The FCC Asked Cowboy Space How Big Its Satellites Are

By Nick David, Editorial Lead, SatNews

The Bottom Line

On August 28 the Space Bureau sent Cowboy Space an eleven-part deficiency letter and gave it until today to answer. Item 11(c) asks for the physical dimensions of the satellites, “including the dimensions of the solar panels.”

Bhatt’s own public numbers imply 400,000 to 500,000 tonnes of spacecraft. The world launched 3,193.4 tonnes in 2025. Neither figure appears in the application, because Schedule S has no field for either one.

The Bureau could not ask about mass directly, so it asked through the orbital debris rules. That is the tell: the form that licenses a constellation cannot measure the one quantity that decides whether it can ever be built.

Baiju Bhatt went shopping for rockets and came back with a rocket program.

The Robinhood co-founder renamed his company Cowboy Space Corporation on May 8 and raised $275 million at a $2 billion post-money valuation, led by Index Ventures, to put megawatt-class AI compute in orbit. Then it did the thing no satellite operator does when the manifest is tight: it started building its own launch vehicle.

Four months later, the Commission has begun asking how big the payload actually is.

On August 28, Carolyn A. Mahoney, acting chief of the Space Bureau’s Satellite Programs and Policy Division, wrote to Cowboy’s chief operating officer and chief legal officer Joseph Yaffe with eleven numbered demands. The opening line sets the posture: the information is required “in order to render the Application suitable for further processing.” The letter closes by citing 47 CFR § 25.112(c), under which failure to respond substantially “may result in dismissal of the application.” The response was due September 11. As of 4:30 p.m. Eastern that day, nothing had been filed: the docket still carried only the five original May 11 documents, and the application still read “pending review.”

Item 11(c) is the one worth reading twice.

Four hundred thousand tonnes

Cowboy’s application is public. ICFS File No. SAT-LOA-20260323-00135, filed May 11 and accepted for filing June 12, requests up to 20,000 satellites under the system name Stampede, in dawn-dusk sun-synchronous shells between 700 and 1,000 kilometers, with the first commercial launch scheduled for 2028. Call sign S00851. Status: pending review.

What the application does not contain is a mass. Not in the technical annex, whose orbital debris section states that the satellites “are still being designed” and promises a modification once the design is final. Not in the Schedule S. No kilogram figure, no spacecraft power figure, no GPU count across any of the three documents on file.

That absence is not Cowboy’s doing. Schedule S has no field for spacecraft mass and none for generated power; “power” in Part 25 means EIRP, the energy leaving an antenna. Run the form and you get frequency bands, beam polarizations, G/T, antenna gain contours, orbital planes. The only dimension it asks about is whether the satellite’s smallest side reaches ten centimeters, and it asks so that radar can find the thing. Cowboy’s answer runs to one sentence: “All Stampede satellites will be larger than 10 cm in their smallest dimension and are therefore presumed to be trackable.”

Cowboy said as much itself, in the first paragraph of its technical annex. The document exists, it explains, to carry information “that cannot be fully captured by the Commission’s Schedule S form or software.”

It runs deeper than a missing field. Cowboy’s Schedule S does not describe 20,000 satellites. It describes four. Section S3 reports four simultaneously operating satellites, four deployed during the license term, across four orbital planes. The legal narrative explains why: the company “has not been able to include all the planned satellites in each expected plane and has therefore included only a single representative satellite for each plane.” That is the basis of the § 25.114(a)(1) waiver request; the Commission’s binding technical form now carries four spacecraft standing in for twenty thousand.

The annex knows the real number. Its interference analysis models 2,885 simultaneously visible satellites at 700 kilometers and 3,333 at 1,000.

Bhatt has supplied the mass figures separately. Each unit, the company has said publicly, runs 20,000 to 25,000 kilograms and delivers one megawatt across roughly 800 GPUs. Multiply that by the 20,000 satellites the narrative does request: 400,000 to 500,000 tonnes of hardware.

