Starcloud raises $250M at $2.3bn because rockets, not GPUs, are the scarce input

Starcloud, which builds satellites that run AI inference in orbit, told TechCrunch it has added a $250 million extension to the $170 million Series A it raised in March, at a $2.3 billion valuation ^[1][2]. The stated use of funds is a larger manufacturing facility and its biggest spacecraft, Starcloud-3, which is meant to fly on SpaceX's Starship ^[3] - but the more revealing reason is that CEO Philip Johnston is accumulating cash to guarantee he can get to orbit at all ^[4].

The constraint is scheduling, not silicon. Johnston told TechCrunch that "one of the biggest costs is now on securing your launch capacity" and that launch is constrained because SpaceX's Falcon 9 program is scheduled to end in 2028 ^[5]. Starcloud has asked the FCC for permission to operate 88,000 spacecraft ^[6], a number that only means anything if there is a ride. Johnston's framing is blunt: "We can see what's coming - we're going to need to book an enormous amount of launch" ^[7]. The alternatives are thin. Blue Origin's New Glenn and ULA's Vulcan are not flying regularly, and Rocket Lab's Neutron is not yet on the pad ^[8]. Launch economics are hard enough in this sector that one orbital data center startup decided to build its own rockets ^[9].

That leaves Starship, which is larger and cheaper per kilogram in theory but unproven in practice ^[8]. This week Elon Musk said SpaceX will delay an attempt to catch a returning Starship by a few months and will try to re-fly a vehicle for the first time at the end of the year or in early 2027 ^[10]. Johnston says he remains confident SpaceX can demonstrate rapid reuse ^[11], while conceding the downside: "Obviously if we can't book any SpaceX launch capacity in 2029, that will be challenging for us" ^[12].

The near-term plan is modest by comparison. Two 8 kW Starcloud-2 satellites are slated for rideshare flights in 2027, doing inference work for customers including US government agencies, and the company is weighing a dedicated Falcon 9 purchase plus contracts with other providers ^[13][14].

The chip story runs on a similar clock. Nvidia joined the extension, which was led by Manhattan West Ventures and included Cisco, Benchmark, EQT, Soma, NFX, 776, Cedar Capital, Goanna Capital and Standard Capital ^[15][16]; a person familiar with the deal told TechCrunch that Nvidia put in $25 million, or roughly a tenth of the extension ^[17][18]. Johnston attributes that to data from Starcloud One, and says Starcloud is the only company known to be operating an Nvidia H100 data center GPU in orbit and the first to train a model on one, with those learnings feeding Nvidia's first purpose-built space GPU, the Vera Rubin Space-1 ^[19][20][21]. That chip has not been built, and Starcloud hopes to fly it in late 2028 ^[22] - the same year Falcon 9 is scheduled to stop flying ^[5][23]. Open engineering questions include chip operating temperature versus radiator size, radiation shielding placement, and surviving launch loads ^[24].

For scale: 25 employees, a 100,000 square foot site in Woodinville, Washington, and $420 million raised across the round, or about $16.8 million per current employee ^[25][26][27].

Watch whether the capital converts into a signed Starship contract or a dedicated Falcon 9 booking, and watch the Starship re-flight attempt. Everything downstream, including the Vera Rubin Space-1 flight, is gated on that.