Starcloud

Scroll sideways for more →

Starcloud, Inc. is an American space infrastructure startup that builds space-based data centers: satellites carrying data-center-class GPUs that are powered by solar arrays and cooled by radiating waste heat into space. The Redmond, Washington company launched Starcloud-1 in November 2025, putting the first NVIDIA H100 GPU into orbit, and in December 2025 it trained a small language model on that satellite, which the company describes as the first AI model trained in space.^[1][4][5] In March 2026 it closed a $170 million Series A led by Benchmark and EQT Ventures at a $1.1 billion post-money valuation; the company and GeekWire called it the fastest company in Y Combinator history to reach unicorn status, 17 months after its demo day, while TechCrunch described it as "one of the fastest."^[1][2][3]

History

Starcloud was founded in January 2024 under the name Lumen Orbit by Philip Johnston, Ezra Feilden, and Adi Oltean.^[2][10] Johnston, the CEO, previously worked at McKinsey & Co. on satellite projects for national space agencies; Feilden, the CTO, spent a decade on satellite design at Airbus Defence and Space (SSTL) and Oxford Space Systems, with a PhD in materials engineering from Imperial College London; Oltean, the chief engineer, was a principal software engineer on SpaceX's Starlink team and before that spent two decades at Microsoft, where he worked on large production GPU clusters.^[16][18]

The company went through Y Combinator's Summer 2024 batch and published a white paper, "Why we should train AI in space," in September 2024. The paper argues that gigawatt-scale AI clusters will outgrow terrestrial power grids and that a dawn-dusk sun-synchronous orbit, where a spacecraft rides the day-night line and its solar panels face the sun almost continuously, can supply near-constant power without batteries. Its cost model projects that a 40 MW compute cluster operated for ten years would cost roughly $8.2 million in space against $167 million on land, assuming Starship-class launch prices, and sketches an eventual 5 GW orbital data center with a solar array roughly 4 km by 4 km.^[15]

In December 2024 the startup announced an $11 million seed round from NFX, Y Combinator, FUSE, Soma Capital, and scout funds associated with Andreessen Horowitz and Sequoia. In February 2025 it added $10 million more, bringing the seed total to $21 million, among the largest seed rounds for a YC company at the time, and simultaneously changed its name from Lumen Orbit to Starcloud because Lumen Technologies holds the rights to the "Lumen" name for data centers.^[10]

Johnston has said the company was "fairly roundly pilloried" in its early days: "If you go back to some of the comments on X when we announced, people said it was impossible and we couldn't do it."^[2]

Starcloud-1

Starcloud-1, a 60 kg demonstration satellite about the size of a small refrigerator, launched on November 2, 2025 on SpaceX's Bandwagon-4 Falcon 9 rideshare mission from Cape Canaveral, one of 18 payloads deployed to a mid-inclination low Earth orbit.^[11][12] The satellite was built by Astro Digital on its Corvus-Micro bus, with Starcloud integrating the compute payload, and operates at about 325 km altitude with a planned lifetime of roughly 11 months followed by a controlled deorbit.^[11] It carries the first NVIDIA H100 GPU flown in space, which NVIDIA said represented about 100 times more GPU compute than had previously operated in orbit.^[4] A second GPU on board, an NVIDIA A6000, failed during launch, according to TechCrunch.^[1]

Within weeks of commissioning the satellite produced a series of firsts that Starcloud, NVIDIA, and Google publicized in December 2025:^[5]

• It trained NanoGPT, the minimal GPT implementation by Andrej Karpathy, on the complete works of Shakespeare, which the company describes as the first language model trained in space. Karpathy wrote: "NanoGPT - the first LLM to train and inference in space. It begins."^[5][13]
• It ran Gemma, Google's open-weights model family derived from Gemini, in orbit; Starcloud and several outlets describe this as running "a version of Gemini" in space. Google DeepMind CEO Demis Hassabis congratulated the team, and Google DeepMind product director Tris Warkentin said that "seeing Gemma run in the harsh environment of space is a testament to the flexibility and robustness of open models."^[5][13]
• It ran inference on synthetic-aperture radar imagery from Earth-observation company Capella Space, an early customer workload aimed at tasks such as spotting wildfires or lifeboats at sea without first downlinking raw data.^[1][5]