Jonathan McDowell’s annual space report puts total payload mass launched to orbit in 2025 at 3,193.4 tonnes, up from 2,541.1 the year before, in a year Falcon 9 flew 165 times. Cowboy’s own numbers describe between 125 and 157 times everything the planet lifted last year. Spread over a decade, they would consume the entire global launch industry twelve to sixteen times over, flying nothing else.

SpaceX’s million-satellite application follows the same pattern and lands somewhere worse. That filing is thin on mass too; the American Astronomical Society told the docket in March that the satellites’ size “is not specified in the Application or the Opposition.” The company has nonetheless described publicly a plan to launch one million tonnes of satellites a year, at 100 kilowatts of compute per tonne, adding 100 gigawatts annually. That is 313 times the 2025 world total, every year.

Demand implies a supplier who could plausibly fill the order. At these masses, no such supplier exists or is being built.

Kilograms per kilowatt

SpaceX’s public claim implies roughly 10 kilograms of spacecraft per kilowatt, which is the hundred-kilowatts-per-tonne arithmetic reduced. Slava Turyshev at NASA JPL posted a preprint in April and got 34 to 59. A single megawatt, in his accounting, needs 5,640 square meters of solar array and 2,500 square meters of radiator.

Now run Cowboy’s numbers through the same conversion. Twenty to twenty-five tonnes for one megawatt is 20 to 25 kilograms per kilowatt. That sits below Turyshev’s entire range: Cowboy’s heaviest unit comes in lighter per kilowatt than his most optimistic case, while running two to two and a half times heavier than the figure SpaceX puts in its own marketing.

Whoever is closest to the welding assumed the heavier number. Mass estimates here appear to track how near a company is to building hardware, and every published constellation figure inherits whichever assumption its author picked without saying so.

Turyshev’s solar-array figure is also why item 11(c) bites. Panel dimensions are the closest thing to a power disclosure the debris rules can reach. The Bureau asked for them alongside maximum cross-sectional area, a collision-probability demonstration that must include the solar arrays, and a casualty-risk calculation under § 25.114(d)(14)(vii)(D)(2)(ii). None of those can be produced without knowing what the spacecraft is made of and what it weighs. The Commission wanted the number; lacking a field for it, the Bureau went around its own form to get at it.

A Stampede unit at 20 to 25 tonnes outweighs the roughly 18.5 tonnes a Falcon 9 can put in low Earth orbit while still recovering its booster. No Falcon will ever carry one, because a Stampede unit is not a payload. Cowboy has designed a spacecraft that no existing rocket could deliver even if one were free.

Slightly bigger than Falcon 9

Almost nothing has been published about the vehicle. Bhatt says it will be “slightly more powerful than the SpaceX’s workhorse, Falcon 9, though still smaller than Starship,” with a booster designed for eventual reuse and a first flight targeted for 2028. No engine name, no thrust figure, no launch site, no test article. Cowboy’s own head of launch and propulsion, Warren Lamont, described something else again to GeekWire on September 5: “the last big clean-sheet launch vehicle in my lifetime, so it’s going to be a very big heavy-lift vehicle.” Slightly more powerful than Falcon 9 and very big heavy-lift are not the same rocket.

Cowboy has never flown an operational spacecraft, though it holds three experimental licenses and promises a power-beaming and optical-link demonstration this year, with a first GPU satellite in the first quarter of 2027. What the filing does disclose about readiness is mostly balance sheet and partnership: $340 million raised to date, money from the Department of War’s Operational Energy Capability Improvement Fund for space-to-Earth power beaming, and an arrangement with NVIDIA to fly its Space-1 Vera Rubin Module. None of that is hardware in orbit.

Look past the schedule to the architecture, which is the interesting part. Cowboy’s website puts it plainly: “the rocket’s upper stage is the satellite itself.” No fairing. No separation. No bus. The launch vehicle stays in orbit and becomes the product, which is elegant engineering and also means every megawatt of compute consumes an entire upper stage and a dedicated launch. The satellite is not cargo but the vehicle itself, spent.

