FundingRounds, leads & terms
Orbital data center startup Starcloud reaches $2.3B valuation after $250M raise led by Manhattan West
Orbital data center startup Starcloud more than doubled its valuation to $2.3 billion in five months with a $250 million extension. Its only hardware in orbit so far is a single Nvidia H100, and that satellite's radiator could not keep the chip at full power.
From the The Peel (Turner Novak) podcast episode of November 20, 2026 and cited sources: Starcloud's valuation rose fast. [3] [7] [8] [9]
SOURCES[3] Business Wire: Starcloud Raises $250M at $2.3B businesswire.com[7] HPCwire: $170M Series A release hpcwire.com[8] TechCrunch: Starcloud raises $170M Series A techcrunch.com
AI-assisted, not human-edited. Methodology Corrections Who owns this site
- Original source
- The Peel (Turner Novak) podcast episode, independent perspective
- Brief by
- AT Pulse research desk AI-assisted; publisher and contact on the desk page
- When
- November 20, 2026 podcast episode published
- Covered
- Nov 21 edition
- Evidence
- 13 sources across 10 sites; no transcript
- 1The Split: How Starcloud Went $0 to $2.3B in 31 Monthsthespl.it
- 2IEEE Spectrum: Orbital data center hypespectrum.ieee.org
- 3Business Wire: Starcloud Raises $250M at $2.3Bbusinesswire.com
- 4IEEE Spectrum: Nvidia H100 in spacespectrum.ieee.org
- 5NVIDIA blog: Starcloudblogs.nvidia.com
- 6GeekWire: Starcloud training AI in spacegeekwire.com
Reach: not measured yet
Starcloud's valuation rose fast. In March 2026 it raised a $170M Series A at $1.1B, led by Benchmark and EQT Ventures. In August a $250M extension at $2.3B followed, led by Manhattan West, with Nvidia and Cisco Investments joining as new investors. Both valuations were announced by the company, and no regulatory filing confirming them has surfaced. [3] [7] [8] [9] [10]
The technical proof so far is small. Starcloud-1 is a 60 kg satellite carrying one H100. In orbit it trained NanoGPT on a Shakespeare text corpus and ran Gemma inference. That shows an unmodified data-center GPU can run in space. It does not show useful training throughput, and no benchmarks, uptime figures or radiation error rates have been published. [4] [5] [6] [2]
Heat, launch capacity and mass will decide whether this works. In a vacuum, heat can only leave by radiation, so large radiators are needed. Independent analyses estimate that radiators and solar arrays outweigh the servers by about four to one, and they find ground data centers still cheaper. Johnston says securing launch capacity has become one of the company's biggest costs. [2] [9] [12] [13]
Nvidia, Cisco and EQT are backing orbital compute, though Starcloud has disclosed no revenue or commercial service. Implication: the valuation prices in launch access and future scale.
Read the full assessment
The two Starcloud-2 satellites planned for 2027 will be the first real test.
Executive brief
Starcloud has had one GPU in orbit, and IEEE Spectrum reports that its radiator was too weak to run that chip at full power (IEEE Spectrum). Even so, investors have priced the company at $2.3 billion. In August 2026 the Redmond startup announced a $250M Series A extension (Business Wire), five months after a $170M round at $1.1B. A November 2026 interview with CEO Philip Johnston on The Split presents this as 31 months from zero to $2.3B (The Split). Launch capacity, heat removal and cost per kilogram decide whether this works.
What changed and event timeline
First H100 reaches orbit
Starcloud-1 is a 60 kg satellite carrying one Nvidia H100. It launched on SpaceX's Bandwagon 4 rideshare into roughly 350 km orbit, with a planned three-year mission ().
Training run in orbit
Starcloud said it trained Karpathy's NanoGPT on Shakespeare and ran inference with Google's Gemma on the orbiting H100 (;).
Unicorn round
The company announced a $170M Series A at a $1.1B valuation, led by Benchmark and EQT Ventures. Total raised reached $200M (;).
$250M extension at $2.3B
Nvidia and Cisco Investments joined as new investors (;).
Founder interview
The Split episode description says Starcloud built a "$100M satellite" for $2.5M and that 40 VCs passed on its Series A. Both claims come from the company and have not been verified elsewhere ().
Capabilities and access
- Starcloud-1 (in orbit): one Nvidia H100 with 80 GB memory. Workloads tested include Earth-observation and SAR image analysis and Gemma inference (IEEE Spectrum).
