Astrodynamics & SE Podcast · 10k+ views

Catalog of Earth Satellite Orbits (NASA Earth Observatory) · Licensing of Private Remote Sensing Space Systems: Final Rule (NOAA)

Medium Earth orbit is everything between about 2,000 km and 35,000 km, and it is the compromise almost nobody takes. On paper it looks ideal: one satellite at 8,062 km sees roughly 28% of the Earth's surface, so 20–30 spacecraft give near-global coverage instead of the several hundred a LEO system needs, and round-trip latency of 110–150 ms is fast enough for enterprise traffic, video calls, and most cloud applications. The reason the band is nearly empty is radiation. The inner proton belt peaks near 3,000 km and the outer electron belt is most intense between roughly 15,000 and 25,000 km, so a MEO spacecraft sits inside or between the two most damaging regions in near-Earth space. Two things live there anyway: navigation constellations at 20,200 km, and SES's O3b broadband satellites at 8,062 km, parked in the relatively quiet gap between the belts.

The strength is coverage per satellite. Twenty-something spacecraft is a manageable fleet to build, launch, and operate, the ground antennas track slowly enough that a modest gimballed dish works, and handovers happen every tens of minutes rather than every few minutes. The weakness is that MEO spacecraft carry the heaviest shielding and the most conservative parts of any operational satellites, which makes them heavy, expensive, and a generation behind on electronics. Total ionizing dose over a long MEO mission runs one to two orders of magnitude above a LEO mission of the same length. The specific failure mode is deep dielectric charging: high-energy electrons from the outer belt penetrate a few millimeters of aluminum, deposit charge inside cable insulation and circuit boards, and eventually produce an internal arc that no amount of surface grounding prevents. Nothing decays out of MEO either, so disposal means spending propellant to reach a graveyard orbit.

Pick MEO when you need global or near-global coverage from a fleet you can actually afford to build, and when 110–150 ms of latency is acceptable. That describes navigation, which has no realistic alternative, and enterprise and government broadband trunking where the customer wants better than GEO latency without funding a LEO constellation. Do not pick it for consumer broadband, where LEO wins on latency and on capacity per dollar, or for anything short-lived and cheap, because the radiation environment forces exactly the expensive, rad-hard, long-life spacecraft that a cheap mission is trying to avoid. Also check your latitude coverage before assuming global: the original O3b constellation is equatorial, so it serves roughly ±45° of latitude and needs inclined planes to reach higher latitudes.

2,000–35,000 km by definition, with real traffic at 8,062 km and 20,200 km · inner proton belt peaking near 3,000 km and outer electron belt most intense between roughly 15,000 and 25,000 km · about 28% of the Earth's surface visible from one satellite at 8,062 km · 20–30 satellites for near-global coverage against several hundred in LEO · 110–150 ms round-trip latency in service · GPS orbital period of 11 h 58 m.

MEO carries an interference constraint that LEO and GEO do not, because a MEO satellite crosses the geostationary arc as seen from a ground station and would otherwise transmit straight into a GEO receiver. ITU rules require non-geostationary systems to protect the arc, and O3b complies operationally rather than by design: as a satellite approaches the arc from a gateway's point of view, the network hands that beam off to another satellite and switches the first one off, which costs capacity and complicates scheduling. Navigation is a different regime. GNSS signals live in internationally protected radionavigation satellite service allocations around 1,559–1,610 MHz, and because billions of receivers depend on them, adjacent-band proposals get fought hard, which is what drove the long dispute over Ligado's terrestrial use of nearby L-band spectrum. Disposal is a licensing item too: nothing decays from MEO, so operators must reserve propellant to raise a retired satellite clear of the constellation.

GPS at 20,200 km, Galileo at 23,222 km, GLONASS at 19,100 km, and BeiDou's MEO shell at about 21,500 km. On the commercial side, SES's O3b and second-generation O3b mPOWER satellites at 8,062 km sell trunking to cruise ships, remote enterprise sites, cellular backhaul, and government customers. The newest entrant is defense: the US Space Force has begun buying missile warning and tracking satellites for MEO as a middle layer between the traditional GEO assets and the proliferated LEO layer.

MEO spacecraft cost like GEO spacecraft and cover like a small constellation, which is an awkward place to be commercially. Rad-hard parts, heavier shielding, long design lives, and higher launch energy all push unit cost into the hundreds of millions once launch is included, and there are few suppliers with the qualification heritage to build them. That works when the customer is a government funding a public utility, which is why every navigation system is state-funded, and it works for a niche commercial operator selling high-value trunking where LEO capacity is thin and GEO latency is disqualifying. It does not work for consumer services. If you are evaluating a MEO business, the question is whether the customers who cannot use GEO are also customers who will not be served adequately by LEO within the asset's fifteen-year life.

4 minThe Orbits Explained - What is LEO, MEO & GEO?Global Invacom Group · 50k+ views

15 minHow Deadly Are The Van Allen Radiation Belts?Scott Manley · 500k+ views

Types of orbits (ESA) · Mitigating In-Space Charging Effects — A Guideline, NASA-HDBK-4002B (NASA)

Geostationary orbit is a circular equatorial orbit at 35,786 km, where the period matches the Earth's 23 h 56 m sidereal rotation and the satellite therefore appears to hang motionless in the sky. That single property built a fifty-year industry, because a user antenna can be bolted to a wall and aimed once. Three satellites spaced around the ring cover everything below roughly ±70° latitude. The price is distance: the round-trip path of 143,000 km sets a physics floor of 477 ms, and real services measure 550–650 ms once processing and terrestrial backhaul are included. Getting there is its own project, since launchers normally drop payloads in a geostationary transfer orbit and the spacecraft circularizes itself, which costs roughly 1,500–1,800 m/s depending on the launch site's latitude.

A fixed point in the sky means no tracking antenna, no handover, no constellation, and a coverage footprint that is stable for fifteen years, which is exactly the product for broadcast, for a government that wants one asset parked over one theater, and for maritime and aviation routes that do not move. Station-keeping is modest at roughly 50 m/s a year, almost all of it north-south, correcting the inclination drift the Sun and Moon impose. The weaknesses are latency and geometry. Half a second of round-trip delay disqualifies interactive applications, high latitudes see the satellite low on the horizon or not at all, and there is no cheap version of a GEO spacecraft, since anything worth putting there is worth building to last fifteen years. The failure mode that matters commercially is obsolescence: you commit to a payload design three to four years before launch, then fly it for fifteen, so a market that shifts underneath you leaves an expensive asset serving demand that moved.

Choose GEO when the user terminal must be cheap, fixed, and non-tracking, when one footprint over one region is the product, or when the mission is broadcast, weather imaging of a full disk, or a government lease over a specific theater. Do not choose it for consumer broadband, where latency has already lost that market to LEO, or for anything where the payload design has to keep up with a fast-moving market, unless the payload is software-defined and can be reconfigured on orbit. If you need wide coverage without half a second of delay, MEO is the middle answer; if you need low latency and can fund the satellite count, LEO wins outright. Once you have committed to GEO, the next decision is how to circularize from GTO: a chemical apogee burn takes days to a couple of weeks and starts revenue almost immediately but costs several hundred kilograms of propellant, while electric orbit raising uses a small fraction of that mass and takes four to six months. Take chemical when the slot has to be occupied quickly or when a launch is already paid for; take electric when the mass saving drops you a launch class, which is usually worth more than half a year of deferred revenue.

35,786 km altitude and a 23 h 56 m period · three satellites cover everything below about ±70° latitude · 477 ms round-trip physics floor and 550–650 ms measured in service · roughly 50 m/s a year of station-keeping, almost all north-south · ITU slots spaced about 2° apart · 1,500–1,800 m/s from GTO to GEO, or four to six months of electric spiral · 15-year design life, ending with a roughly 300 km raise to a graveyard orbit.

GEO is the only orbit where the position itself is a licensed, scarce, internationally coordinated asset. Slots and their associated frequencies are secured through ITU filings under the Radio Regulations, coordinated with every administration whose systems might be affected, and typically spaced about 2° apart for similar services, which is what limits how many satellites can share a band over a given region. A filing must be brought into use within seven years or it lapses, a rule that exists because administrations were filing speculatively for slots they had no intention of using. In the US the FCC grants the license on top of the ITU filing, requires a performance bond of a few million dollars that is released as milestones are met, and sets a deadline to launch and operate within five years. End-of-life disposal is a license condition: raise the satellite roughly 300 km above the ring into the graveyard orbit and passivate it, which costs a few months' worth of station-keeping propellant and has to be reserved from the start.

The commercial fleet operators are Intelsat, SES, Eutelsat, Viasat, EchoStar, and Telesat, alongside regional operators like Arabsat, Yahsat, and APT. Civil users include the NOAA GOES-R weather satellites and Europe's Meteosat, which image the full disk continuously because they never move relative to it. Defense users include the AEHF and WGS constellations. Astranis is the notable attempt at a smaller, cheaper GEO satellite, selling a dedicated few-hundred-kilogram spacecraft to a single country or operator rather than a share of a large one.

A GEO mission is a fifteen-year annuity bought up front: $150M to $500M for the spacecraft, $60M to $150M for launch, and three to four years from order to service. That works when the revenue is predictable over the whole period, which describes broadcast and government leases and describes almost nothing else now. Commercial GEO orders fell from more than twenty satellites a year in the 2000s to roughly a dozen a year worldwide, because consumer broadband moved to LEO and video distribution moved partly to terrestrial streaming. The industry's answer is software-defined payloads that can reallocate beams and capacity in orbit, which converts some of the obsolescence risk into a flexibility argument. If you are underwriting a GEO satellite today, the question is whether the demand it serves will still be there in year twelve, because that is where the return lives.

5 minGeosynchronous Orbits are WEIRDminutephysics · 1m+ views

2 minHow Do Geosynchronous Satellites Systems Work?ChromaCommunications · 50k+ views

10 minGraveyard Orbits Where Old Satellites Are ForgottenScott Manley · 100k+ views

Regulation of satellite systems (ITU) · Advanced Propulsion for Geostationary Orbit Insertion and North-South Station Keeping (NASA)

Cislunar space is everything between GEO and the Moon, plus the Earth-Moon Lagrange points, and escape trajectories are everything past that. The delta-v is not the hard part: a translunar injection costs about 3.12 km/s from LEO, escape sits only a little higher at a characteristic energy of C3 = 0, and holding a near-rectilinear halo orbit around the Moon takes under 10 m/s a year, which is less than geostationary station-keeping. What changes is operations. Launch windows recur roughly monthly and last days rather than being a date you pick, light-time delay puts 2.6 seconds of round trip between you and the Moon and 6 to 44 minutes to Mars, tracking time on deep-space antennas is rationed rather than purchased, and there is no orbital infrastructure of any kind: no GNSS, no relay network you can assume, no rescue, no second attempt.

The Lagrange points and lunar halo orbits are genuinely good places to be for certain jobs. L2 gives a telescope a stable thermal environment with the Earth, Moon, and Sun all on one side, and a near-rectilinear halo orbit keeps a station in continuous line of sight to Earth while staying cheap to maintain. The weakness is that everything else about the environment is hostile to normal engineering practice. You are outside the magnetosphere, so galactic cosmic rays and solar particle events hit the spacecraft directly, and shielding against a large event is impractical for anything uncrewed. Navigation depends on ground-based radiometric tracking, which means your position knowledge is only as good as your allocation of antenna time. The failure mode is the one CLPS has demonstrated repeatedly: landers that work through cruise and then fail in the last minutes, because terminal descent has to work on the first try with no way to abort and no help available.

This is not a choice most missions face. Go beyond GEO when the destination is the point: lunar surface science, a Gateway-supported program, an astrophysics observatory that needs L2, a planetary mission, or a defense payload watching the volume between GEO and the Moon. If you go, plan around the operational constraints rather than the delta-v. Book deep-space tracking early and design the mission to survive with less of it than you asked for. Assume the launch window is a few days a month and that missing one costs a month. Carry enough autonomy that the spacecraft can hold a safe state through a light-time round trip plus a ground shift change. If your objective can be met from GEO or from a high Earth orbit, do that instead, because the ground segment cost and the operations risk both step up sharply once you leave Earth orbit.

About 3.12 km/s from LEO to a translunar injection · C3 = 0 as the escape threshold · under 10 m/s a year to hold a near-rectilinear halo orbit · 2.6 s round-trip light time to the Moon and roughly 6 to 44 minutes to Mars · lunar launch windows recurring monthly and lasting days · CLPS lunar lander task orders in the $79–200M range.

Cislunar is where the regulatory framework is thinnest, and the gap is authorization rather than spectrum. The Outer Space Treaty makes states responsible for authorizing and continuously supervising their nationals' space activities, but in the US no agency clearly holds that authority for a novel activity like operating on the lunar surface. The FAA licenses the launch, the FCC licenses the radio, and NOAA licenses remote sensing of the Earth, and a lunar lander doing something none of those cover falls between them. Spectrum has its own quirk: the ITU's protected deep-space bands apply beyond two million kilometers, so a lunar mission is coordinated as near-Earth space research and shares its allocations more widely than a Mars mission does. Layered on top are COSPAR planetary protection, where the Moon is a light-touch Category II but Mars is not, the Artemis Accords as a non-binding framework signed by more than fifty states covering interoperability and safety zones, and separate presidential-level approval for launching nuclear power sources.

NASA's CLPS program has bought lunar deliveries from Astrobotic, Intuitive Machines, and Firefly, whose Blue Ghost landed successfully in 2025 after earlier attempts fell short. Rocket Lab's CAPSTONE demonstrated a near-rectilinear halo orbit with a 12U CubeSat, which is the orbit Gateway will use. Beyond the Moon, JWST and Euclid operate at Sun-Earth L2, China's Chang'e program has run sample return and far-side missions with the Queqiao relay satellites providing communications, and Impulse Space's Helios kick stage is aimed at commercial cislunar delivery. Military interest is real but new and small, centered on tracking objects in the volume beyond GEO.

Almost all the money here is government money, and the buying model recently changed. CLPS replaced cost-plus lander development with fixed-price task orders in the $79–200M range, accepting that some missions fail in exchange for many more attempts per dollar, and the early results have been exactly that mix. That model works for delivery. It has not yet produced a commercial customer who pays for lunar activity because they want the result, which is the thing missing from every cislunar business plan. The nearer-term commercial opportunities are the supporting layers rather than the destinations: transfer stages that take payloads from a rideshare drop-off to a lunar trajectory, communications and navigation relays, and tracking services. If you are evaluating a cislunar company, look at how much of its revenue depends on a single national program continuing at its current funding level, because for most of them the answer is all of it.

10 minRocket Lab's Tiny Rocket Sent A Satellite to The Moon: What does 'Near Rectilinear Halo Orbit' Mean?Scott Manley · 100k+ views

14 minWhy Spacecraft Are Using These Crazy Routes To The Moon - Weak Stability and Ballistic Capture.Scott Manley · 100k+ views

2 minNear Rectilinear Halo Orbit Explained and Visualizedaisolutionsinc · 50k+ views

Basics of Space Flight, Chapter 4: Trajectories (NASA) · Cislunar Near Rectilinear Halo Orbit for Human Space Exploration (NASA)

A CubeSat is a spacecraft built to a published mechanical standard: one unit (1U) is a 10 cm cube massing up to about 2 kg, and real satellites are stacks of them, most often 3U, 6U, or 12U. Cal Poly and Stanford wrote the specification in 1999 so student projects could share a ride, and the part that mattered turned out to be the deployer rather than the cube. A spring-loaded dispenser holds the satellite, bolts to the launch vehicle or an ISS airlock, and gives the launch provider one qualified interface instead of a new interface for every payload, which is what made secondary payloads routine instead of a favor. Because the interface was settled, a components industry grew up around it, and you can now buy structures, reaction wheels, star trackers, radios, and deployable arrays as catalog parts that fit together. The same standard sets the ceiling: a 6U spacecraft carries a few centimeters of optical aperture, generates roughly 20–60 W, and gets its data down at tens of megabits per pass. That is enough for a radio payload, a coarse imager, or a technology demonstration, and not enough for anything that needs real power or a real optic.

The strengths are cost and schedule: a 6U bus runs roughly $100–500k against $2–15M for an ESPA-class platform, a rideshare seat for a small CubeSat is under about $150k, and a team that buys rather than builds can go from contract to launch in about a year. The weaknesses come straight from the volume and power budget, since aperture, transmit power, and propellant all scale with the box you are allowed to occupy. The failure mode to plan around is a satellite that deploys and never answers. Surveys of university-class missions by Michael Swartwout have put infant mortality — dead on arrival or lost within 30 days — near 40%, which is what you get from single-string avionics, no propulsion to fix a bad orbit, and a few minutes of contact a day to debug with. Commercial operators do much better, mostly by flying the same design dozens of times and by accepting that some units are written off.