The rule that moved, and hasn’t landed

Cowboy asked to be excused from four sets of rules. The argument running through all of them is that Stampede moves its data by optical laser link and therefore warehouses no scarce spectrum. A waiver, the narrative says, “would give Cowboy Space the ability to flexibly scale its operations over time to respond to market demand and technological developments.” Beyond the 2028 first launch, the application discloses no dated schedule for reaching 20,000.

On the bond, Cowboy quoted the Commission back to itself, citing an October 2025 rulemaking notice that had proposed dropping the requirement for systems granted spectrum access outside a processing round.

The Commission then adopted that proposal. FCC 26-47, adopted July 22 in SB Docket 25-306, replaces Part 25’s licensing framework with a new Part 100 and splits NGSO milestones by whether an applicant opts into a processing round. Those that do keep the six-and-nine schedule and post a $10 million surety bond within 30 days of grant, declining to zero at 90 percent deployment. Those that do not get the ITU timetable written into § 100.146: one satellite within seven years, 10 percent in nine, half in twelve, the remainder in fourteen. For them the Commission requires no bond at all.

The front of that schedule is not soft. The seven-year mark is the ITU’s bringing-into-use deadline, met by notifying the Commission that a single satellite has operated for 90 continuous days, and missing it triggers automatic termination of the license. That is the same provision, carried into § 100.302, that the Bureau waived for Amazon Leo in June. What the lighter path relieves is the back end: the pace of the build-out, and the money posted against it.

That rule is not yet in force. It takes effect 60 days after Federal Register publication, and as of September 11 it had not been published. Every other Commission item adopted around it has appeared in the Register, 26-48 through 26-51 and 26-56 among them. Several schedules also await Paperwork Reduction Act clearance, and the Space Bureau has promised separate guidance on applications pending at the transition.

Cowboy’s classification is unsettled too, and the Bureau has said so. A footnote to the August letter observes that the frequencies Cowboy wants “are included in three separate processing rounds, and Cowboy Space would be required to protect operations authorized in those processing rounds,” citing cut-offs from 2016, 2020 and June of this year. The spectrum-light theory is the load-bearing element of every waiver in the file, and the Bureau has flagged it rather than accepted it.

What is not in doubt is the design of the new regime. Deployment discipline is now indexed to spectrum appetite. A 400,000-tonne optical constellation reaches the lighter schedule precisely because it does not want much spectrum, and can then take fourteen years, bondless, to discover whether the mass was ever liftable.

What the Bureau charged Amazon

Set that against the most recent milestone ruling the Bureau actually issued.

In DA 26-553, released June 5, it waived the automatic-termination provision at § 25.161(a)(2) as applied to Amazon Leo’s 50 percent milestone, waived the prerequisite that the interim milestone be met before final-milestone compliance, and waived a compatibility showing for temporarily reassigned satellites. The extension request was dismissed as moot. July 30 stood, and missing it still counts as a violation. The surety bond is forfeited. Satellites undeployed at the deadline lose processing-round priority for 20 months, to March 30, 2028, reducible to 15 if Amazon certifies that half the fleet is built and launches are secured. Amazon had 396 production satellites in orbit in early July against the 1,616 the milestone requires.

That is the Bureau’s revealed exchange rate for a licensee with hardware already flying: the obligation stays, the bond goes, and the delay is charged against spectrum priority.

Cowboy sits in a different posture, and the difference cuts the other way. Amazon was a licensee in default, and the Commission had a bond and a priority position to take away. Cowboy is a pending applicant with nothing built and nothing to forfeit, which is exactly why the conditions written at grant carry the whole weight.

The Commission is not hearing this in silence. Three filings landed in July. The Center for Space Environmentalism opposed on the 6th and Earthjustice petitioned for a programmatic environmental impact statement on the 8th. Secure World Foundation filed on the 10th, noting that Cowboy “has not previously operated a satellite constellation at any scale,” and asked the Commission to deny or defer every waiver requested, naming the Schedule S completeness relief specifically. Then on September 9, NASA objected to a separate 100,000-satellite orbital data center application, arguing that its own missions operate in and transit those altitudes. It had raised the same objection against Blue Origin’s Project Sunrise in May.