Johnston told TechCrunch that the H100 "is probably not the best chip for space, to be honest, but the reason we did it is we wanted to prove that we could run state of the art terrestrial chips in space."^[1]

Satellite roadmap

Scroll sideways for more →

Starcloud-2 is the company's first commercial mission, designed to run customer cloud workloads with persistent storage and 24/7 access in sun-synchronous orbit.^[14] It will carry multiple GPUs, including an NVIDIA Blackwell B200, an AWS server blade, a module running a cloud platform from AI infrastructure provider Crusoe, and, according to TechCrunch, a bitcoin-mining computer.^[1][3][5] Johnston says it will fly the largest commercial deployable radiator sent to space and generate about 8 kilowatts, roughly 100 times Starcloud-1's power.^[3][7] He also told SpaceNews the satellite will be the company's first to earn more revenue than it costs to build and launch, with hosted payloads from defense and Earth-observation customers covering development costs.^[3] The launch date has moved: CNBC reported an October 2026 target in December 2025, March 2026 coverage said "later this year," and by late May 2026 SpaceNews reported it was slated for a Falcon 9 launch in January 2027.^[1][3][5][7] At least two more satellites of the same class are planned.^[1]

Starcloud-3 is a three-ton, 200-kilowatt-class spacecraft sized for SpaceX's Starship and its "PEZ dispenser" deployment system, originally designed for Starlink satellites. Johnston estimates a single Starship could deploy about 50 of them, roughly 10 MW of computing per launch, and expects Starship to take commercial payloads in the mid-to-late 2028 timeframe.^[1][3] Series A funds are going toward a 3,000 square meter production facility in Woodinville, Washington; the first chassis had been welded by March 2026.^[3] Johnston expects Starcloud-3 to be the first orbital data center that is cost-competitive with terrestrial ones, on the order of $0.05 per kWh if commercial launch costs fall to about $500 per kilogram. "We're not going to be competitive on energy costs until Starship is flying frequently," he told TechCrunch, adding, "If it ends up being delayed, we'll just carry on launching the smaller versions on Falcon 9."^[1]

Starcloud-4 refers to the long-term concept of Starship-deployed spacecraft with solar and cooling arrays approximately 4 km on a side, supporting a 5-gigawatt data center.^[4][6]

Constellation plans and the FCC filing

On March 13, 2026 the U.S. Federal Communications Commission accepted for filing a Starcloud application to operate as many as 88,000 satellites as orbital data centers, in narrow orbital shells between 600 and 850 km altitude, in dusk-dawn sun-synchronous orbits for near-continuous solar power.^[6] The constellation would rely on optical intersatellite links with broadband systems such as Starlink, Amazon's Project Kuiper, and Blue Origin's TeraWave for most communications, with Ka-band spectrum requested only for telemetry, tracking, and control. The filing says the satellites are "designed for full demisability" (burning up completely on reentry) and commits to brightness-mitigation measures for astronomy.^[6] The application is far larger than any operating constellation (Starlink had about 10,000 satellites at the time) but smaller than SpaceX's own January 2026 filing for up to one million orbital data center satellites.^[6]

On May 26, 2026 Starcloud announced a contract with SpaceX's Starlink for more than 50 Starlink Mini Laser terminals across more than 25 satellites, with the first hardware expected on orbit within a year. Each Starcloud satellite will carry two of the terminals, which provide up to 25 Gbps of intersatellite connectivity at ranges up to 4,000 km, connecting Starcloud's spacecraft into Starlink's mesh network instead of routing everything through ground stations.^[7][8] "That's what turns individual satellites into a functioning distributed data center," Johnston said.^[8] Johnston has argued that 25 Gbps, modest by terrestrial standards, gives a 200-kilowatt orbital node more external connectivity per watt than an average ground data center.^[7]