Pick a CubeSat when the payload genuinely fits in a few liters and tens of watts, and when the mission tolerates losing a unit. That covers RF payloads (AIS, ADS-B, signal monitoring), GNSS radio occultation for weather, coarse multispectral imaging, in-orbit demonstration of a component you intend to fly on something bigger, and any constellation whose value comes from having many satellites rather than good ones. Do not pick one because it is cheap if the payload needs an aperture, a kilowatt, or a gigabit downlink, because you will spend two years discovering the ceiling and then buy a small satellite bus anyway. If your payload is between 20 and 100 kg, go straight to a microsatellite platform: the standard stops paying for itself once you outgrow the deployer, and the launch cost difference no longer dominates the program.

1U is a 10 cm cube massing up to about 2 kg, with 3U, 6U, and 12U the common sizes · a 6U bus runs roughly $100–500k · a small CubeSat rideshare seat costs under about $150k · a 6U generates roughly 20–60 W · typical mission life 1–3 years · infant mortality near 40% on university-class missions · FCC streamlined smallsat license fee $30,000.

The FCC's streamlined smallsat process (part 25.122, in force since 2020) is what makes CubeSat licensing tractable: a $30,000 application fee, a target of about six months, and eligibility for systems of no more than 10 satellites under 180 kg each, operating six years or less including deorbit, below 600 km without propulsion, at least 10 cm in the smallest dimension, and carrying a unique telemetry marker so the tracking networks can tell them apart. Universities often go the amateur route instead, coordinating through the IARU and operating under Part 97, which costs almost nothing but bars commercial use of the downlink. Either way the spacecraft still needs an ITU filing submitted by a national administration, and a CubeSat that images the Earth needs a NOAA remote sensing license on top of the radio license. The practical planning number is that licensing and frequency coordination take longer than building the satellite, so start them before you cut metal.

Planet's Dove and SuperDove imagers (3U, around 200 on orbit), Spire's Lemur-2 fleet (3U, AIS and GNSS radio occultation), Swarm's quarter-U SpaceBEEs, NASA's MarCO twins (6U, relayed InSight's Mars landing in 2018), CAPSTONE (12U, flown to a lunar near-rectilinear halo orbit in 2022), and TBIRD (6U, 200 Gbps laser downlink). Bus and component vendors include NanoAvionics, EnduroSat, Blue Canyon Technologies, GomSpace, ISISpace, and AAC Clyde Space; deployers come from Cal Poly's original P-POD lineage plus Nanoracks, Exolaunch, and Planetary Systems.

The bus is a catalog purchase and the mission cost is dominated by everything around it: payload, integration and test, licensing, launch, and a year or more of operations and ground station time. Component prices have fallen a long way and are now flattening, because the parts are already commercial-grade and the remaining cost is qualification and low volume. The strategic point for anyone building a business here is that the CubeSat form factor is a means, not a market. The companies that made money from it (Planet, Spire) did so by flying enough units to sell a data product, and both moved to larger spacecraft as soon as their payloads outgrew the box. If a plan needs CubeSats specifically because the unit price is low, check whether the payload's aperture and power requirements have been sized honestly, because that is where these programs usually break.

3 minCrazy Engineering: CubeSatsNASA Jet Propulsion Laboratory · 100k+ views

5 minCubeSats explained by NASASciNews · 10k+ views

2 minWhat is a CubeSat?Canadian Space Agency · 10k+ views

CubeSat 101: Basic Concepts and Processes for First-Time CubeSat Developers (NASA) · CubeSat Design Specification Rev. 14.1 (Cal Poly)

A merchant bus is a 100–500 kg spacecraft platform sold as a product: structure, solar arrays and batteries, attitude determination and control, propulsion, avionics, and a TT&C radio, with a published mechanical, electrical, and data interface that your payload bolts to. The category exists because of a launch adapter. The EELV Secondary Payload Adapter ring was built to sell spare capacity on big government launches, its ports were rated at 181 kg (320 kg on ESPA Grande), and satellite builders sized their products to fit, so a launch accessory ended up defining a class of spacecraft. A typical platform supplies 100–1,000 W to the payload, points to a few tens of arcseconds or better, carries enough propellant for orbit maintenance and disposal, and is designed for 5–7 years. The shift over the last decade is that you now buy this from a configuration sheet instead of writing a spacecraft specification, and several vendors publish prices and lead times.

Buying a bus moves a payload company out of the spacecraft business, which is usually the right trade: the vendor has flown the design, carries the qualification paperwork, and delivers in 12–24 months against three or four years for a clean-sheet build. The cost is that you inherit someone else's design decisions, and the interface is a real constraint — payload volume, view angles, thermal rejection, and peak power draw are fixed before you start. The failure mode to underwrite is a first-article bus. A platform that has flown twice has no meaningful on-orbit statistics, and the losses in this class have overwhelmingly been in attitude control and power management software rather than in structures, which is exactly the part a data sheet does not tell you about. The second, less obvious risk is corporate: bus vendors keep getting acquired by primes, and a supplier that becomes part of your competitor's parent company is a schedule risk you did not price.

Buy a merchant bus when the payload runs 20–200 kg and needs somewhere between 100 W and 1 kW, which covers most imaging, RF sensing, and communications payloads that will not fit a CubeSat. If the payload fits in a few liters and tens of watts, a CubeSat is four to ten times cheaper and you should take the ceiling. If it needs more than about 2 kW or a 15-year life, you are buying a large platform and the merchant bus market cannot help you. The volume threshold is the other decision: below roughly 20 identical satellites, buying is clearly right; past 50 or so, the margin you pay the bus vendor starts to exceed the cost of standing up your own line, which is why every large constellation operator eventually builds its own. And weight vendor track record heavily over spec sheets, because on this class of spacecraft the difference between a good and a bad bus shows up in year two, not at delivery.

ESPA port limit 181 kg, ESPA Grande 320 kg · bus prices roughly $2–15M, with Apex publishing $3.5–9.5M for its Aries platform and $13.5M with a 15-month lead time for the GEO version · delivery 12–24 months · payload power 100–1,000 W · design life 5–7 years · pointing to tens of arcseconds or better.

The question that decides the paperwork is who holds the license. Operate the satellite yourself and you need a full FCC Part 25 authorization (or a foreign equivalent plus US market access) and an ITU filing submitted by a national administration, because the streamlined smallsat path stops at 180 kg, 10 satellites, and six years — a 300 kg bus or an eleventh spacecraft drops you into the full process. The FCC's July 2026 licensing overhaul made that process more predictable: a 30-day completeness determination, a 15-day public notice, and a 60-day shot clock on decision, with annual band-specific processing rounds opening 1 January for Ka, Ku, V, and Q band, and a $10 million surety bond for systems that opt into a round, declining to zero once 90% of the constellation is deployed. Export control is the other gate. Most commercial buses have sat under the EAR as ECCN 9A515 rather than ITAR's USML Category XV since the 2014 reform, but classification follows capability, and a bus with hardened parts or certain propulsion still lands on the Munitions List. Non-US vendors market "ITAR-free" platforms for exactly this reason, and it is a genuine advantage when the customer is a foreign government.

Blue Canyon Technologies (RTX), York Space Systems, Terran Orbital (bought by Lockheed Martin in October 2024 at roughly $450M enterprise value), NanoAvionics (Kongsberg), Apex with its Aries, Nova, and GEO Aries platforms, Airbus Arrow and Longbow, Surrey Satellite Technology, and Astranis for small GEO communications satellites. Loft Orbital sits one layer up, buying buses from Airbus and Apex and selling payload slots plus the license and operations as a service.

The economics are product economics rather than program economics: a vendor's margin depends on how many identical units it ships, so the whole industry is chasing order books and configuration discipline. Published prices are themselves the change — a platform with a list price and a lead time is a very different purchase from a proposal cycle. Consolidation has been the dominant trend, with primes buying merchant bus makers to secure supply for defense constellation programs, and that is worth diligencing directly: ask whether your supplier will still be independent when your second batch is due. Costs keep falling on parts and on avionics, where commercial components and automotive-grade electronics do the work, but integration and test labor does not scale the same way, so do not model a bus price curve as steep as a component price curve.

14 minInside the Astranis Satellite Factory — How to Connect MillionsS3 | Science, Startups, & Stories · 10k+ views

7 minHow to Build a SatelliteAstranis Space Technologies · 10k+ views

2 minHow a satellite is built.Airbus Defence and Space · 10k+ views

State-of-the-Art of Small Spacecraft Technology: Complete Spacecraft Platforms (NASA) · Setting the Standard: Recommendations on "Launch Unit" Standard SmallSat Sizes between CubeSats and ESPA-Class (The Aerospace Corporation)

A constellation bus is a spacecraft designed around the production line rather than around the mission. The satellite is flat so it stacks without a dispenser, the harness and avionics are simplified to cut touch labor, parts are commercial or automotive grade instead of rad-hard, and the design is frozen for a production block and then revised in place, the way a car model gets a facelift. SpaceX's IPO filing put the Redmond line at about 70 Starlink satellites a week between December 2025 and April 2026, which is roughly 3,600 a year, and the current V2 Mini is a 740–800 kg spacecraft. The consequence for engineering is that reliability targets move from the unit to the fleet: you accept a five-year design life, expect to lose a percentage every year, and design the constellation so losing any one satellite is invisible to users. Roughly 12,600 Starlink satellites had been launched by August 2026 against about 10,900 still on orbit, which is the same statement expressed as attrition.

Building at rate is the only way to field thousands of satellites, and it drops unit cost by an order of magnitude compared with a bespoke bus. It also compresses technology cycles: a line that turns over its whole fleet every five years can put a new generation on orbit while a traditional program is still in critical design review. The weaknesses are the mirror image. Per-unit reliability is genuinely lower, so a design flaw discovered on orbit is a flaw in hundreds of satellites at once and the only fix is to build the next block differently. Commercial parts limit you to benign radiation environments, which is why this approach works in LEO and not at 20,000 km. And the failure mode that gets missed in business plans is not a satellite failing, it is the line stopping: a constellation with an idle factory is a depreciating asset with a hard expiry date.

Choose a production-line bus when the mission needs more than roughly a hundred spacecraft and the value comes from coverage, revisit, or capacity rather than from any individual satellite's capability. That means broadband, data transport meshes, proliferated missile tracking, and high-revisit sensing. Do not use it for anything that needs a large aperture, high radiation tolerance, or a 15-year life, because the design choices that make the line fast are the ones that cap those. The financial test is whether you can fund the steady state, not the first build: if the constellation is 6,000 satellites on a five-year life, you need to produce and launch about 1,200 satellites a year forever, and a plan that treats deployment as a one-time capital event has understated the business by a large multiple. If the fleet is under about 50 spacecraft, buy a merchant small satellite bus instead; the line will never pay for itself.

Starlink V2 Mini launch mass roughly 740–800 kg · about 70 satellites a week off the Redmond line from December 2025 to April 2026, near 3,600 a year · roughly 12,600 Starlink satellites launched all-time against about 10,900 on orbit in August 2026 · five-year design life · SDA Transport Layer vehicles averaging about $14M on Tranche 1 and about $21.5M on later awards · a 6,000-satellite fleet on a five-year life needs about 1,200 replacements a year.

License milestones turn directly into a manufacturing schedule, which is the regulatory fact that matters most here. An FCC NGSO authorization requires 50% of the authorized satellites on orbit within six years of grant and 100% within nine, and the ITU's milestone regime under Resolution 35 requires 10% of the system deployed within two years of the end of the seven-year bringing-into-use period, 50% within five years, and 100% within seven, with a missed milestone capping the recorded assignment at the number actually flown. So a 3,000-satellite filing is a commitment to a build rate before it is a commitment to a market. Disposal is the other per-unit cost: the FCC's 2022 rule expects LEO satellites to be gone within five years of mission end, so every spacecraft carries deorbit propellant and every unit cost includes it. US government constellations sidestep the FCC entirely, since federal systems get their spectrum through NTIA rather than a commercial license, which trades milestone pressure for appropriations risk.

SpaceX's Starlink line in Redmond, Washington; Amazon Leo's production facility in Kirkland; Airbus OneWeb Satellites in Florida, which built roughly 650 spacecraft for the OneWeb constellation; the Space Development Agency's Transport Layer, with more than 60 Tranche 1 vehicles on orbit by July 2026 from York Space Systems, Lockheed Martin, Northrop Grumman, and Rocket Lab, and a 2026 award of about $1.5B to Lockheed and Northrop for 72 more; and China's Guowang and Qianfan constellations, which are scaling their own lines.

Unit cost falls with cumulative volume, but the learning curve is shallower on government programs than commercial ones because requirements churn: SDA's Transport Layer averaged roughly $14M per space vehicle on Tranche 1 and about $21.5M on later awards, against a program goal near $15M. The structural feature investors most often miss is that replacement capex never stops, so the right way to model a constellation is as a factory with a depreciating installed base rather than as infrastructure that gets built once. That in turn explains the vertical integration: if you own the launch vehicle, the satellite line, and the user terminal, the recurring cost of standing still is yours to compress, and if you rent any of them, someone else prices your steady state. When diligencing a constellation, the questions worth asking are what a marginal satellite costs today, how that has moved over the last hundred units, and what the launch cost per satellite is at the current stacking density.

4 minThe Spacecraft Assembly LineSSPIVideo · 10k+ views

1 minAirbus OneWeb Satellite ManufacturingClark Lindsey · 10k+ views

Key Technology, Programmatic Drivers, and Lessons Learned for Production of Proliferated Small Satellite Constellations (Small Satellite Conference) · Satellite Constellations 2026 - Survey, Trends and Economic Viability (Small Satellite Conference)

A large platform is the traditional big satellite: typically 2,000–6,500 kg at launch, 15–25 kW of power at end of life, a 15-year design life, and a price between $150M and $1B. Everything about it follows from the fact that it has to work for fifteen years with no maintenance in an orbit that never clears itself. Parts are radiation-hardened and qualified for the full dose, every critical function is redundant, propellant is sized for station-keeping across the whole life, and the spacecraft goes through months of thermal-vacuum and vibration testing before it ships. The payload gets deployable reflectors several meters across and kilowatts of transmit power, which is capability no small bus can supply. The schedule follows from the same place: three to four years from order to launch, plus four to six months of electric orbit raising if the operator chose electric propulsion over chemical to save several hundred kilograms of propellant mass.

Aperture, power, and service life are the product, and they are not substitutable. A 6 kW payload with a 9 m reflector does things no constellation of 200 kg satellites can do, and fifteen years of revenue from one asset amortizes a large capital cost at a rate that looks good to a lender. Against that, the schedule risk is severe and one-sided: a satellite ordered in 2026 flies technology frozen around 2027 and is still in service in 2042, so the operator is betting on a market fifteen years out. It is also a single point of failure with no partial credit. ViaSat-3 F1 launched in 2023 and lost more than 90% of its designed terabit of capacity because a reflector did not deploy, which produced a $421M insurance claim and a hole in the operator's capacity plan that took years to fill. That is the failure mode: not degraded performance, but a working satellite that cannot do the job it was bought for.

Choose a large platform when the mission needs a big aperture or several kilowatts on a payload, when the coverage requirement is one region served continuously, or when the asset is worth more parked over a specific place than distributed. Broadcast, government leases over a theater, weather from GEO, and maritime and aviation broadband along fixed routes all still fit. Do not choose one for interactive consumer services, where the 550–650 ms GEO round trip loses to LEO regardless of how much capacity you build. And check the small GEO options before defaulting to a full-size platform: Astranis, AscendArc, Swissto12, and Apex's $13.5M GEO Aries now cover missions that used to require a $300M satellite, and about 20 of the 47 GEO orders placed since 2020 have gone to small platforms. If the payload genuinely needs 20 kW, none of them help and you are back with the primes.

2,000–6,500 kg at launch · 15–25 kW power at end of life · 15-year design life · $150M–1B per satellite · three to four years from order to launch, plus four to six months of electric orbit raising · 11 GEO satellites ordered industry-wide in 2025, the best year since 2016.

A GEO satellite without a slot and frequency rights is scrap metal, so the filing usually precedes the order. The ITU sequence is advance publication, then a coordination request, then notification and recording in the Master Register, and the fees are almost irrelevant next to the hardware: in 2024 an advance publication cost 570 CHF, a coordination request ran from about 5,710 to 67,000 CHF, and a notification from about 15,910 to 116,000 CHF. What binds is time and priority. You get seven years from filing to bring the assignment into use, and for a GEO network that means keeping a satellite capable of transmitting or receiving on those frequencies at the notified position for a continuous 90 days, then informing the Bureau within 30 days. Priority runs by filing date, so a later filer has to reach agreement with every earlier one whose frequencies and service area overlap, and that bilateral coordination between administrations routinely takes years. Slots sit about 2° apart. National licensing runs in parallel: an FCC Part 25 authorization for a US-licensed satellite, or a foreign license plus FCC market access to serve US customers.