Ten flights between them

Scarcity is real, though not unanimous. Arianespace CEO David Cavaillolès called it “absolutely a situation of scarcity” at SATELLITE 2026 in March. SpaceX’s vice president of commercial sales, Stephanie Bednarek, called her company’s manifest packed from 2026 through 2028 on the same panel, while pointing elsewhere for the binding constraint: “rockets don’t seem to be the limiting factor. It’s more about payload processing space.” Firefly CEO Jason Kim told Benzinga in August that demand for government and commercial constellations “outstrips the rocket capacity available, both in the U.S. and globally.”

Consider the queue. Falcon 9 has flown 103 times in 2026 through September 6. The four other Western medium- and heavy-class vehicles have flown ten times between them: Atlas V four, Ariane 6 four, Vulcan and New Glenn one each. Vulcan has been grounded since February, when a solid-rocket-booster nozzle burned through 20 seconds after liftoff on USSF-87, a mission that reached orbit anyway. New Glenn lost NG-3 to a second-stage anomaly in April, then lost its pad to an explosion during a prelaunch test in May; the rebuild is estimated at more than a year. Neutron, Terran R and Nova have flown zero times, and Neutron slipped again this month. Firefly and Northrop’s Eclipse is not expected before 2027. SpaceX has meanwhile stopped taking rideshare bookings past late 2028, as this publication reported in July.

None of that adds up to 400,000 tonnes. It does not add up to four thousand.

The Kent lease

There is a fair reading in which none of this matters, and it deserves stating at full strength: filings are free. Across the sector they now total roughly 1.26 million spacecraft — SpaceX’s million, Orbital Compute’s 100,000, Starcloud’s 88,000, Blue Origin’s 51,600, Cowboy’s 20,000 — against a handful of demonstration slots. Nothing has been authorized. A reasonable person concludes these are option tickets written by companies with nothing to lose by asking. This is a year in which SES is guiding to about €700 million of capex and cancelling two GEO satellites it inherited from Intelsat.

Then Cowboy raised $275 million of a $340 million total. Starcloud raised a $250 million extension at a $2.3 billion valuation in August. On September 4, Cowboy announced it had leased a 291,035-square-foot former Costco distribution center in Kent, Washington, to build the hardware. Rocket programs and factories are not option tickets. Somebody has decided the arithmetic can be closed and is spending accordingly, which leaves the question of at what mass per kilowatt, and who is right about it.

The number, asked for

Six weeks ago this column argued that launch inventory had become an input to spectrum rights. The August letter suggests the Bureau has reached the same place from the other side, and found its own form waiting there.

Eleven questions went to San Carlos. Most are requests for the system as it would actually fly, rather than the four representative spacecraft the software could hold: the maximum number of satellites operating in Ka-band at any one time, the number of satellites per plane delivered as a spreadsheet, whether Cowboy holds an optical intersatellite link contract with any operator and which one. Item 11(c) reaches for the mass through the only door available to it.

That door should not be the only one. Casualty risk and cross-sectional area are proxies, extracted late and one applicant at a time, from a debris rule written for another purpose. The Commission is holding orbital data center applications for more than a million spacecraft, and it has just built them a lighter deployment path on the theory that a system which does not hoard spectrum does not need watching. Mass is the thing that theory does not measure.

So the fix is a field added, not a waiver denied. What does the Commission believe it is licensing when it grants 20,000 satellites whose combined mass it has never been told, and how would anyone at the agency know the difference between a constellation and a decade of global launch?

The answer to that starts with the kilograms per kilowatt. The Bureau has asked for something close to it, and the response was due today.

---

About the Author

A storyteller at heart, Nick David covers space policy, satellite markets, defense, and the technologies reshaping how humanity operates beyond Earth. With a background in creative direction, brand strategy, and editorial storytelling, he brings a modern lens to complex subjects and a relentless curiosity about what comes next.