Business model and partners

Johnston describes two business models. Near-term, Starcloud sells processing to other spacecraft in orbit: Earth-observation satellites, and synthetic-aperture radar in particular, generate far more raw data (about 10 GB per second, per Johnston) than downlinks can comfortably carry, so running inference next to the sensor cuts response times for applications such as wildfire detection.^[1][4] Longer-term, as launch costs fall, Starcloud intends to sell energy and infrastructure as a service, letting customers run their own computing in orbit rather than operating its own AI workloads, analogous to leasing terrestrial data center capacity.^[3][7]

The company is a member of the NVIDIA Inception startup program and a graduate of the Google for Startups Cloud AI Accelerator; press reports list NVIDIA, AWS, and Google as partners and Crusoe as an early Starcloud-2 customer.^[3][4][5] Starcloud claims orbital data centers will have roughly 10 times lower energy costs than terrestrial ones and 10 times lower lifetime carbon-dioxide emissions, with launch as the only environmental cost; these are company projections.^[4][5] Johnston has predicted that "in 10 years, nearly all new data centers will be being built in outer space."^[4]

Funding

Scroll sideways for more →

The Series A valued Starcloud at $1.1 billion post-money and brought total funding to about $200 million.^[1][2][3] The first tranche was led by Benchmark with participation from EQT, whose parent investment group operates more than 70 terrestrial data centers, and an extension was co-led by both firms; Benchmark general partner Chetan Puttagunta joined the board.^[2] Angel investors included retired U.S. Air Force General Stephen Wilson, former Boeing CEO Dennis Muilenburg, and former Starbucks CEO Kevin Johnson.^[2][3] In early May 2026, a source confirmed to SpaceNews, following earlier reporting by The Information, that Starcloud was in talks to raise at least $200 million more at roughly double the $1.1 billion valuation; the company declined to comment.^[9]

Competition

Several companies are pursuing orbital compute. SpaceX filed with the FCC in January 2026 for up to one million data center satellites,^[6] a program since organized as Starmind; Johnston argues the two companies can coexist because SpaceX will mainly serve internal xAI (Grok) and Tesla workloads while Starcloud positions itself as a neutral infrastructure and energy provider.^[1][3][9] Google announced Project Suncatcher, a "moonshot" to fly TPU-equipped solar satellites, in November 2025; Aetherflux, founded by Robinhood co-founder Baiju Bhatt, targets an orbital data center satellite in early 2027; Lonestar Data Holdings works on lunar data storage; and Axiom Space, Kepler Communications, Sophia Space, and Aethero (which flew NVIDIA's first space-based Jetson GPU in 2025) are developing related systems.^[1][2][5] Johnston claims a head start: "We're two years ahead in terms of having any kind of data and telemetry from how these chips perform on orbit."^[2]

Reception and criticism

Starcloud's plans depend on unproven technology and heavy capital spending, and on a rocket that is not yet in commercial service; TechCrunch noted that every large space data center project is cost-prohibitive until a new generation of rockets flies at high cadence, which may not happen until the 2030s.^[1] Large training workloads also need hundreds or thousands of GPUs working in tandem, which in orbit would require either very large spacecraft or reliable laser links between formation-flying satellites.^[1] Morgan Stanley analysts have flagged radiation, the difficulty of in-orbit maintenance, debris hazards, and regulatory questions around data governance and space traffic.^[5] Johnston himself names the two hardest engineering problems as building a large, low-cost deployable radiator and making high-performance chips survive a higher-radiation environment.^[9] Microsoft president Brad Smith, asked about the approach, said the company might eventually pursue satellites but was not betting on it: "We're keeping our feet on the ground."^[2]