Airbus Eurostar Neo and the software-defined OneSat, Thales Alenia Space's Spacebus Neo and Spacebus 4000 and its Space Inspire line, Boeing 702, Lockheed Martin LM2100, Lanteris (formerly Maxar Space Systems) 1300-series, Northrop Grumman GEOStar, and CASC's DFH-4 and DFH-5 in China. Specific spacecraft worth looking up: Hughes' Jupiter 3 at about 9,200 kg, the largest commercial communications satellite built; Eutelsat Quantum, the first fully reprogrammable commercial satellite; and the ViaSat-3 series.

The commercial GEO order book collapsed when LEO broadband arrived and has partly recovered: 47 orders since 2020, 11 of them in 2025, with most of the recent ones replacements, government programs, and a leasing deal rather than new capacity for growth. Roughly half now go to small GEO platforms, which is the low end of the market leaving. The primes' answer to fifteen-year obsolescence is the software-defined payload, where beams, coverage, capacity, and frequency plan are reprogrammed on orbit instead of being fixed at build. It is the right idea and it has been expensive to execute: Airbus has booked around ten OneSat orders without delivering one, and took roughly €600M of charges in its space division with OneSat named as the principal cause. If you are underwriting a business that depends on a large platform, the two numbers that matter are capacity delivered per dollar of capex over the asset's whole life, and how much of the fifteen-year revenue is already contracted.

4 minAssembly, Integration and Test (AIT) for Communication Satellite in Airbus Defence and SpaceAirbus Defence and Space · 10k+ views

4 minInside A Satellite Clean RoomTom Scott · 500k+ views

3 minSpacebus NEO: 100% electric !Thales · 5k+ views

Neosat boosting Europe's telecommunications by satellite (ESA) · Software-defined satellite enters commercial service (ESA)

A LEO broadband constellation is a fleet of hundreds to thousands of satellites at 300–600 km, each acting as a moving cell tower with a phased array underneath it that paints beams onto the ground and hands users off as it passes. Being 60 times closer than a geostationary satellite buys two things: round-trip latency of 20–50 ms instead of 550–650 ms, and a link budget good enough that a flat consumer terminal closes it. Capacity comes from spectrum reuse, so total system throughput is roughly the usable spectrum multiplied by the number of beams you can put over paying customers at once. Laser crosslinks let a satellite over open water route traffic to one over a gateway, which decouples coverage from ground infrastructure. Starlink is the reference implementation: about 12,600 satellites launched by August 2026 with roughly 10,900 still on orbit, more than 10 million subscribers as of the first quarter of 2026, and $11.4B of connectivity revenue in 2025.

It is the only satellite architecture that competes with terrestrial broadband on user experience, and it reaches places fiber and cellular will not for decades: ships, aircraft, rural households, disaster zones, and forward military units. The structural weakness is that capacity is geographically stranded. A satellite over the Pacific serves no customers, and demand concentrates over land and over cities, so the fleet has to be sized for the peak cell rather than the average one, and most of the capacity you launched is idle most of the time. That shows up commercially as sold-out cells: the operator stops taking customers in the areas where demand is highest. The other weakness is that nothing about the constellation is a one-time cost. Satellites live about five years, so a fleet of several thousand needs roughly a fifth of itself replaced every year forever, and the moment the factory or the rocket stops, the asset starts evaporating.

For a buyer, choose LEO broadband when latency or mobility matters: interactive applications, moving platforms, and anywhere terrestrial service is absent. If the users are fixed and the traffic is broadcast or bulk transfer along a stable route, a GEO high-throughput satellite may still deliver more bits per dollar of capex. For anyone considering building one, the honest test is vertical integration. The only operator with demonstrated positive unit economics owns its launch vehicle, its satellite line, and its user terminal, and each of those you rent instead is a margin someone else takes and a schedule you do not control. If you cannot own at least the satellite line and the terminal, the realistic plans are wholesale capacity into other people's networks, a regional or sovereign system funded as infrastructure, or a niche the incumbents will not serve.

Roughly 10,900 Starlink satellites on orbit in August 2026 out of about 12,600 launched · more than 10 million subscribers as of Q1 2026, up from 5.0 million a year earlier · $11.4B of connectivity revenue in 2025 · average revenue per user about $66/month in Q1 2026, down from $86 a year earlier · 20–50 ms round-trip latency · Amazon Leo at roughly 330 satellites against 3,232 required by July 2029.

Spectrum, not technology, is what gates this business. User links run in Ku and Ka band and gateways are moving into V and Q, and every non-geostationary system has to protect the geostationary networks that were there first through the equivalent power flux density limits in Article 22 of the Radio Regulations — limits that WRC-27, meeting in Shanghai from 11 October to 12 November 2027, is scheduled to review, which is the largest open regulatory question in the sector. Priority runs by filing date and processing round, and it is enforceable. When the FCC waived Amazon's milestone of 1,616 satellites by 30 July 2026 in a conditional decision on 6 June 2026, it kept the 3,232-satellite deadline of July 2029 and ruled that satellites launched after the original date carry no interference protection until Amazon reaches 50% deployment, meaning they must not interfere with Starlink. The milestones themselves are hard: 50% of an FCC-authorized NGSO system within six years and 100% within nine, and under ITU Resolution 35, 10% within two years of the end of the seven-year bringing-into-use period, 50% within five, and 100% within seven. Filings are cheap relative to satellites — an ITU advance publication ran 570 CHF in 2024 and a coordination request 5,710 to about 67,000 CHF — which is exactly why more than 300 constellations covering over a million satellites were filed between 2017 and 2022, more than a hundred times the operational fleet. A paper filing is not worthless, since priority is real and transferable, but Resolution 35 caps the recorded assignment at the number of satellites actually deployed, so a filing you cannot build shrinks to the size of what you built. The FCC's July 2026 overhaul added annual band-specific processing rounds opening 1 January for Ka, Ku, V, and Q, a $10M surety bond that declines to zero at 90% deployment, and a 60-day shot clock on decisions. And a US license only sells service in the US: every other country requires its own landing rights, which is why coverage maps have holes that have nothing to do with orbits.

SpaceX Starlink, with V3 satellites carrying about 1 Tbps of downlink each against roughly 80 Gbps on a V2 Mini; Amazon Leo, renamed from Project Kuiper in 2025, at roughly 330 satellites in mid-2026 and building toward 3,232 Gen 1 spacecraft with a further 4,500 Gen 2 approved in January 2026; Eutelsat OneWeb, a 648-satellite system sold wholesale to enterprises and telcos rather than to consumers; Telesat Lightspeed; and China's Guowang and Qianfan constellations.

This is a manufacturing and distribution business wearing a space company's clothes. Revenue growth has been fast and average revenue per user has fallen as service expanded into price-sensitive markets, from $86/month in early 2025 to about $66/month in early 2026, which means subscriber growth has to outrun ARPU decline to keep revenue compounding. The cost side is dominated by the replacement rate, launch cost per satellite (which improves with stacking density, not just with rocket price), and terminal subsidy. The reason vertical integration keeps showing up in this category is that all three of those are internal transfer prices for one operator and external invoices for everyone else. When evaluating a competitor's plan, the questions that separate real from optimistic are what a marginal satellite and a marginal terminal cost, how many paying users sit under a typical beam, and what fraction of the fleet is over water at any moment.

28 minHow does Starlink Satellite Internet Work?📡☄🖥Branch Education · 5m+ views

21 minStarlink Technology OverviewPhil Trubey · 100k+ views

Low Earth Orbit Satellites: Potential to Address the Broadband Digital Divide (Congressional Research Service) · A Survey on Non-Geostationary Satellite Systems: The Communication Perspective (arXiv)

A high-throughput satellite replaces the wide beams of a classic communications satellite with dozens or hundreds of narrow spot beams, each covering a few hundred kilometers. Beams far enough apart can reuse the same frequencies without interfering, so total capacity scales with the reuse factor rather than with the spectrum you were licensed, and a narrower beam has more gain, which puts more power on a smaller terminal. That is what took a single geostationary satellite from a few gigabits per second to Hughes' Jupiter 3 at more than 500 Gbps and the ViaSat-3 class designed for over 1 Tbps. The trade is that the beam layout is baked into the reflector and feed geometry at build, so the satellite is a fifteen-year bet on where demand will be. Latency is set by geometry and cannot be engineered away: 477 ms is the round-trip speed-of-light floor through GEO and services measure 550–650 ms.

Capacity per dollar of capital is genuinely good when the beams sit over paying traffic, and one satellite, one launch, one license, and one set of gateways is a far simpler operation than a constellation. Fifteen years of revenue from a single asset also finances well. The weaknesses are latency, inflexibility, and concentration. GEO lost consumer broadband on latency and is not getting it back. Beams cannot be moved unless the payload is software-defined, so a demand forecast that was wrong in year three is wrong for twelve more years. And the whole investment rides on one deployment sequence: ViaSat-3 F1 reached orbit in 2023 and lost more than 90% of its designed capacity when a reflector failed to deploy, triggering a $421M insurance claim and leaving the operator's capacity plan years behind. There is no partial credit in this architecture.

Choose GEO high throughput when traffic concentrates over a fixed area and the application tolerates half a second of delay. In-flight and maritime connectivity along established routes, video distribution and broadcast, cellular backhaul, enterprise VSAT networks, and government leases where one asset parked over one theater is exactly the product all still work. Do not choose it for consumer internet or anything interactive, because LEO wins on latency by an order of magnitude and the price gap has closed. If the coverage requirement might move, buy a software-defined payload or a small GEO platform instead of a large fixed-beam satellite, and accept lower peak capacity in exchange for being able to re-point. The honest way to run the comparison against LEO is capex per gigabit per second delivered over the asset's whole life, counting the fraction of beams that will actually be sold.

Jupiter 3 at about 9,200 kg launch mass and more than 500 Gbps · ViaSat-3 designed for over 1 Tbps per satellite · Viasat's estimate of $1.2–1.4B all-in for two ViaSat-3 satellites including launches, insurance, and ground infrastructure · a $421M insurance claim after ViaSat-3 F1's reflector failed · 550–650 ms round-trip latency against a 477 ms physics floor · 15-year design life.

Geostationary operators are the incumbents, and their regulatory position is an asset in itself. Under Article 22 of the Radio Regulations, non-geostationary systems must keep their emissions below equivalent power flux density limits designed to protect GSO networks, which is what stops a LEO constellation from simply overwhelming a GEO link in shared Ku and Ka spectrum. WRC-27, meeting in Shanghai from 11 October to 12 November 2027, is scheduled to review those limits, and that review is the single largest regulatory question facing GEO operators; SES and other incumbents have campaigned publicly to keep them, and the LEO operators want them relaxed. The rest of the GEO position is slot and market rights. Slots sit about 2° apart, so only so many satellites can serve one market on the same frequencies; an assignment must be brought into use within seven years of filing by keeping a capable satellite at the notified position for a continuous 90 days; and selling capacity in any country requires that country's landing rights. An incumbent holding landing rights in dozens of markets owns something a new entrant cannot replicate on a satellite-build timescale, and that is a large part of why the traditional operators remain valuable even as consumer broadband leaves.

Hughes JUPITER 3 (EchoStar 24), the ViaSat-3 series with F1 crippled in 2023 and F2 entering service in 2026 over the Americas, Eutelsat KONNECT VHTS, SES-17, the Inmarsat Global Xpress fleet now inside Viasat, and Intelsat's EpicNG satellites now inside SES. Consolidation has been the sector's response to LEO: Viasat completed its acquisition of Inmarsat in 2023, and SES completed its acquisition of Intelsat in July 2025 to form a roughly 120-satellite GEO and MEO fleet.

The metric that decides these programs is capex per delivered gigabit per second across fifteen years, and it improved enormously from wide-beam satellites to HTS, which is why the architecture persisted. What changed is the competition: LEO capacity is now abundant and priced against terrestrial broadband, so the GEO business retreated to the segments where a fixed footprint and a long-lived asset are advantages rather than liabilities. Consolidation followed, and the surviving operators sell multi-orbit service, reselling or partnering for LEO capacity while keeping the GEO fleet for broadcast, mobility, and government. If you are evaluating a GEO program today, look at how much of the capacity is contracted before launch, how much of the revenue is government or mobility rather than consumer, and whether the payload can be re-pointed if the forecast turns out wrong.

4 minKA-SAT The 1st high throughput satellite for EuropeTooway Official · 50k+ views

3 minFaster and More Powerful Satellite Internet Is on Its Way, Here’s HowSeeker · 100k+ views

13 minIntroduction to Ka Band Satellites Part 1YedaCenter · 10k+ views

Modernizing Spectrum Sharing for Satellite Broadband: Report and Order (Federal Communications Commission) · Resolution 76 (Rev.WRC-23): Protection of geostationary networks from the aggregate epfd of non-GSO systems (ITU Radio Regulations)

Direct-to-device means a satellite talking to a phone nobody modified: no dish, no external antenna, no special handset. The difficulty is entirely in the link budget. A handset transmits well under a watt into an antenna with essentially no gain that cannot be pointed, and it is designed to reach a tower a few kilometers away rather than a satellite 500 km up, so the satellite has to supply the whole margin with aperture and receive sensitivity. There are two ways to buy that margin, and the industry has split along them. One is an enormous antenna on a few large satellites: AST SpaceMobile's BlueBird Block 2 is roughly 6,100 kg and unfolds a phased array of about 2,400 sq ft (223 m²), the largest commercial antenna structure in low Earth orbit. The other is a modest array on very many satellites, which is Starlink Direct to Cell, at roughly 650 dedicated spacecraft as of 31 March 2026. Both work by transmitting in spectrum licensed to a terrestrial mobile operator, so the satellite behaves like another cell in the carrier's network and the phone never knows the difference.

The strength is that the addressable market is every phone already in someone's pocket, with no hardware to sell, install, or subsidize. That is a fundamentally different distribution problem from every other satellite service. The weakness is capacity. One beam covers a cell hundreds of kilometers across and shares a few megabits per second across everyone inside it, which makes this a coverage product rather than a capacity product, and it is why first services are messaging, emergency calling, and low-rate data rather than broadband. The failure mode is commercial rather than technical: the spectrum belongs to the carrier, so an operator that has not bought its own spectrum is a supplier to the company that owns the customer relationship, and its economics are whatever the carrier agrees to pay. SpaceX's response was to buy spectrum outright, which tells you how the incumbent reads the risk.

If you are a mobile operator, buy direct-to-device as a coverage feature. It closes dead zones, it is a genuine safety product, and it costs far less than building towers into terrain that has none. Do not sell it as a broadband substitute, because a shared beam over a whole region cannot support the traffic your terrestrial network carries, and customers who expect otherwise will churn. If you are an investor, the question that decides the outcome is who holds the spectrum and on what terms, since that determines whether the satellite operator captures the revenue or collects a wholesale fee. And if you need real throughput to a mobile user rather than a fallback, buy a terminal: a flat-panel LEO terminal delivers hundreds of megabits, and the whole reason direct-to-device is hard is that it refuses to use one.

A handset transmits well under 1 W into a near-omnidirectional antenna · AST's BlueBird Block 2 at about 6,100 kg with a roughly 2,400 sq ft (223 m²) array · roughly 650 Starlink direct-to-cell satellites as of 31 March 2026 · $632M of Starlink Mobile revenue in 2025 across 7.4 million monthly unique devices · $17B paid for 65 MHz of EchoStar AWS-4 and H-block spectrum, approved by the FCC on 12 May 2026 · AST authorized for 248 satellites, with 124 required by 2 August 2030.

This category exists because a regulator invented it. The FCC adopted its Supplemental Coverage from Space framework on 14 March 2024, effective 30 May 2024, the first rules anywhere that let a terrestrial mobile licensee lease its spectrum to a satellite operator so the satellite can serve ordinary handsets, with secondary mobile-satellite allocations added in specific bands and interim 911 routing requirements attached. The framework also fixes where the power sits: the spectrum is the carrier's, so the satellite operator is a tenant unless it buys its own. SpaceX did, paying $17B for 65 MHz of EchoStar's AWS-4 and H-block licences plus $2.6B in stock for 15 MHz of AWS-3, which the FCC approved on 12 May 2026 subject to a $2.4B escrow tied to EchoStar's abandoned terrestrial buildout. That is more money for spectrum than the satellites cost, which is the clearest single statement of what binds in this business. AST took the tenant route with variations, contracting long-term access to up to 45 MHz of Ligado's lower mid-band and coordinating 700 and 800 MHz low-band with AT&T, Verizon, and FirstNet; the FCC granted it commercial authority for a 248-satellite constellation in April 2026 with milestones of 124 satellites by 2 August 2030 and the full system by 2 August 2033. Every other country runs its own version of the same question, so a global service has to be assembled market by market through whoever holds the local mobile license.

Starlink Direct to Cell with T-Mobile, commercially launched as T-Satellite in 2025 and now partnered with more than 30 mobile operators; AST SpaceMobile's BlueBird Block 2 satellites with AT&T, Verizon, Vodafone, and Rakuten as partners, the first of which launched in December 2025; Apple's Emergency SOS via satellite over Globalstar; Skylo's narrowband service over geostationary L-band, used for messaging on Android and iOS handsets; and Lynk Global.

The revenue so far looks like a feature, not a service. SpaceX reported $632M of Starlink Mobile revenue in 2025 across 7.4 million monthly unique devices, and its prospectus projects average revenue per user near $8 a month worldwide, about $18 in high-income markets and $2 in lower-income ones, with its T-Mobile agreement worth around $100M over several years. Those are wholesale-feature numbers rather than standalone-carrier numbers. Costs run the other way: AST's satellites are large and expensive, and SpaceX's spectrum purchase is a capital commitment on the scale of a constellation. So the investable question is whether next-generation satellites raise per-cell throughput enough to sell data rather than messaging, and whether the operator owns spectrum when that happens. If it stays a messaging and emergency feature, the economics belong to the carriers, and the satellite operators are infrastructure suppliers with a single-digit dollar monthly wholesale rate.

17 minStarlink Direct to Cell: How it works, common misconceptions, and answers to your questionsDISHYtech · 100k+ views

2 minT-Mobile & Starlink Launch T-Satellite – How It Works & Why It MattersTEKZOID · 10k+ views

Non-Terrestrial Networks (NTN) (3GPP) · Single Network Future: Supplemental Coverage from Space, Report and Order (Federal Communications Commission)

Protected military SATCOM is what a communications link costs when it has to keep working while someone is actively trying to stop it. The mechanism is a stack of measures, each addressing a different attack. Uplinks run at extremely high frequency near 44 GHz with fast frequency hopping, which forces a jammer to spread its power across a wide band; nulling antennas steer a pattern null onto the jammer's bearing so its energy never enters the receiver; onboard processing demodulates and regenerates the signal instead of repeating it, so a jammer cannot simply swamp a transponder; and crosslinks route traffic satellite to satellite so it never transits a foreign teleport. On top of that, the strategic constellation is hardened against nuclear effects, because its primary job is nuclear command and control, which means shielding, hardened parts, and a design that survives the electromagnetic pulse and radiation from a high-altitude burst. AEHF is the current US system: six satellites of about 6,170 kg each, at roughly $15.5B for the program, or around $1.3B per spacecraft.

It works, which is the point, and nothing commercial comes close: an unprotected commercial link can be jammed by an adversary with a truck and a power amplifier, and this cannot. The weaknesses are cost, capacity, and pace. Protected channels are narrow by design, so a satellite that cost more than a billion dollars carries a small fraction of the traffic of a commercial satellite costing a fifth as much. Development runs a decade, so the fielded system reflects threat assumptions from the previous decade. And the failure mode nobody likes to discuss is concentration: a six-satellite constellation of extremely valuable assets in geostationary orbit is a small, well-known, and fixed target set, which is precisely the argument the proliferated architecture makes against it.

Reserve protected SATCOM for traffic that has to survive a contested environment: nuclear command and control, strategic warning, and the command links that must work when everything else is being jammed. For high-volume theater traffic, wideband systems like WGS carry far more bandwidth per dollar with none of the hardening, and commercial leases carry more still. The live architectural choice is exquisite versus proliferated. SDA's Transport Layer buys resilience through numbers rather than hardening, at roughly $14M per Tranche 1 space vehicle and about $21.5M on later awards against $1.3B for an AEHF satellite, and had more than 60 Tranche 1 vehicles on orbit by July 2026. If the requirement is resilience against counterspace attack, proliferation is the better buy. If the requirement is a link that survives nuclear effects, nothing in LEO does that job today, and you are buying the exquisite satellite whether you like the price or not.

AEHF at about 6,170 kg per satellite and roughly $15.5B for six spacecraft, near $1.3B each · EHF uplinks near 44 GHz with downlinks near 20 GHz · military X-band at 7.25–7.75 GHz down and 7.9–8.4 GHz up · SDA Transport Layer vehicles at roughly $14M on Tranche 1 and about $21.5M on later awards · more than 60 Tranche 1 transport satellites on orbit by July 2026 · about $1.5B awarded in 2026 for 72 more.

This is the one communications category where spectrum is not a race. The bands are set aside for government use in the ITU allocations — EHF uplinks at 43.5–45.5 GHz with downlinks at 20.2–21.2 GHz, and military X-band at 7.25–7.75 and 7.9–8.4 GHz — so a protected system never enters a commercial processing round, never fights a bringing-into-use deadline, and in the US gets its assignments through NTIA and the interagency process rather than an FCC license. What binds instead is export control and alliance management. Crypto, anti-jam waveforms, and nulling antennas sit firmly on the US Munitions List, so allied access runs through government-to-government agreements rather than a sales contract: the UK, Canada, the Netherlands, and Australia bought into AEHF on exactly that basis, and the Protected Tactical SATCOM program is being structured so allies can host payloads. The second constraint is that protected bands are narrow and demand keeps growing, which is why the US Space Force is pushing protection into the waveform. The Protected Tactical Waveform provides anti-jam performance over ordinary transponders, including commercial ones, turning a spectrum-allocation problem into a signal-processing problem and letting commercial capacity carry protected traffic.

Milstar and its successor AEHF, built by Lockheed Martin with Northrop Grumman payloads, with AEHF-6 launched in 2020; WGS, the wideband X and Ka system built by Boeing that carries the bulk of theater traffic without hardening; Evolved Strategic SATCOM as the intended AEHF replacement for nuclear command and control; Protected Tactical SATCOM, with a prototype payload flying and Viasat and Intelsat awarded the first two operational PTS-Global satellites in June 2026; the SDA Transport Layer as the proliferated LEO alternative; and allied systems including UK Skynet 6 and France's Syracuse 4.

Cost per protected bit is the worst in the industry by a wide margin, and it always will be, because hardening and anti-jam processing consume mass, power, and bandwidth that would otherwise carry traffic. Governments buy it anyway because the alternative is a command system an adversary can switch off. The interesting money has moved to the two layers around the satellite: terminals, which are the practical constraint on how much protected capability a force can actually use, and waveforms, which let commercial satellites carry protected traffic and so break the link between protection and a billion-dollar spacecraft. For anyone building a business adjacent to this, the durable positions are terminal manufacturing, waveform and modem intellectual property, and commercial capacity qualified to carry the protected waveform. Building the exquisite satellites themselves is a two-or-three-prime business with decade-long cycles and no commercial market underneath it.

9 minAdvanced EHF Communications SatelliteNorthrop Grumman · 10k+ views

3 minAnti-Jam MilSatCom Overview - Protected (Anti-Jam) Satellite CommunicationsNorthrop Grumman · 10k+ views

5 minAnti-Jam MilSatCom - See Low Cost Terminals in ActionNorthrop Grumman · 5k+ views

Defense Primer: Nuclear Command, Control, and Communications (NC3) (Congressional Research Service) · DOD Satellite Communications: Reporting on Progress Needed to Provide Insight on New Approach (U.S. Government Accountability Office)

Optical Earth imaging is the oldest commercial remote-sensing business, and its whole structure follows from one equation. Ground sample distance scales with altitude and inversely with aperture, so at visible wavelengths a 30 cm pixel from 500 km takes roughly a meter of telescope, while a 3 m pixel needs only a few centimeters. That split the industry cleanly in two. On one side sit small, cheap, numerous satellites: Planet flies about 200 SuperDoves of roughly 5 kg each at 3 m resolution and images the world's land surface every day. On the other sit large, expensive ones: the six WorldView Legion satellites now operated by Vantor deliver 30 cm-class imagery and up to 15 revisits a day over some locations, collecting several million square kilometers daily. Adjacent to both is spectral imaging, where the value is chemistry rather than sharpness — hyperspectral satellites such as Planet's Tanager, Pixxel's constellation, and GHGSat's methane monitors identify materials and gas plumes at spatial resolutions a customer would otherwise consider useless. The sensing and navigation sheet covers how these sensors work; what matters here is the constellation, the revisit rate, and who pays.

A visible-band image is interpretable by any human and by every off-the-shelf vision model, which is not true of radar, and that keeps optical the default product for anything a person has to look at. Costs at the low-resolution end have collapsed. The weakness is the atmosphere, and it is not fixable: cloud blocks the sensor entirely, and roughly two-thirds of the planet is cloud-covered at any moment, so a promised daily revisit can mean a useful image every several days over the places customers most want to watch. Darkness removes half the orbit. The commercial failure mode is subtler and has hurt more companies: the price of an image fell faster than the capital cost of collecting one. High-resolution constellations were built on the assumption of a large commercial market that never materialized at the projected size, and most of the revenue in the category still comes from government contracts.

Choose optical when a human or an analytic model needs to identify what something is, when the target area has reasonable cloud statistics, and when the question tolerates gaps. If the requirement is a guaranteed look at a specific time, put SAR in the architecture, because optical cannot promise a cloud-free pass. If the question is about material composition rather than shape — methane leaks, crop stress, mineralogy, effluent — buy spectral imagery and accept 30 m pixels, since resolution is not what answers that question. Before committing to a constellation, check cloud climatology over the target area and multiply the nominal revisit by the clear-sky fraction; that number, not the advertised revisit, is what the customer actually receives. And check the archive first: archive imagery runs 40–60% below tasking, and a large share of requests are answered by something already collected.

A 30 cm pixel from 500 km takes roughly a 1 m aperture at visible wavelengths · Planet's roughly 200 SuperDoves at 3 m from about 5 kg spacecraft, covering the land surface daily · six WorldView Legion satellites giving up to 15 revisits a day and up to 3.6 million km² of 30 cm-class imagery daily · 50 cm archive imagery around $8–18 per km², with 30 cm tasking above $50 and minimum orders near 25 km² · BlackSky's NRO imagery contract worth up to $1.021B over ten years on an $85.5M five-year base · Planet revenue near $300M in fiscal 2026.

Imaging carries a license that communications satellites do not. Under the Land Remote Sensing Policy Act, a US operator needs a remote sensing license from NOAA's Office of Space Commerce, and the 2020 rewrite replaced case-by-case national security conditions with three tiers keyed to whether comparable data is already available from sources the US does not regulate. Tier 1 systems, whose data is substantially the same as what foreign operators already sell, get minimal conditions and NOAA has issued dozens of them; Tier 2 adds a single condition; Tier 3 covers genuinely unique capability and can carry temporary restrictions. In 2023 NOAA removed most of the remaining temporary conditions, which is what allowed US companies to sell their best products, including imagery below 25 cm. The statute still contains shutter control — the government's authority to restrict collection or dissemination on national security grounds — and it is the reason imagery contracts carry government-priority clauses. On top of the imaging license sit an FCC authorization for the downlink, usually X-band or Ka-band, and export control on the payload: space-qualified optics above certain performance thresholds remain on the US Munitions List even though most spacecraft moved to the Commerce Control List in 2014, so selling a high-resolution imaging satellite abroad is a State Department decision.

Planet's roughly 200 SuperDoves at 3 m, about 21 SkySats at 0.5 m, and Pelican high-resolution satellites; Vantor (formerly Maxar Intelligence) with six WorldView Legion satellites plus the older WorldView fleet; Airbus with two operational 30 cm Pléiades Neo satellites and a 20 cm-class Pléiades Neo Next planned from 2028; BlackSky, built around high revisit rather than maximum resolution; the free Landsat and Sentinel-2 programs, which set the floor price at zero for anything above 10 m; and on the spectral side Planet's Tanager, Pixxel, GHGSat, and NASA's EMIT instrument on the ISS.

The revenue is mostly government, and the contracts are large and long: the NRO's Electro-Optical Commercial Layer awards run ten years, with BlackSky's worth up to $1.021B on an $85.5M five-year base and Maxar's carrying a year-five value above $300M. Commercial demand exists in insurance, agriculture, energy, and finance, but it has consistently been smaller than the projections used to raise capital for high-resolution constellations. Two structural facts drive the economics. Free imagery from Landsat and Sentinel sets the price at zero above about 10 m, so a commercial operator has to be sharper, faster, or spectrally different to charge anything. And the value has migrated from pixels to answers: customers increasingly buy detections, counts, and change alerts rather than scenes, which favors operators with large archives and analytic pipelines over those with the best telescope. If you are underwriting a company here, look at the fraction of revenue that is recurring subscription rather than per-scene, and how much of it renews without a government appropriation.

16 minWhat Satellites Can See From Space Is TroublingAstrum · 1m+ views

8 minTiny satellites that photograph the entire planet, every day | Will MarshallTED · 100k+ views

4 minHow do you observe the Earth with satellites?FutureLearn · 10k+ views

Fundamentals of Remote Sensing (NASA ARSET) · Commercial Remote Sensing Regulatory Affairs (NOAA Office of Space Commerce)

A SAR satellite carries its own illumination, so it images through cloud and at night, which is why defense buys it and why it has been the fastest-growing segment of commercial remote sensing. The sensing and navigation sheet covers how the technique works; the interesting thing at satellite level is what limits a SAR spacecraft, and the answer is power rather than aperture. A 100 kg radar satellite has to radiate kilowatts during a collection while its solar arrays produce a couple of hundred watts averaged over the orbit, so it charges a battery for most of the orbit and spends it in bursts of seconds. Battery capacity and thermal duty cycle, not antenna size, set how many square kilometers the satellite can collect per orbit, and the resulting data volume then runs straight into downlink capacity. That is why the small-satellite SAR companies converged on similar designs: a deployable antenna a few meters long, a large battery, and an aggressive X-band or laser downlink. ICEYE's satellites are around 85 kg and Capella's around 100 kg, against about 2,300 kg for ESA's Sentinel-1.

Guaranteed collection is the product. If a customer needs to know what is at a location on a specific day, only radar can promise it, and that reliability is worth more to defense, insurance, and maritime customers than sharper pixels would be. Coherent processing also gives products optical cannot: interferometric SAR measures ground motion to a few millimeters by comparing the phase of repeat passes, which is how subsidence, dam deformation, mine walls, and volcanic inflation get monitored, and coherent change detection reveals that soil was disturbed or a vehicle moved between two passes. The weaknesses are interpretation and cost per collection. SAR imagery does not look like a photograph and needs trained analysts or trained models, which limits the buyer pool. And because collection is power-limited, a SAR satellite's daily capacity is small compared with an optical imager sweeping a wide swath, so the cost per square kilometer stays higher.

Choose SAR when the answer has to arrive regardless of weather or daylight, when the target area is persistently cloudy, or when the question is about millimeter-scale motion or disturbance rather than appearance. Maritime domain awareness, flood mapping, force monitoring, and infrastructure deformation are the strongest fits. Do not buy SAR when a human analyst needs to identify something visually and the sky is usually clear, because optical is cheaper per square kilometer and every model already understands it. For InSAR specifically, check that the constellation actually flies repeat orbits with tight baselines, since not every SAR satellite can produce interferometric pairs, and a vendor offering deformation products without that geometry is selling something weaker than it sounds. The common architecture is both: optical for wide-area survey and identification, SAR for guaranteed revisit and for change.

ICEYE satellites around 85 kg, with 48 launched since 2018 for the company and its customers · Capella spacecraft around 100 kg, with four on orbit in 2026 · Sentinel-1 at about 2,300 kg, distributing data free and open · commercial spotlight resolution to 25 cm, and Umbra has publicly released a 16 cm image · InSAR measures ground displacement to a few millimeters across repeat passes · a 100 kg SAR satellite radiates kilowatts in bursts from a couple of hundred watts of orbit-average power.

SAR carries the same NOAA remote sensing license as optical, but it has historically been the harder case, because the tier system keys conditions to whether comparable data is already sold by operators the US does not regulate — and for high-resolution radar there were fewer foreign competitors to point at. That is why Umbra flew under temporary conditions capping what it could sell at 25 cm and barring release of phase history data, the raw complex signal that makes interferometry and coherent change detection possible. The restriction was on a capability rather than on a picture, which is the useful thing to understand about remote sensing licensing: regulators care about what the data enables. When NOAA lifted those conditions, Umbra was able to sell sub-25 cm products and phase history for the first time, and released a 16 cm image. The radar itself needs spectrum too, transmitting in the Earth exploration-satellite (active) allocations, principally 9.3–9.9 GHz at X-band, which are shared with radiolocation services and coordinated accordingly. Export control shapes who can buy the satellites rather than the data: SAR payloads sit at the sensitive end of the spacecraft controls, which is part of why a Finnish operator selling spacecraft directly to national governments has been a viable business model where a US one would face a State Department license for each sale.

ICEYE, the largest commercial SAR constellation, which sells satellites to governments as well as data; Capella Space, acquired by IonQ in July 2025 and operating four spacecraft while building the next batch; Umbra, which holds the commercial resolution record; Synspective in Japan; ESA's Sentinel-1, whose free C-band data underpins most civil and academic InSAR; and Airbus TerraSAR-X and the Italian COSMO-SkyMed constellation among the earlier government-backed systems.