Environmental groups have also targeted the category. In July 2026 a coalition represented by Earthjustice petitioned the FCC to require environmental review before licensing orbital data center constellations, citing debris risk, reentry pollution, and night-sky brightness; the petition responded to the wave of filings that includes SpaceX's million-satellite proposal and, by extension, constellations like Starcloud's.^[17]

See also

• Space-based data centers
• SpaceX Starmind
• NVIDIA H100
• Crusoe
• AI data center
• Y Combinator
• NVIDIA Vera Rubin

References

1. ^Tim Fernholz, "Starcloud raises $170 million Series A to build data centers in space," TechCrunch, March 30, 2026. techcrunch.com/...-ato-build-data-centers-in-space
2. ^Lisa Stiffler, "Orbital AI: Seattle-area startup Starcloud hits $1.1B valuation to build space-based data centers," GeekWire, March 30, 2026. geekwire.com/...-to-build-space-based-data-centers
3. ^Jason Rainbow, "Starcloud achieves unicorn status with $170 million raise for orbital data centers," SpaceNews, March 30, 2026. spacenews.com/...on-raise-for-orbital-data-centers
4. ^Angie Lee, "How Starcloud Is Bringing Data Centers to Outer Space," NVIDIA Blog, October 15, 2025. blogs.nvidia.com/...starcloud
5. ^Pia Singh, "'Greetings, earthlings': Nvidia-backed Starcloud trains first AI model in space as orbital data center race heats up," CNBC, December 10, 2025. cnbc.com/...ai-model-in-space-orbital-data-centers
6. ^Jeff Foust, "Starcloud files plans for 88,000-satellite constellation," SpaceNews, March 15, 2026. spacenews.com/...for-88000-satellite-constellation
7. ^Jason Rainbow, "Starcloud orders Starlink lasers for orbital data center network," SpaceNews, May 26, 2026. spacenews.com/...s-for-orbital-data-center-network
8. ^"Starcloud to Integrate SpaceX's Starlink Mini Lasers Into Its Orbital Data Center Constellation," Business Wire press release (via Morningstar syndication), May 26, 2026. morningstar.com/...bital-data-center-constellation
9. ^Jason Rainbow, "Starcloud seeks more orbital data center funding shortly after unicorn status," SpaceNews, May 1, 2026. spacenews.com/...ding-shortly-after-unicorn-status
10. ^Alan Boyle, "Lumen Orbit changes its name to Starcloud and raises $10M for space data centers," GeekWire, February 26, 2025. geekwire.com/...t-starcloud-10m-space-data-centers
11. ^"Starcloud 1 (Lumen 1)," Gunter's Space Page (Gunter Krebs), accessed August 2026. space.skyrocket.de/...starcloud-1
12. ^Jeff Foust, "SpaceX launches fourth Bandwagon rideshare mission," SpaceNews, November 2, 2025. spacenews.com/...ourth-bandwagon-rideshare-mission
13. ^"Starcloud-1," Starcloud, Inc. (official site), accessed August 2026. starcloud.com/starcloud-1
14. ^"Starcloud-2 | In-Space GPU Cluster & Cloud Computing Satellite," Starcloud, Inc. (official site), accessed August 2026. starcloud.com/starcloud-2
15. ^Ezra Feilden, Adi Oltean, Philip Johnston, "Why we should train AI in space," Lumen Orbit (now Starcloud) White Paper v1.03, September 2024. starcloudinc.github.io/wp.pdf
16. ^"Starcloud: Data Centers in Space," Y Combinator company directory, accessed August 2026. ycombinator.com/...starcloud
17. ^Chelsea Gohd, "'Reckless' space-based data centers lack environmental review, drawing criticism," Space.com, July 9, 2026. space.com/...nvironmental-review-drawing-criticism
18. ^"Data Centers in Space," Starcloud, Inc. homepage and team pages, accessed August 2026. starcloud.com