SAR was the category that changed most with small satellites: a capability that used to require a two-tonne government spacecraft now flies at 85–100 kg, and prices per image have fallen by roughly an order of magnitude in a decade. Demand is dominated by defense and intelligence, with insurance, maritime, and infrastructure monitoring as the real but smaller commercial markets, and flood mapping in particular has become a repeatable product because it needs exactly the all-weather property optical lacks. Two moats matter. Constellation size sets revisit, which is what a monitoring customer is actually buying, and the analytic layer turns a hard-to-read image into an answer a non-specialist can act on. A third, less discussed, is that selling satellites rather than imagery converts a subscription business into a capital-equipment business with government buyers, and several operators now run both. Free Sentinel-1 data caps what anyone can charge at coarse resolution, so commercial value concentrates in high resolution, fast tasking, and guaranteed revisit.

16 minSatellites Use 'This Weird Trick' To See More Than They Should - Synthetic Aperture Radar Explained.Scott Manley · 1m+ views

2 minCapella Overview and Introducing Evolved Satellite DesignCapella Space · 10k+ views

2 minICEYE's Dwell Imaging Mode Explained By Darren Muff [New Product]ICEYE · 5k+ views

Synthetic Aperture Radar (SAR) (NASA Earthdata) · S1 Mission (ESA SentiWiki)

RF signal geolocation finds transmitters from orbit and works out where they are: ships running with their AIS transponders switched off, GPS jammers, air-defense radars, VSAT terminals, marine radars, push-to-talk radios. The mechanism is timing rather than imaging. Three or more satellites fly in a loose formation, typically spaced about 125 km apart along track, and each records the same emission; the difference in arrival time between a pair of satellites defines one curve on the ground and the difference in Doppler shift defines another, and where the curves from several pairs cross is the emitter. What comes out is a coordinate with a confidence ellipse, a timestamp, and a parametric description of the signal — center frequency, bandwidth, modulation, pulse pattern — rather than a picture. Geometry and signal strength set the accuracy: a longer baseline and a longer, stronger burst both tighten the fix, and commercial systems deliver errors on the order of a kilometer, not meters. Because the payload only listens, the technique works at night, through cloud, and against targets that are deliberately trying not to be seen.

The economics are the strength. A satellite that only has to receive needs no large telescope and no kilowatt transmitter, so the spacecraft are tiny — HawkEye 360's Pathfinder satellites were 13.4 kg each and later clusters run a few tens of kilograms — and a full cluster of three costs roughly what a single sub-meter imaging satellite costs to build many times over. That buys answers imagery cannot give at any price, because a hidden emitter is far easier to detect than to resolve. The failure mode is the negative result. Detection requires the target to transmit during the eight or so minutes a cluster is overhead, so emission control defeats the sensor completely, and an operator who sees nothing cannot tell whether the emitter was silent or whether nobody was watching. A kilometer-wide ellipse in a busy shipping lane can contain a dozen vessels, which is why the data is almost always fused with AIS and with imagery rather than used alone. Dense, overlapping signal environments also make it hard to decide which recorded burst on one satellite matches which on another, and mismatched pairs produce confident, wrong fixes.

Buy RF geolocation when the question is "where is the emitter" and you do not need to see the object. It is the right tipping-and-cueing layer above imaging: a cluster pass covers a very wide swath cheaply, and you spend the expensive sub-meter image only on the coordinate it hands you. If the question is what the object is or what it is doing, buy optical or SAR imaging instead, because a one-kilometer fix will not identify a hull. Do not buy it for persistent watch over a small area, since a fleet of ten clusters gives a handful of looks a day at a given point rather than continuous coverage, and do not buy it expecting message content, because commercial operators detect and locate emissions but do not sell demodulated communications. If your target reliably runs silent, this whole category is the wrong purchase and you are back to imaging.

Clusters of three satellites spaced about 125 km apart along track · 13.4 kg per Pathfinder satellite, a few tens of kilograms on later clusters · more than 30 satellites on orbit as of 2026 · tuning from 70 MHz to 6 GHz, extended to about 18 GHz with a downconverter · geolocation accuracy on the order of a kilometer · $98.7M of HawkEye 360 revenue in 2025, 61% of it from US customers.

This category slipped through the licensing regime that governs imaging. NOAA assessed HawkEye 360's Pathfinder cluster and concluded that a commercial remote sensing operator's license was not required, on the reading that the statute covers sensing the Earth for imagery rather than collecting radio emissions, so RF geolocation operators do not carry the resolution and shutter conditions that constrain optical and SAR licensees. What does bind is ordinary FCC authorization: the payload receives, which needs no transmit license, but the satellite's own telemetry and data downlink is a Part 25 space station license like any other, and the operator must show it will not interfere with the services it is listening to. Export control is the real constraint on the business. Spacecraft with signals-intelligence payloads sit on the US Munitions List under Category XV, so both the hardware and much of the technical data are ITAR-controlled, which is why the customer list is US and allied governments and why cross-border sales move at the speed of State Department approvals. Operators also generally avoid demodulating content: intercepting communications is a criminal matter under US wiretap law, and staying at the detect-and-locate layer keeps the product clear of it.

HawkEye 360 is the reference system, flying clusters of three from Cluster 1 through Cluster 13, which launched in January 2026, with SFL Missions under contract for Clusters 14 to 16; it publishes maps of GPS interference around conflict zones and sells maritime dark-vessel detection, and it won a $75M multi-year European defense ministry contract for air-defense and GPS monitoring. Unseenlabs in France sells maritime RF detection to European navies and fisheries authorities, and Spire adds RF collection alongside its AIS and weather payloads. The counterexample matters as much: Kleos Space listed on the ASX, launched four clusters, wrote off satellites that failed on orbit, ran out of cash in July 2023, and was delisted that August.

The capital intensity is low for space and the operating leverage is high, which is unusual in this sheet. Satellites cost single-digit millions each, a cluster is a few tens of millions including launch, and the same collected pass can be sold to several customers, so revenue scales with analytics and subscriptions rather than with the fleet. HawkEye 360 reported $98.7M of revenue in 2025, roughly double 2024, and $2.7M of net income against a $29M loss the year before, which makes it one of the few commercial remote-sensing businesses to reach profitability at all. The risk sits in customer concentration: US customers, mostly the US government, were 61% of that revenue and Japan another 16%, so this is a defense budget business with a commercial cost structure. Demand has been pulled up by GPS jamming becoming routine over Ukraine, the Baltic, and the eastern Mediterranean, which turned interference mapping into a recurring subscription. If you are evaluating a competitor, the questions are how many independent clusters it can afford to keep in orbit, whether it owns a signal archive deep enough to recognize a specific emitter it has seen before, and what happens to pricing when governments decide to fly the capability themselves.

5 minHawkEye 101: Intro to Radio Frequency | Episode 1HawkEye 360 · 5k+ views

6 minHawkEye 101: Exploiting RF Signals | Episode 2HawkEye 360 · 5k+ views

3 minIntroduction to TDOA geolocation with RFeye SiteCRFS - Extraordinary RF Technology · 5k+ views

Making the Invisible Visible: Precision RF-Emitter Geolocation from Space by the HawkEye 360 Pathfinder Mission (Small Satellite Conference) · Geolocation of RF Emitters with a Formation-Flying Cluster of Three Microsatellites (Small Satellite Conference)

Missile warning and tracking satellites carry infrared telescopes that watch for the heat of a launch and then try to hold a track on what the launch produced. The two halves of that job are not equally hard. A boosting rocket is a very bright, very hot plume, and it is easy to detect from geostationary orbit in a narrow shortwave-infrared band where the atmosphere is opaque enough to hide most of the Earth's own glare, which is what warning satellites have done since the 1970s. Once the booster burns out the target becomes a cold body a meter or two across, tens or hundreds of times dimmer, and it has to be picked out in mid- and long-wave infrared either against the black of space or, for a hypersonic glide vehicle flying at 20–60 km, against the moving clutter of the Earth's limb. Holding that track continuously and accurately enough to hand off to an interceptor is the capability the current architecture is being rebuilt to provide, and it is why the money moved from a handful of exquisite satellites in high orbit to a large mesh in low orbit. A satellite at about 1,000 km is roughly 35 times closer to the target than one at 35,786 km, and infrared irradiance falls with the square of range, so it sees the same object about 1,300 times brighter and can do the job with a far smaller aperture.

Infrared from orbit is the only sensor that sees a launch anywhere on Earth within seconds, with no basing rights, no radar horizon, and no ambiguity about whether something lit off. The proliferated low-orbit version adds two things the old architecture lacked: enough viewing geometry to hold a maneuvering target and enough satellites that losing several does not blind the system, which matters when the assets themselves are targets. What it costs is scale. A LEO satellite sees a given point for only a few minutes, so continuous custody of a target anywhere on Earth takes on the order of a hundred spacecraft rather than six, plus optical crosslinks to move the track between them and a fire-control-quality data path to the shooter. The failure modes are track handover and false alarms. A track dropped between two satellites is worse than no track, because a partial trajectory produces a confident, wrong impact prediction, and infrared sensors sensitive enough to see a cold body also see aircraft afterburners, industrial fires, and sun glint off cloud tops, so the discrimination burden lands on processing rather than optics.

This is not a purchase most organizations make, but the architecture choice inside it is live. Use high orbit when you want persistent stare over a hemisphere with a small number of assets and the target is a bright boosting missile: three or four geostationary satellites cover most of the globe continuously and nothing in low orbit matches that simplicity for strategic warning. Use proliferated low orbit when the target is dim, maneuvering, or both, because closing the range by a factor of 35 is worth more than any realistic aperture increase and because the mesh survives attrition. In practice the answer is both layers, which is what the US is buying: geostationary and polar satellites for strategic warning, a low-orbit tracking layer for custody and fire control, and a medium-orbit layer being added between them. If you are a supplier deciding where to compete, note that the bus is now a commodity and the infrared payload, its cryocooler, and the focal plane are not.

Roughly $1.5B per SBIRS geostationary satellite against about $49M per Tranche 3 Tracking Layer satellite · 35,786 km versus about 1,000 km, so roughly 35 times the range and 1,300 times the signal · more than 60 Tranche 1 vehicles on orbit by mid-2026, most of them Transport Layer · $3.5B for 72 Tranche 3 tracking satellites awarded in December 2025 · a further $1.75B in July 2026 for 36 accelerated satellites across four planes, delivery by the end of 2028 · 104 Tranche 3 tracking satellites in total.

Nothing here goes through the FCC. These are US federal systems, so their spectrum is assigned by NTIA through the interagency process rather than licensed commercially, their ITU filings are made by the administration on the government's behalf, and their frequencies sit in military allocations that no commercial operator can buy into. That is one reason the tracking layer standardized on optical crosslinks: a laser link needs no spectrum authorization at all and cannot be jammed the way an RF link can. Export control is absolute. Infrared focal plane arrays and missile-warning sensors sit squarely on the US Munitions List, spacecraft carrying them fall under Category XV, and the resulting track data is classified, so there is no commercial resale of this product and allied access runs through government-to-government agreements rather than contracts. The practical effect on the supply base is that a company wanting into this market needs a facility, cleared staff, and an ITAR compliance program before it can quote, which is a much higher barrier than the satellite bus itself.

SBIRS, whose geostationary satellites run around $1.5B each and which replaced the Defense Support Program, with hosted payloads in highly elliptical orbit covering the poles; Next-Generation OPIR, a program estimated at roughly $14B, with geostationary satellites from Lockheed Martin and polar payloads from Northrop Grumman. In low orbit, SDA's Tracking Layer runs from the Tranche 0 demonstrators through Tranche 1, which had more than 60 vehicles on orbit by mid-2026 across both its transport and tracking halves, and Tranche 2, to the Tranche 3 awards to Lockheed Martin, Rocket Lab, Northrop Grumman, and L3Harris in December 2025 and the accelerated award of $955M to L3Harris and $798M to Sierra Space in July 2026. The Missile Defense Agency's HBTSS prototypes, built by L3Harris and Northrop Grumman and launched in February 2024, were the specific test of whether a low-orbit sensor can produce a track good enough to guide an interceptor. A medium-orbit layer is now being added on top of both.

This is the clearest case in the sheet of a procurement model changing the industry underneath it. The traditional path was a cost-plus development contract with one prime, a decade of schedule, and unit costs above a billion dollars, which produced excellent satellites and very few of them. SDA buys instead through fixed-price other transaction agreements, two or more vendors per tranche, and two-year build cycles, which has pulled in suppliers that would never have won a classical missile-warning program and pushed the unit price to around $49M. The consequence for anyone building a business here is that revenue is recurring by construction: a low-orbit constellation with a five-year design life needs replacing continuously, so tranches are a standing order rather than a one-time buy, and Golden Dome has made that order larger. The risks are the ones that come with any single-customer defense business — the tranche cadence depends on annual appropriations, fixed-price agreements move schedule and cost risk onto the supplier, and a program restructure can strand a production line that was sized for a rate nobody now wants.

18 minHow America Detects Every Missile Launch On EarthSam Eckholm · 1m+ views

1 minSBIRS Mission Overview - Lockheed MartinMatthew Travis · 50k+ views

Hypersonic Missile Defense: Issues for Congress (Congressional Research Service) · Missile Warning Satellites: Space Development Agency Should Be More Realistic and Transparent About Risks to Capability Delivery (GAO)

A navigation constellation broadcasts a signal that says, in effect, "this is satellite 14 and my atomic clock reads exactly this." A receiver that hears four of them solves four equations for three position coordinates and its own clock offset, which is why every GNSS receiver is also a precision timing device and why the timing product matters at least as much as the positioning one. Making that work needs rubidium and caesium clocks on orbit that drift by about a nanosecond a day, since light travels 30 cm in a nanosecond and clock error turns straight into ranging error, plus a ground segment that continuously measures each satellite's true orbit and clock error and uploads corrections, and enough satellites that four are always above the horizon everywhere. GPS runs 31 operational satellites in six planes at 20,200 km with an 11 h 58 m period, which puts each satellite over the same ground track twice a day; Galileo sits at 23,222 km, GLONASS at 19,100 km, and BeiDou's medium-orbit shell near 21,500 km. The signals are free and open, nobody is billed, and the result is the only piece of space infrastructure that most of the world's economy depends on without knowing it exists. The sensing-and-navigation sheet covers what happens on the receiver side, including jamming and spoofing; this entry is about the constellation and who pays for it.

Global coverage, sub-meter ranging accuracy, timing to tens of nanoseconds, and a marginal cost to the user of nothing make this the highest-leverage infrastructure per dollar in space. Medium orbit is what makes the fleet small: 24 to 31 satellites cover the planet where a low-orbit system would need hundreds. The weakness is power at the receiver. A signal transmitted from 20,200 km arrives far below the thermal noise floor and is recoverable only through spread-spectrum processing gain, so a few watts from a jammer denies it across tens of kilometers, and that is now routine over Ukraine, the Baltic, and parts of the Middle East. The second weakness is institutional. These are 15-year assets built on decade-long procurements, so capability arrives slowly: GPS III took the constellation from around $500M per satellite on the first units down to roughly $200M on the last, the tenth and final one launched in April 2026, and the first GPS IIIF has already slipped from February to November 2026 over a traveling-wave-tube amplifier. The ground segment has been slower still, which is the usual reason a new signal sits unusable on orbit for years after it launches.

As a user you do not choose this, you choose how much to depend on it. The useful decision procedure is about the timing function, because that is where the exposure is largest and least visible: if a network, exchange, substation, or base station is taking time from GPS with no holdover, budget for a rubidium or chip-scale atomic clock that can free-run through an outage, and treat GNSS as the thing that disciplines it rather than the thing that provides it. If your platform operates anywhere contested, assume denial and design to inertial with GNSS as an aid. For a government deciding whether to build its own system, the honest test is sovereignty rather than performance, since GPS, Galileo, GLONASS, and BeiDou all work and a modern receiver uses all of them: you build your own when you cannot accept that another government could degrade or deny the signal, which is exactly the argument Europe used for Galileo. Low-orbit PNT is the interesting new option, but treat it as a resilience layer that supplements the medium-orbit systems rather than a replacement.

31 operational GPS satellites in six planes at 20,200 km · 11 h 58 m orbital period · user range error under 1 m and timing accuracy of tens of nanoseconds · roughly $500M per satellite on the early GPS III units falling to about $200M on the last · 15-year design life · four global systems, plus a 258-satellite commercial low-orbit constellation planned against the 24 to 31 satellites a medium-orbit system needs.

Navigation spectrum is the most heavily protected in the civil radio spectrum, because a receiver looking for a signal below the noise floor has no defense against an interferer that a normal service would shrug off. The signals sit in radionavigation satellite service allocations near 1,559–1,610 MHz, with GPS L1 at 1575.42 MHz, L2 at 1227.60 MHz, and L5 at 1176.45 MHz inside an aeronautical radionavigation band reserved for safety-of-life use, which is why ICAO rather than any spectrum regulator effectively sets the terms for aviation. Coordination between systems is done by treaty rather than by license: the 2004 US–EU agreement on GPS and Galileo settled signal structures so the two would be compatible and interoperable rather than mutually interfering, and it is the reason Galileo's open signal sits where it does. None of these constellations is FCC-licensed; each is filed by its own administration at the ITU and its frequencies assigned nationally. Export control still binds the user side: military M-code and its predecessor keying are ITAR-controlled, and receiver exports remain restricted above the long-standing speed and altitude thresholds that keep GNSS out of foreign missiles. A commercial low-orbit PNT operator faces the hardest version of all this, because a signal existing receivers can already use has to live in the same protected bands as the government systems, and getting permission to transmit there is a slower problem than building the satellites.

GPS, with 31 satellites operational and the final GPS III launched in April 2026 ahead of a GPS IIIF production run contracted to Lockheed Martin under options worth up to $7.2B; Galileo, run by the EU and EUSPA, now launching on Ariane 6 after the first Galileo mission on that vehicle in December 2025 and moving toward a second-generation fleet; GLONASS; and BeiDou, which adds geostationary and inclined geosynchronous satellites to its medium-orbit shell for extra regional performance. Regional systems include Japan's QZSS and India's NavIC. On the low-orbit side, Xona Space Systems launched its first Pulsar in-orbit validation satellite in June 2025 against a planned 258-satellite constellation, ESA is building an 11-satellite LEO-PNT demonstrator, and Iridium sells a timing and location service derived from its existing constellation.

This is a public good with no direct revenue, and the economics only make sense at the level of the economy it enables. Operating and modernizing GPS costs the US well over a billion dollars a year across satellites, the ground segment, and the user equipment programs, and every other system is funded the same way, by a government that decided it needed one. The downstream market — chipsets, receivers, correction services, and the applications built on them — runs into the hundreds of billions of euros a year by EUSPA's accounting, and none of that money flows back to the constellation operators. The commercially interesting layer is resilience rather than positioning, because the free signal is good enough and only fails under attack. That is what the low-orbit entrants are selling: much stronger received power from 20 times closer, faster geometry change that resolves carrier-phase ambiguities in seconds instead of minutes, and authentication that makes spoofing detectable. Whether it becomes a business depends on whether aviation, autonomy, and critical infrastructure will pay a subscription for something they are used to getting free, which is the same question every backup-utility business faces.

5 minHow an atomic clock works, and its use in the global positioning system (GPS)engineerguy · 1m+ views

7 minGPS and Relativity | How Time Dilation Affects GPS Accuracy | Special and General RelativityFlightInsight · 10k+ views

The Global Positioning System: space, control, and user segments (GPS.gov) · Global Positioning System Standard Positioning Service Performance Standard, 5th Edition (U.S. Department of Defense)

A ground station network is the set of tracking antennas that talk to a spacecraft, plus the scheduling, modems, and backhaul that turn a few minutes of visibility into data on a server. For anything in low orbit the fundamental limit is geometry. A satellite at 550 km circles the Earth about 15 times a day, but from a single mid-latitude site only 4 to 6 of those orbits come high enough over the horizon to use, and each usable pass lasts 8 to 12 minutes above a 5 to 10 degree elevation mask. Move the antenna to a high-latitude site and the numbers change completely: from Svalbard at 78 degrees north, a polar-orbiting satellite is visible on 10 to 14 of its 15 orbits, which is why the two most valuable pieces of ground-station real estate on Earth are Svalbard and Antarctica. What flows through that window is set by band and antenna size, roughly 100 Mbps to 1.2 Gbps on an X-band downlink and up to about 2.5 Gbps on Ka. Geostationary and deep-space missions have the opposite problem: the satellite is always visible from the right site, so the scarce resource is antenna time on a shared network rather than passes.

Renting antenna time removed what used to be a hard prerequisite for operating a satellite. A ground-station-as-a-service provider sells scheduled contacts by the minute across a network of licensed sites, so a company with one 6U CubeSat gets the same global coverage as an agency with its own stations, and the fixed cost of building, permitting, and staffing a site disappears. The weakness is that contact time, not sensor performance, is what usually limits revenue on a high-rate mission. A modern SAR or high-resolution optical satellite can collect far more data per orbit than it can push through 10 minutes of X-band, so the spacecraft spends most of its life with a full recorder, and adding aperture or collection capacity buys nothing until you add downlink. The other weakness is that a shared network is shared: contention for the same Svalbard passes at the same 10:30 local time crossing is real, on-demand pricing is two to three times reserved pricing, and a provider's coverage map is only as good as the countries that granted it site licenses.

Rent by default. If you fly fewer than a few dozen satellites, ground-station-as-a-service is cheaper and faster than building anything, and the pricing makes the comparison easy: reserved capacity runs about $3 per minute for narrowband and $10 for wideband, so even a heavy user buying 3,000 minutes a month of wideband spends around $360k a year, which is less than the lifetime cost of two staffed sites. Build your own when one of three things is true: you have enough constellation traffic that antenna-minute bills exceed the amortized cost of sites you would use around the clock, your data is classified or export-controlled and cannot transit a commercial provider's backhaul, or latency from collection to customer is your product and you need a station where the satellites are, not where the provider has one. Most large operators end up with both, owning stations at the high-value polar sites and renting the rest. If contact time is the binding constraint and you have already bought all the passes there are, the next lever is an optical downlink or an inter-satellite relay rather than another antenna.

4–6 usable passes a day from one mid-latitude site against 10–14 from Svalbard, out of about 15 orbits · 8–12 minutes per pass above a 5–10 degree elevation mask · X-band downlinks of roughly 100 Mbps–1.2 Gbps and Ka-band up to about 2.5 Gbps · reserved ground-station-as-a-service pricing near $3 per minute narrowband and $10 wideband · on-demand pricing near $10 and $22 · a 40 MHz instantaneous bandwidth threshold dividing the two tiers.

Every antenna needs a license in the country it stands in, and that portfolio of licensed sites is the actual product a ground-station provider sells. In the US an earth station is authorized under FCC Part 25, with a separate showing under the rules covering US stations that communicate with non-US satellites, and the same pattern repeats jurisdiction by jurisdiction, which is why a provider's coverage map has gaps that have nothing to do with orbital mechanics. The spectrum itself is shared rather than exclusive. The Earth exploration-satellite downlink band at 8,025–8,400 MHz sits alongside terrestrial fixed service in most countries, so a new site has to be coordinated against existing microwave links and sometimes cannot be sited where you want it. The clearest example of what that risk looks like is C-band: when the US reallocated 3.7–3.98 GHz to terrestrial 5G, teleports that had operated there for decades had to move or filter. Government work adds a second gate, since classified missions need an accredited facility and cleared operators, which is a large part of why defense customers still build stations instead of renting them.

KSAT operates the Svalbard and Antarctic sites that anchor polar coverage and sells passes worldwide. AWS Ground Station sells antenna-minutes with data delivered straight into a customer's cloud account. Viasat Real-Time Earth, Atlas Space Operations, Leaf Space, and RBC Signals compete on network coverage and price. NASA's Near Space Network and the Deep Space Network's 34 m and 70 m antennas are the science and exploration equivalent, and deep-space antenna time is rationed rather than sold. The instructive failure is Microsoft's: Azure Orbital Ground Station launched in 2020, was retired in December 2024, and its 10 tracking antennas were sold off to be operated by others, which is a useful signal about how thin the margins are in renting antenna time.

The cost structure is a fixed asset sold by the minute, so utilization is everything. A site costs money whether or not anyone books it, its capacity is capped at roughly the number of visible passes per day, and the price per minute is set by competition, so a provider's economics come down to how many customers it can stack onto the same antenna and how many sites it can share fixed engineering across. That is a hard business, and one of the two hyperscalers that entered it exited within four years. The durable advantages are geographic and regulatory rather than technical: a licensed antenna at 78 degrees north is scarce in a way that a licensed antenna in Virginia is not. On the cost curve, software-defined modems, cheaper antennas, and cloud backhaul keep pushing the price per minute down, and optical downlink is the real threat to the wideband tier, since a laser terminal moves in seconds what an X-band pass moves in ten minutes. If you are building a satellite business, model downlink as a per-image or per-scene cost from the start, because it scales with data volume and not with the number of satellites.

2 minIntroduction to AWS Ground StationAmazon Web Services · 10k+ views

2 minGround Stations Explained How Does Satellite Data Travel from Space to EarthGeoawesome · 5k+ views

18 minLEO Satellite Networks: Brief Introduction to Communications ChallengesIain Explains Signals, Systems, and Digital Comms · 10k+ views

State-of-the-Art of Small Spacecraft Technology: Ground Data Systems and Mission Operations (NASA) · Deep Space Network (NASA)

A flat-panel user terminal is the antenna that lets an ordinary customer use a satellite that is moving. A geostationary dish can be bolted to a wall and aimed once, because the satellite never moves in the sky. A low-orbit satellite crosses from horizon to horizon in a few minutes at several degrees per second, and the terminal has to follow it, then hand over to the next one without dropping the session. Doing that mechanically is possible but slow, noisy, and unreliable in the field, so the answer is an electronically steered phased array: hundreds or thousands of small radiating elements spaced about half a wavelength apart, roughly a centimeter at Ku band, each fed through a phase shifter so the beam can be pointed by changing the timing of the signals rather than by moving anything. That makes the terminal a semiconductor product. Most of its cost sits in beamformer chips, the RF front end, and a multilayer board with thousands of controlled-impedance feeds, so its price follows silicon economics rather than aerospace ones, and it falls with volume in a way that a satellite bus does not.

The strength is that once the array exists, everything downstream gets easier: no moving parts to fail, instant beam switching for handover, the ability to track two satellites at once during a make-before-break handover, and a package a customer can install on a roof in an hour. The weaknesses are cost, power, and heat. A Ku-band array needs an element roughly every centimeter, so aperture and element count scale together and there is no cheap way to build a big array; the electronics draw roughly 50 to 100 W continuously, which matters on a boat, an aircraft, or an off-grid site; and that power comes out as heat in a sealed flat panel sitting in the sun. The commercial failure mode is subsidy. If the terminal costs more to build than the customer pays, every new subscriber consumes cash up front, so growth is funded from the balance sheet until manufacturing volume closes the gap. That is exactly the trap that stalled earlier consumer satellite ventures, and it is why terminal cost, not satellite cost, is the gate on reaching consumers.

If your users are fixed, in one country, and happy with a dish on a pole, a conventional parabolic antenna is still cheaper by a wide margin and you should use it. Choose a flat-panel electronically steered terminal when the satellite moves, when the user moves, or when the installation cannot tolerate a mechanism: consumer low-orbit broadband, ships, aircraft, vehicles, and anything that has to switch between satellites or between constellations. If you are building a constellation, treat the terminal as a program on the critical path with a cost target attached, not as an accessory. A useful rule of thumb: your terminal bill of materials at your projected annual volume, minus what the market will pay for hardware, is the cash you will spend per subscriber before service revenue starts, and if that number is in the thousands you do not have a consumer business no matter how good the satellites are.

Starlink standard kits selling for roughly $349–599, against a production cost SpaceX put above $1,500 in the early years and $3,000 at the 2020 beta · high-performance business and maritime terminals at about $2,500 · aviation terminals around $150,000 for the hardware · elements spaced about half a wavelength apart, roughly 1 cm at Ku band, giving thousands of elements in a consumer panel · roughly 50–100 W of continuous power draw · satellites crossing at several degrees per second with a handover every few minutes.

User terminals are earth stations, and licensing millions of them individually would be impossible, so regulators grant blanket authorizations: one FCC license covers an unlimited number of identical terminals operating with a specified satellite system, which is why terminals are locked to a network and cannot simply be pointed at someone else's satellite. Terminals that move are a separate regime. Earth stations in motion on aircraft, ships, and vehicles carry their own rules on pointing accuracy and off-axis emissions, because a mispointed moving array interferes with adjacent satellites along the geostationary arc. Every country requires its own type approval and its own landing rights for the service, so terminals are geofenced in firmware and a unit that works in one country goes dark across a border, which is the main reason coverage maps have political holes. Aviation adds a certification cost that dwarfs the radio approval, since mounting anything on a pressurized fuselage needs a supplemental type certificate from the aviation authority, and that process usually takes longer than building the terminal.

Starlink's standard kit, the smaller Starlink Mini, and the high-performance dish sold for business, maritime, and aviation use are the volume reference. Amazon Leo has published a family of three terminals, from a small low-rate unit to an enterprise model, with a stated goal of building the standard one for a few hundred dollars. On the specialist side, ThinKom's mechanically scanned variable-inclination arrays dominate commercial aviation connectivity, Kymeta sells metamaterial-based electronically steered antennas for vehicles, and All.Space and ALCAN Systems build multi-beam and low-cost steerable panels. Military terminals from Viasat, L3Harris, and others cover the protected and multi-orbit end, where a single terminal has to work across several constellations and bands.

This is a consumer electronics business attached to a space business, and the two have opposite cost curves. Satellites get modestly cheaper with volume; phased arrays get dramatically cheaper, because the cost sits in silicon, board fabrication, and assembly yield, all of which improve with scale and process generation. SpaceX reported cutting terminal production cost by more than half within about six months of starting deliveries, from $3,000 to under $1,500, and the retail price has since sat in the $349–599 range with periodic promotions below that. The strategic consequence is that whoever ships the most terminals gets the cheapest terminals, which compounds. For a challenger the practical options are to buy terminals from a merchant supplier and accept a higher bill of materials, to target markets where the customer tolerates a $2,500 or $150,000 terminal because the alternative is nothing, or to sell wholesale into networks that already own the customer relationship. Watch the beamformer chip suppliers rather than the satellite manufacturers if you want an early read on where this cost curve goes next.

20 minPhased Arrays - Steering and the Antenna Pattern | An Animated Intro to Phased ArraysMarshall Bruner · 100k+ views

12 minHow Starlink Tracks 40 Satellites at Once (With No Moving Parts)The Backstory Lab · 1k+ views

Earth Stations in Motion: Report and Order (Federal Communications Commission) · Adaptive Antennas and Phased Arrays (MIT Lincoln Laboratory)

An optical terminal moves data between spacecraft, or from a spacecraft to the ground, on an infrared laser beam instead of a radio wave. The reason it works so much better is beam width. A radio antenna a few tens of centimeters across spreads its energy over degrees, so almost all of the transmitted power misses the receiver; a small telescope at 1,550 nm produces a beam tens of microradians wide, which puts nearly all of the power on the far terminal and buys orders of magnitude of link margin for the same watts. That margin is spent on data rate, and optical links now run at about 100 Gbps against a few Gbps for a good RF link. The cost of a narrow beam is that both ends have to find each other and stay pointed while separated by thousands of kilometers and closing at several kilometers a second, which makes acquisition and tracking the hard engineering problem rather than the optics or the modem. On a crosslink between satellites there is no atmosphere in the way and the link is excellent; on a downlink the beam has to come through cloud and turbulence, and cloud is opaque at these wavelengths, so the ground segment becomes a network of geographically separated sites rather than one antenna.

Three things make optical worth the trouble. It needs no spectrum license, so it sidesteps the coordination process that gates every RF service. It is very hard to jam or intercept, because an adversary has to physically be inside a beam that is meters wide at useful ranges. And a mesh of crosslinks decouples coverage from ground infrastructure, so a satellite over open ocean can still deliver its data by routing through neighbors. The weaknesses are pointing, weather, and supply. Acquisition takes seconds and a broken link has to be reacquired, so links across a moving constellation need constant re-planning; a cloudy ground site is a dead site, and getting to high availability takes five to ten stations spread far enough apart to be uncorrelated in weather; and the terminals themselves have been the bottleneck, because building thousands of precision optomechanical assemblies a year is a manufacturing problem the industry underestimated. The failure mode that bites operators is not a lost link but a link budget that closes in the lab and not in orbit, since vibration from reaction wheels and thermal distortion of the optical bench both degrade pointing in ways ground testing does not reproduce.

Use optical crosslinks whenever the constellation has to move data faster than it can fly it to a ground station, or when the traffic must not touch a foreign teleport. That is why every large low-orbit system has adopted them and why the military tracking and transport layers specified them. Use a laser downlink when a single high-rate satellite is collection-limited by contact time, since one optical pass can move what a dozen X-band passes would, but only if you can also fund several ground sites — a single optical ground station is a demonstration, not a service. Stick with RF when availability matters more than throughput, when the spacecraft is too small or too jittery to hold arcsecond pointing, or when the link has to work to a mobile user on the ground, because a laser cannot track a phone. In practice most systems fly both and use RF as the always-available control path with optical carrying the bulk data.

About 100 Gbps per optical link, with Starlink links reported peaking near 200 Gbps · three laser terminals per Starlink satellite and tens of thousands flying · more than 42 petabytes a day carried across that mesh · NASA's TBIRD downlinking at 200 Gbps from a 6U CubeSat in 2023 · deep-space optical returning 267 Mbps from about 31 million km · five to ten separated optical ground stations needed for high availability.

Optical links sit above 3,000 GHz, outside the frequency range the ITU allocates and coordinates, so a laser crosslink needs no filing, no coordination, and no spectrum fee. The flip side is that it gets no protection either: nobody has to coordinate with you, and there is no regulatory remedy if someone else's beam or a bright source interferes. What does bind is safety and space-object protection on the ground leg. In the US, anyone transmitting a laser upward has to coordinate with the Laser Clearinghouse so the beam is shuttered when a satellite crosses it, and lasers projected into navigable airspace require notification to the aviation authority, which is a per-site operational constraint rather than a one-time approval. Export control applies to the hardware: laser communication terminals are controlled items, which limits who can supply whom and is one reason the US, Europe, and Japan each built separate supplier bases. Standards have done the work that regulation usually does, with the Space Development Agency publishing an optical terminal specification so that terminals from different vendors interoperate, and that specification has effectively become the reference for the Western merchant market.

Starlink is the largest deployment by a wide margin, with three terminals per satellite, links running at about 100 Gbps, and more than 42 petabytes a day moving across the mesh; SpaceX has also built a smaller terminal for third-party spacecraft, rated around 25 Gbps at up to 4,000 km, with Muon Space among the first outside users. The Space Development Agency's Transport and Tracking layers require standards-compliant terminals across every vendor. The European Data Relay System uses geostationary laser relays to get Sentinel imagery down without waiting for a polar pass. On the demonstration side, NASA and MIT Lincoln Laboratory's TBIRD downlinked at 200 Gbps from a 6U CubeSat in 2023, and the Deep Space Optical Communications payload on Psyche returned 267 Mbps from roughly 31 million km, which is a rate no deep-space radio link approaches.

The terminal is a small, cheap-sounding box that has been very hard to build in quantity, and that gap defines the market. Constellation operators need thousands of units a year at prices that fit a satellite costing tens of millions, and the merchant suppliers built for tens of units a year at prices that fit a flagship mission. Mynaric, the best-known independent supplier and the source of terminals for Rocket Lab's $1.3B Space Development Agency satellite contracts, ran into exactly that squeeze and was bought by Rocket Lab in a deal announced in March 2025 and completed in April 2026 for about $155M, explicitly to fix a production constraint. The pattern is the same one that shows up elsewhere in this sheet: the operators that need the most terminals end up owning the factory. For anyone else, the practical question is whether a supplier can hold price and yield at rate, because a terminal that is 20% late or 20% over budget stops a satellite line, and there are only a handful of qualified sources. Long term, optical is the main threat to the wideband ground-station business, since a laser downlink moves in seconds what an X-band pass moves in ten minutes.

3 minFree-Space Optical Communications - TRENDING IN OPTICSEdmund Optics · 10k+ views

3 minTesting Space Lasers for Deep Space Optical Communications (Mission Overview)NASA Jet Propulsion Laboratory · 100k+ views

4 minLunar Laser Communications Demonstration (LLCD)MIT Lincoln Laboratory · 1k+ views

State-of-the-Art of Small Spacecraft Technology: 9.0 Communications (NASA) · Non-Coherent Optical Communications Physical Layer, CCSDS 141.0-B-2 (CCSDS)

Space situational awareness is the business of knowing where every object in orbit is, predicting where it will be, and telling operators when two of them are going to get close. Radars track low orbits and optical telescopes track high ones, the observations are fitted into orbits, and the resulting catalog is propagated forward and screened for close approaches. ESA counted 46,180 tracked objects and more than 17,000 tonnes of material in orbit as of 31 July 2026, of which roughly 16,000 are working satellites out of about 27,500 ever placed there; statistical models put the untracked population at around 54,000 objects larger than 10 cm, 1.2 million between 1 and 10 cm, and 140 million from 1 mm to 1 cm. The operational product is not the catalog but the conjunction warning, and the thing that makes it hard is uncertainty rather than detection. Three days out, an object's along-track position is known to hundreds of meters at best, so the screening question is never "will they hit" but "is the probability high enough to spend propellant and interrupt service", and most maneuvers are made against uncertainty rather than against a known collision.

The system works well enough that catastrophic collisions remain rare despite a fleet that has roughly tripled in a decade, and improvements are cheap relative to what they protect: better sensors shrink the covariance, and a smaller covariance directly means fewer maneuvers, more service uptime, and less propellant burned. The weakness is that the whole thing is advisory. Screening tells two operators they have a problem; nothing tells either of them who moves. Coordination happens by email, phone call, and bilateral agreement, and it fails exactly when it matters most, which is when one of the two objects is dead and cannot move at all. Roughly half of the tracked population is debris that nobody controls, so a warning about it is only actionable for the other party. The second weakness is data quality: the public catalog is deliberately imprecise for some objects, operators know their own satellites far better than any external tracker does, and reconciling those two views is a standing source of false alarms and missed ones.

Everyone flying anything needs conjunction screening, so the real decision is how much to buy beyond the free government service. Take the public catalog and the standard conjunction messages as a baseline. Pay for a commercial provider when one of three conditions holds: you fly enough satellites that maneuver decisions are a real operating cost and tighter covariances save you propellant and downtime, you operate somewhere the public catalog is thin, such as geostationary or very low orbits where objects decay quickly, or you need to know when another spacecraft maneuvers rather than only where it was. If you run a large constellation, build automated screening and maneuver planning into the operations system from the beginning, because the maneuver rate grows faster than the fleet does and manual handling stops scaling early. And if you are relying on a warning to save you, remember that no warning exists for the 1.2 million objects between 1 and 10 cm, which are large enough to destroy a satellite and too small to track.

46,180 tracked objects and more than 17,000 tonnes of material in orbit as of 31 July 2026 · about 16,000 working satellites out of roughly 27,500 ever placed in orbit · modeled populations of 54,000 objects above 10 cm, 1.2 million from 1–10 cm, and 140 million from 1 mm to 1 cm · positional uncertainty of hundreds of meters three days ahead, which is what forces a maneuver · relative velocities around 10 km/s in a low-orbit conjunction · roughly 300,000 Starlink collision-avoidance maneuvers reported across 2025.

There is no space traffic authority, and that is the central regulatory fact about this category. National licensing does part of the job: the FCC requires an orbital debris mitigation plan and a demonstrated collision-avoidance capability before it will authorize a US system, and other regulators have similar conditions, but those are conditions on a license rather than rules of the road, and they bind only the operators each regulator licenses. The UN guidelines on long-term sustainability are voluntary. Tracking data itself comes with strings: the US public catalog is published by the Space Force on space-track.org, higher-accuracy data requires a sharing agreement, and some objects are deliberately withheld or degraded for national security reasons, so the free data is not the best data. The US attempt to move civil space traffic coordination out of the military into the Commerce Department produced TraCSS, which reached initial operating capability and then had its funding proposed for cancellation in the fiscal 2026 budget request on the argument that commercial providers could do the job. Whether a neutral civil coordinator exists is still unsettled, and it matters because the alternative is the largest operator's own system becoming the de facto standard.

The US Space Surveillance Network combines radars including the S-band Space Fence at Kwajalein with the GEODSS optical telescopes for high orbits, and its 19th Space Defense Squadron issues the conjunction messages most of the world's operators act on. Commercially, LeoLabs runs a network of phased-array radars sized for low orbit, Slingshot Aerospace and ExoAnalytic Solutions operate global optical telescope networks that watch geostationary objects, and COMSPOC sells the astrodynamics software behind many of these catalogs. Operators have started building their own: SpaceX unveiled a tracking system of its own in January 2026 and signed a cooperative agreement with NOAA on automated collision avoidance. The events that shaped the field are the 2007 Chinese anti-satellite test, the 2009 Iridium 33 and Cosmos 2251 collision, and the 2021 Russian anti-satellite test against Cosmos 1408, each of which added thousands of tracked fragments.

The hard part of this business is that the baseline product is free. A government funds the sensors, publishes a catalog, and issues warnings at no charge, which caps what anyone can charge for the same thing. Commercial providers therefore sell the margin above it: tighter covariances, faster revisit on a specific object, custody of maneuvering spacecraft, coverage where the public catalog is weak, and analytics that turn conjunction messages into maneuver decisions. The customers who actually pay are defense and intelligence agencies, large constellation operators for whom avoided maneuvers have a measurable cost, and insurers and regulators at the margin. Sensor hardware is no longer the barrier, since a phased-array radar site costs a few million dollars and a telescope network far less, so the durable asset is the catalog history and the software that maintains it. The unsolved commercial problem is the same as the unsolved policy problem: everyone benefits from a well-maintained shared picture, nobody's individual business case justifies paying for all of it, and if public funding for a neutral system goes away the picture ends up owned by whoever flies the most satellites.

2 minDodging debris to keep satellites safeEuropean Space Agency, ESA · 10k+ views

4 minDARPA's Space Surveillance Telescope: What It DoesDARPAtv · 10k+ views

NASA Spacecraft Conjunction Assessment and Collision Avoidance Best Practices Handbook (NASA) · Space Situational Awareness: DOD Should Evaluate How It Can Use Commercial Data (GAO)

An orbital transfer vehicle is a small spacecraft that rides up with a payload and then takes it the rest of the way. It exists because rideshare drops everyone at the same place at the same time: a Transporter mission releases a hundred-odd satellites into one orbit, they separate from each other only through slow natural drift, and none of them is where it actually wanted to be. Two distinct classes solve two distinct problems. Low-thrust last-mile tugs carry a stack of payloads, use electric or small chemical propulsion to shift altitude and phase within low orbit over days to weeks, and release each customer individually into its own slot. High-energy chemical kick stages do the opposite job, delivering several tonnes of delta-v-hungry transfer in a single day, from low orbit to geostationary transfer, to GEO directly, or onto a lunar trajectory. The physics that separates them from a dedicated launch is worth understanding: a tug changes altitude and phase very cheaply, and altitude differences then let satellites drift apart in right ascension for free, but changing inclination costs roughly 130 m/s per degree in low orbit, so no tug will fix an orbit plane you should have launched into. Engine details belong in the propulsion-systems sheet; what matters here is the delivery service.

The last-mile version is the one part of in-space logistics that is a real, flown business. It buys precise altitude, useful separation between satellites in a constellation, individual deployment rather than a shared dispenser window, and often a hosted-payload slot for a customer who does not want a spacecraft at all. What it costs is time and mass. Electric transfer takes weeks to months, during which the payload is out of service and exposed to an environment it was not built to sit in, and the tug's own dry mass and propellant come out of the rideshare capacity the customer paid for. The failure mode is a stranded payload: the tug becomes a single point of failure between a working satellite and its orbit, and early missions in this category have left customer spacecraft undeployed or deployed late after tug anomalies. The high-energy class has a different problem, which is that nothing at the size that matters has flown yet. Impulse Space's roughly 300 kg Mira has, but the several-tonne stage that would carry a satellite from low orbit to geostationary has not, and that is the mission the business case rests on. The performance claims are credible and the physics is ordinary, but a $300M geostationary satellite will not be the first payload on a new stage, so the class has to buy its way onto orbit with demonstration flights before anyone valuable rides it.

Buy last-mile transfer when the rideshare goes to roughly the right orbit and you need a specific altitude, a specific phase, or separation from the hundred satellites you launched with. At $0.5–3M on top of the seat, it is far cheaper than the $7.5–8.5M a dedicated small launch costs, and it usually beats it on schedule too because rideshares fly more often. Buy dedicated launch instead when you need a plane or a local time of ascending node that no rideshare serves, because that is an inclination change and a tug cannot afford it. Skip the tug entirely when your satellite already carries propulsion, since an onboard electric thruster does the same altitude and phasing job over the same timescale for less money; the tug wins when you want to fly a bus with no propulsion at all, or when you need to be on station before commissioning starts. For high-energy transfer, treat it as a way to buy a GEO or lunar delivery off a cheap low-orbit launch rather than paying for a heavy-lift direct injection, and wait for a demonstrated flight before putting anything expensive on one.

Last-mile deployment costing roughly $0.5–3M on top of a rideshare seat, against $7.5–8.5M for a dedicated small launch · 22 ION Satellite Carrier missions flown by March 2026 · Mira at roughly 300 kg wet · Helios advertised at 5-plus tonnes from low orbit to GEO in under 24 hours · inclination changes costing about 130 m/s per degree in low orbit · D-Orbit revenue of roughly $52M in 2025.

Orbital transfer sits in a genuine regulatory gap. FAA Part 450 licenses launch and reentry, and a tug that comes back needs a reentry license, but nothing in that framework covers maneuvering between the two, so there is no US agency that clearly authorizes on-orbit transportation as an activity. What fills the gap in practice is the FCC, because the tug needs a space station license for its own telemetry and command link and that license carries orbital debris and disposal conditions, which makes a spectrum regulator the de facto approver of the mission profile. Each payload the tug deploys needs its own authorization and its own registration by its own country, so a tug operator is effectively checking its customers' paperwork, and regulators have begun asking whether it did. Carrying foreign customers adds export-control exposure, since integrating with and maneuvering a foreign spacecraft can involve controlled technical data. The unsettled question is mission authorization for novel on-orbit activity generally, which affects this entry, servicing, refueling, and debris removal alike, and which the US has been debating for several years without resolving.

D-Orbit's ION Satellite Carrier is the flight-proven case, with 22 missions by the Wayfinder flight that launched on Transporter-16 in March 2026, deploying customer satellites and hosting payloads on the same vehicle. Exolaunch's Reliant and Momentus's Vigoride cover the same last-mile niche, and Firefly's Elytra is a launch company selling transfer as an extension of its own rocket. On the high-energy side, Impulse Space's Mira is a roughly 300 kg vehicle that has flown, while Helios is a methalox kick stage designed to move 5-plus tonnes from low orbit to GEO in under a day, with the first Caravan mission targeted for the third quarter of 2026 after slipping from an earlier date and further Caravan flights planned for 2027 and 2028. Orbit Fab plans to host a refueling depot on a Mira, which is how this entry connects to on-orbit refueling.

This is the only part of in-space logistics with meaningful revenue today, and it is still a small business: D-Orbit, the clear leader, reported roughly $52M in 2025 after raising a €150M Series C. The cost structure is awkward because the vehicle is expendable. Each tug flies once, so its full build cost is amortized across a single mission's worth of customers, which puts a floor under pricing that reusability would remove — and reusability requires refueling, which does not exist yet. Competition comes from three directions at once: launch providers adding transfer to their own vehicles, satellite builders integrating cheap electric propulsion so no tug is needed, and dedicated small launch at the top end. The realistic path to a bigger business is either volume in constellation deployment, where one tug serves one customer many times, or the high-energy market, where the value per kilogram delivered is much higher because the alternative is a heavy-lift launch. Getting there takes a successful demonstration flight and then a first commercial customer willing to be second, and until that happens the high-energy half of this category is a set of well-engineered plans rather than a market.

2 minSpace Tug, an autonomous spacecraftAirbus Defence and Space · 10k+ views

4 minInside look: Spaceflight Sherpa-FXGeekWire · 1k+ views

State-of-the-Art of Small Spacecraft Technology: 10.0 Integration, Launch, Deployment, and Orbital Transport (NASA) · Commercial Space: Federal Regulation, Oversight, and Utilization (Congressional Research Service)

Satellite servicing means flying a spacecraft up to another one and doing something useful to it. Almost every candidate client was built decades before anyone planned to service it, so there is no docking port, no grapple fixture, and no cooperative navigation aid. The trick that made it work is that geostationary satellites share two features by construction: a liquid apogee engine with an open nozzle at the aft end, and a launch vehicle adapter ring around it. Northrop Grumman's Mission Extension Vehicle inserts a probe into the apogee engine nozzle throat, expands it to grip, and pulls itself down onto the ring, which gives a rigid mechanical connection to a satellite that was never designed to offer one. Once attached, the servicer takes over station-keeping and attitude control for the combined stack, so the client's exhausted propellant stops being the thing that ends its life. The newer approach separates the two functions: a Mission Robotic Vehicle with robotic arms carries small Mission Extension Pods, installs one on each client, and moves on, so one expensive robot serves many clients instead of one servicer being consumed per satellite.

The value is easy to state. A geostationary satellite usually dies with a working payload and an empty tank, and replacing it costs $300–600M including launch, more for a high-throughput satellite, plus three years of build time, so buying six more years for tens of millions is a good trade whenever the payload is still worth operating. Docking is also demonstrated rather than theoretical, which puts this ahead of everything else in this part of the sheet. The weaknesses are the client population and the operation itself. Most aging geostationary satellites are aging commercially as well as physically: a fifteen-year-old wideband transponder payload competes against high-throughput satellites carrying hundreds of gigabits, so extending it prolongs an asset that is already losing. And the docking is genuinely hard. The servicer closes on a client that is drifting and slowly tumbling, with no cooperative markers, and the failure mode is not a missed approach but a collision that destroys two satellites and salts a valuable orbital slot with debris. That risk is why every step so far has been slow, ground-supervised, and heavily rehearsed.

Life extension is worth buying when three conditions hold at once: the payload still sells, the orbital slot is worth keeping, and propellant is the only thing that has run out. If the payload is obsolete, extension is money spent to keep a bad asset alive and a replacement is the right answer. If the satellite has a failed component rather than an empty tank, extension does not help either, since nothing on offer today repairs electronics. Check the timing too, because a servicing vehicle typically needs a year to climb to GEO and then time to reach each client, so this is a decision made three or more years before the fuel runs out, not a rescue. For defense customers the calculation is different and often better: the product being bought is inspection, relocation, and the ability to keep a maneuvering asset maneuvering, and that has value independent of the client's commercial revenue.

MEV-1 docked Intelsat 901 in February 2020 and MEV-2 followed in 2021, each on a fixed multi-year term · the Mission Robotic Vehicle launched 21 July 2026 carrying three Mission Extension Pods, with about a year of orbit-raising ahead of it · each pod adds at least six years of life to a roughly 2,000 kg geostationary satellite · life extension in the tens of millions against $300–600M for a replacement satellite and its launch · $54.5M for a US Space Force geostationary servicing vehicle covering at least five years of operation.

Servicing runs into the same US mission-authorization gap as orbital transfer: no agency licenses on-orbit operations as such, so the FCC's space station license and its debris and disposal conditions become the effective approval, and NOAA gets involved when the servicer carries cameras that image other spacecraft. Regulation has visibly shaped the product. The Mission Extension Pod is designed as a customer-owned, customer-controlled propulsion module, which means once installed it becomes part of the client's licensed spacecraft rather than a separate operator's asset flying in formation, and that avoids a set of licensing and liability questions that would otherwise have to be answered. The orbital slot and its ITU rights stay with the client operator throughout, since the servicer is providing propulsion and not taking over the frequency assignment. The bigger unsettled area is rendezvous and proximity operations, where there are no international rules at all: nothing prohibits approaching another operator's satellite uninvited, several such approaches by state actors have caused diplomatic complaints, and the industry's own norms of behavior are voluntary. Export control binds who can service whom, since docking with a foreign-owned satellite involves controlled technical data.

Northrop Grumman's SpaceLogistics is the only company with a servicing track record: MEV-1 docked Intelsat 901 in February 2020, the first docking with a satellite not designed for it, MEV-2 docked a second Intelsat spacecraft in 2021, and the Mission Robotic Vehicle launched on a Falcon 9 on 21 July 2026 with three Mission Extension Pods and robotic arms developed under a DARPA program, with NASA supplying software and simulation support. Optus is the announced pod customer for its D3 satellite, launched in 2009. Starfish Space is the main challenger, with a $54.5M US Space Force contract for a geostationary Otter and further Otter missions planned for NASA and Intelsat, but its Otter Pup 2 demonstrator shows how slow this is in practice: launched in June 2025, it lost its original docking partner in late 2025, was retargeted in March 2026 at Gilmour Space's ElaraSat, and had still not docked more than a year after launch. NASA's own OSAM-1 servicing mission was canceled in 2024 after years of cost growth.

The first real market is geostationary life extension, and it is small. Public reporting put the MEV-1 service near $13M a year, which is excellent value against a $300–600M replacement, but the number of clients that qualify at any moment is a few dozen worldwide and shrinking as high-throughput satellites make old payloads uncompetitive. The pod architecture is the industry's answer: spreading one robot's cost across many small, cheap pods improves the unit economics enough that a client with a modest remaining revenue stream can justify it. The larger and more likely market is government. Defense buyers will pay for inspection, relocation, and sustained maneuver in geostationary orbit for reasons that have nothing to do with a payload's revenue, and that is where the funded contracts are. For this category to become a real business rather than a demonstration program, three things have to happen: the MRV has to install a pod robotically and prove one servicer can handle many clients, the per-client cost has to fall to single-digit millions, and refueling has to arrive, because until a servicer can be refilled it is itself a consumable and every mission carries its own vehicle cost.

3 minNext Generation of Satellite Servicing Products: Mission Robotic Vehicle and Mission Extension PodsNorthrop Grumman · 10k+ views

1 minNorthrop Grumman robot docks with satellite to extend its missionVideoFromSpace · 10k+ views

2 minMEV Mission ProfileNorthrop Grumman · 1k+ views

In-Space Servicing, Assembly, and Manufacturing (ISAM) State of Play, 2025 Edition (NASA) · In-Space Servicing, Assembly, and Manufacturing: Benefits, Challenges, and Policy Options (GAO)

On-orbit refueling means transferring propellant from one spacecraft to another after launch, and as of August 2026 nobody has done it to an operational client satellite. Two very different problems sit under the same name. Storable hypergolic transfer moves hydrazine or a bipropellant pair between spacecraft in geostationary orbit at ordinary temperatures, which is mostly a plumbing, docking, and interface problem. Cryogenic transfer moves liquid oxygen and methane in bulk, and the fluids fight back: they boil off through any heat leak, the receiving tank has to be chilled down before it will accept liquid, and the chilldown itself costs propellant. Both share the zero-gravity difficulty that liquid does not sit at the bottom of a tank, so the propellant has to be settled with a small acceleration burn or held at the outlet by surface-tension hardware before anything will flow. And both run into the interface problem: almost nothing already in orbit has a fill port. A geostationary satellite has a ground fill-and-drain valve that was safety-wired shut before launch and never meant to be opened by a robot, which is why the industry has split between adding standardized ports to new satellites and building robots that can defeat the wire and connect to the valve that is already there.

The reason to care is that propellant is what limits maneuver. A satellite designed for fifteen years of station-keeping spends its whole life protecting a reserve, and a satellite that expects to be refueled does not, which changes how aggressively it can be flown. For defense that is the entire argument: an asset that can maneuver without husbanding fuel is a different asset. The weakness is arithmetic, and it is worth doing rather than taking on faith, because nobody publishes the comparison. Orbit Fab's published target is around $20M for 100 kg of hydrazine delivered in geostationary orbit. Work that through the rocket equation and 100 kg on a 2,000 kg satellite is roughly 100 m/s of delta-v, which at the 50 m/s a year that north-south station-keeping costs is about two extra years. A Mission Extension Pod delivers at least six years, and launching that same 100 kg of propellant with the satellite in the first place costs a few million dollars at direct-GEO launch prices. So refueling as a life-extension product is currently the expensive option, and it only clearly wins when the satellite is already in orbit, when you could not have known in advance how much delta-v you would need, or when the customer wants maneuver rather than lifetime. The other weakness is that none of it has been demonstrated end to end.

Right now, plan for it rather than buy it. If you are designing a geostationary or maneuvering spacecraft today, fitting a standardized refueling port is cheap insurance: it adds a few kilograms and keeps an option open that costs nothing if you never use it, and the US Space Force has started requiring the capability on new satellites, so for a defense supplier it is becoming a requirement rather than a choice. Do not build a business case that depends on being refueled, because the depots, the tankers, and the price are all unproven. If the goal is simply more years on a geostationary satellite whose payload still sells, life extension with a pod is the better-understood purchase today. If the goal is a spacecraft that maneuvers a lot and unpredictably, refueling is the only mechanism that gets you there, and the honest position is that you are betting on a capability arriving rather than buying one. For cryogenic transfer, the question does not really arise commercially: it is a program requirement inside one architecture rather than a service anyone sells.

$44.5M for the original single Space Force refueling experiment, now split into Tetra-5 in 2026 and Tetra-6 in 2027 · a published target near $20M for 100 kg of hydrazine delivered in geostationary orbit · 100 kg on a 2,000 kg satellite giving roughly 100 m/s of delta-v, about two years at 50 m/s a year of station-keeping (derived here, not published) · zero operational client refuelings performed as of August 2026 · on the order of 8–16 Starship tanker flights per crewed lunar mission.

Refueling falls in the same gap as the rest of this part: FAA Part 450 licenses launch and reentry but not what happens between them, so there is no agency that authorizes propellant transfer as an activity and the FCC's space station license again does the practical gatekeeping. Where Part 450 does bite is at the launch site, because a tanker is a pressurized vessel full of hazardous propellant and range safety limits how much of it can ride with other payloads, which constrains depot architectures more than most people expect. In the absence of a standards body, the US Space Force has effectively created the standard by endorsing two commercial interfaces, Orbit Fab's RAFTI port and Northrop Grumman's Passive Refueling Module, as acceptable solutions, and a government customer declaring which connector counts is what unlocks the rest. Export control applies to the hardware and to docking with any foreign spacecraft. There is also a norms problem with no regulation attached: a vehicle that can rendezvous, dock, and transfer fluid to a satellite is mechanically indistinguishable from one that could do something unfriendly to it, and nothing in international law distinguishes the two.

The US Space Force's Tetra-5, launching in 2026, puts up two small satellites carrying Orbit Fab's RAFTI ports along with an Astroscale US propellant shuttle and an Orbit Fab depot, with the shuttle collecting fuel from the depot and delivering it to a client. Tetra-6 follows in 2027 to test Northrop Grumman's Passive Refueling Module against its ROOSTER-5 spacecraft. Orbit Fab flew a small tanker demonstrator in 2021 and has selected Impulse Space to host a geostationary depot on a Mira vehicle. NASA's Robotic Refueling Mission experiments on the ISS demonstrated the robotic steps of cutting lock wire and accessing a conventional fill-and-drain valve, and a NASA-commissioned analysis concluded there are no technological barriers to refueling geostationary satellites. On the cryogenic side, SpaceX has transferred propellant between internal Starship tanks and a ship-to-ship transfer demonstration has been expected in 2026; the Artemis lunar lander architecture depends on it working at a cadence of many tanker flights per mission.

This is a pre-revenue category with a textbook chicken-and-egg problem. Nobody fits a refueling port because there are no depots, and nobody funds a depot because there are no ports, and the interface has to be agreed before either side can commit. Government is breaking the deadlock the way it usually does, by paying for the demonstrations and then writing the capability into requirements for new spacecraft, which turns a speculative feature into a spec line and gives suppliers a reason to standardize. The first real market is therefore defense: geostationary and maneuvering national security satellites whose value comes from being able to move, where the customer does not need the price per kilogram to beat launching a fuller tank. Commercial demand comes later and only if delivered propellant gets much cheaper than $20M per 100 kg, which requires depots that are launched full and used many times rather than tankers flown per customer. What would move this from development to service is a clean Tetra-5, a second successful transfer to a satellite the operator actually cares about, and a published price that a commercial fleet planner can put in a model. Until then, treat any business plan that assumes refueling as a plan with a dependency it does not control.

9 minDemonstrating Rocket Fuel Transfer in SpaceNASA Johnson · 500k+ views

15 minStarship Orbital RefuelingEager Space · 50k+ views

Lessons for Future In-Space Telerobotic Servicing from Robotic Refueling Mission (NASA) · Guidelines for In-Space Cryogenic Propellant Transfer (NASA)

This is two problems that share a name and have opposite economics. Getting your own satellite out of orbit is cheap and well understood: lower perigee into the atmosphere with about 120 m/s of delta-v from 550 km, or skip the propulsion and fit a drag sail that increases area enough to pull the spacecraft down in a few years, or simply fly low enough that atmospheric drag does the work for free. In geostationary orbit the equivalent is cheaper still, a raise of roughly 300 km into a graveyard orbit that costs about 11 m/s, or around three months out of the 50 m/s a year that station-keeping already consumes. Removing somebody else's dead satellite is a completely different activity. The target is uncontrolled, usually tumbling, has no grapple fixture and no working transponder, and a chaser has to match orbits, characterize the tumble, close in, capture a body that may weigh several tonnes, damp out the combined rotation without breaking anything, and then deorbit the stack. ESA counted more than 17,000 tonnes of material in orbit as of 31 July 2026, most of the mass in spent rocket bodies, and the number of objects successfully removed by anyone so far is zero.

Designing for disposal is one of the highest-return engineering decisions available: a few kilograms of drag sail or a modest propellant reserve costs almost nothing against the satellite, and it is now a licensing requirement rather than a courtesy. The problem is the tail. Satellites that fail early fail before they can deorbit, so a constellation with a 2% infant-mortality rate leaves dead objects behind no matter how good its disposal plan is, and those are exactly the objects nobody can move. Active removal is where the weaknesses concentrate. It is very expensive per object, it has never been completed, and the capture step is the part that keeps failing to arrive: Astroscale's ADRAS-J flew to within 15 m of a spent H-IIA upper stage in 2024 and produced excellent imagery, then completed operations without attempting a capture, which is a fair summary of where the field is. There is also a self-defeating quality to the risk. A capture attempt that goes wrong on a nine-tonne rocket body creates far more debris than it removes, so every mission is deliberately cautious, which keeps costs high and the demonstrated rate near zero.

For your own spacecraft, decide disposal at design time and pick by orbit. Below roughly 400 km, natural decay handles it and you need nothing. Between 400 and 700 km, either reserve the propellant or fit a passive device, and prefer the passive one on a small satellite, because it still works after the spacecraft dies and reserved propellant does not. Above about 700 km, drag is too weak to rely on and you need real propulsion with the reserve protected by the operations plan, not just the design. In geostationary orbit, budget the graveyard raise and do it while the tank still has margin, since the classic failure is an operator squeezing extra months of service and then finding it cannot reach the disposal orbit. Active removal is not something you buy today. If you are planning a large constellation, the useful question is not who will clean up after you but what fraction of your fleet will die before it can deorbit itself, because that number, multiplied by a future removal price, is the liability you are actually creating.

About 120 m/s to deorbit from 550 km, or a few kilograms of drag sail instead · about 11 m/s to raise 300 km into a geostationary graveyard orbit, roughly three months of the 50 m/s a year station-keeping budget · the FCC's 2022 rule expecting disposal within 5 years of mission end in low orbit, against the previous 25-year guideline · €86M contracted for ClearSpace-1 to remove a single object and around ¥13.2B for Astroscale's ADRAS-J2 removal mission · 15 m closest approach achieved by ADRAS-J in 2024, with no capture · more than 17,000 tonnes of material in orbit and no object yet removed.

Disposal is the one part of space environmental policy that regulators have actually enforced, through licensing. The FCC's 2022 rule cut the expected post-mission disposal window in low Earth orbit from the long-standing 25-year guideline to 5 years, and a US license now requires a credible disposal plan with the propellant budget to back it, which is why disposal shows up in mass statements rather than in slide decks. ESA has gone further with its Zero Debris approach and charter, aiming at essentially no new debris generation, and it applies the standard to its own missions first. Removing other people's objects runs into a harder legal wall. Article VIII of the Outer Space Treaty leaves jurisdiction and control over a space object with its state of registry indefinitely, so a dead satellite or a spent stage still belongs to somebody, and capturing it without that state's consent is not permitted regardless of how much of a hazard it is. The Liability Convention then makes the remover's launching state liable if the attempt causes damage. Add that a vehicle able to capture an uncooperative satellite is functionally an anti-satellite weapon, and it becomes clear why removal missions are national programs targeting their own countries' debris rather than a service sold on the open market.

Astroscale is the most advanced operator, with ELSA-d demonstrating magnetic capture of a prepared target in 2021, ADRAS-J performing rendezvous and inspection of a Japanese H-IIA upper stage in 2024 to within 15 m, and ADRAS-J2 contracted under Phase II of JAXA's Commercial Removal of Debris Demonstration at around ¥13.2B to return to the same object and actually remove it, with launch planned for the 2027 fiscal year. ESA's ClearSpace-1 was contracted in 2020 for €86M, had to change targets in 2024 after its original one was struck by debris, and now aims to capture and deorbit the PROBA-1 satellite with a four-armed capture mechanism. On the passive side, drag-sail devices such as ADEO and Terminator Tape are flying as bolt-on disposal hardware for small satellites, and large constellation operators deorbit satellites routinely under their own propulsion, which is by far the largest volume of disposal actually happening.

The two halves of this entry have opposite cost curves and only one of them is a business. Designing for disposal costs a few kilograms and is mandatory, so it is a line item rather than a market. Active removal costs tens of millions per object, and the benefit is spread across every operator in that orbit while the cost falls on whoever pays, which is the textbook definition of something markets do not fund. The buyers so far are space agencies buying demonstrations of their own countries' debris, and that is likely to remain the first real market: a state removing a specific large object it is responsible for, priced as a program rather than a service. The plausible second market is constellation operators buying disposal for the small fraction of their fleet that dies before it can deorbit itself, which becomes real if per-object cost falls to single-digit millions. Getting there needs a servicer that captures several objects per mission instead of one, which in turn needs refueling, so this category sits downstream of a capability that does not exist yet. Proposals to fix the incentive problem — orbital-use fees, deposits refunded on successful disposal, insurance-linked requirements — are all still proposals.

9 min4 Ways To Capture Space DebrisDr Ben Miles · 10k+ views

3 minADRAS-J | Active Debris Removal by Astroscale - Japan | Concept of OperationsAstroscale · 10k+ views

3 minAstroscale's Space Debris Removal Mission, ELSA-d - ConOps VideoAstroscale · 50k+ views

ESA's Annual Space Environment Report (European Space Agency) · Mitigation of Orbital Debris in the New Space Age, Second Report and Order (Federal Communications Commission)