- Starcloud-2 (planned 2027): two satellites with about 8 kW of compute each, launching on rideshare flights.
- Starcloud-3: designed to launch on Starship. The company has asked the FCC for permission covering up to 88,000 spacecraft.
Read the full section
- Starcloud-1 (in orbit): one Nvidia H100 with 80 GB memory. Workloads tested include Earth-observation and SAR image analysis and Gemma inference (IEEE Spectrum).
- Starcloud-2 (planned 2027): two satellites with about 8 kW of compute each, launching on rideshare flights. Planned uses include orbital inference and US government work (TechCrunch). Earlier coverage described about 7 kW, with Blackwell and multiple H100s (IEEE Spectrum).
- Starcloud-3: designed to launch on Starship. The company has asked the FCC for permission covering up to 88,000 spacecraft.
- No public commercial service or API has been announced.
Technical analysis for researchers and developers
- What the demo shows: NanoGPT on a Shakespeare corpus is a small toy workload. It proves an unmodified data-center GPU can run in orbit.
- Cooling: in a vacuum, heat can only leave by radiation.
- Power: orbital solar panels produce up to 8x the output of the same panels on the ground.
Read the full section
- What the demo shows: NanoGPT on a Shakespeare corpus is a small toy workload. It proves an unmodified data-center GPU can run in orbit. It does not show useful training throughput. No benchmark figures, uptime numbers or radiation error rates have been published (NVIDIA blog).
- Cooling: in a vacuum, heat can only leave by radiation. One 700 W H100 needs about 1.4 m² of radiator at 60°C. A 100 MW facility would need about 2,500 such radiators (IEEE Spectrum).
- Power: orbital solar panels produce up to 8x the output of the same panels on the ground. This is the core of Starcloud's case (IEEE Spectrum).
- Hardware roadmap: Nvidia and Starcloud are working on "Vera Rubin Space-1", a GPU designed for space, targeting late 2028 (TechCrunch).
Claims and evidence
- August round: $250M at a $2.3B post-money valuation, led by Manhattan West. The company announced it (Business Wire) and TechCrunch reported the same figures.
- March round: $170M at $1.1B, announced by the company and reported by TechCrunch. The valuation is self-reported (Business Wire).
- Interview claims: the "$2.5M satellite" and "40 VCs passed" lines come from the founder only.
Read the full section
- August round: $250M at a $2.3B post-money valuation, led by Manhattan West. The company announced it (Business Wire) and TechCrunch reported the same figures. The TechCrunch URL slug says "$200 million" while its headline says $250M (TechCrunch). TechCrunch puts Nvidia's stake at $25M without naming a source. No regulatory filing confirming these numbers was found.
- March round: $170M at $1.1B, announced by the company and reported by TechCrunch. The valuation is self-reported (Business Wire).
- Interview claims: the "$2.5M satellite" and "40 VCs passed" lines come from the founder only.
- Revenue: none disclosed.
Context and prior work
- Coverage says it built and launched its first satellite in 21 months on about $3M of pre-seed money (TechCrunch).
- Competing visions are much larger. SpaceX has floated a million data-center satellites, and Blue Origin has proposed 51,600 (The Next Web).
- Google's Project Suncatcher is cited as finding that orbital compute pays off only below about $200/kg launch cost (byteiota).
Read the full section
- Starcloud went through Y Combinator. Coverage says it built and launched its first satellite in 21 months on about $3M of pre-seed money (TechCrunch).
- Competing visions are much larger. SpaceX has floated a million data-center satellites, and Blue Origin has proposed 51,600 (The Next Web).
- Google's Project Suncatcher is cited as finding that orbital compute pays off only below about $200/kg launch cost (byteiota).
Limitations, safety and contested findings
- Thermal limits: IEEE Spectrum reports Starcloud-1's radiator could not support the H100 at full power.
- Launch supply: Johnston says securing launch capacity is now one of the biggest costs.
- Cost and weight: independent analyses find ground data centers are still cheaper.
Read the full section
- Thermal limits: IEEE Spectrum reports Starcloud-1's radiator could not support the H100 at full power. Its editors argue the economics "won't make sense for several years, if ever" (IEEE Spectrum).
- Launch supply: Johnston says securing launch capacity is now one of the biggest costs. Falcon 9 is reportedly scheduled to end in 2028, and Starship is still unproven (TechCrunch).
- Cost and weight: independent analyses find ground data centers are still cheaper. They estimate radiators and solar arrays outweigh the servers by about 4x (byteiota; Forethought).
Business and practitioner implications
- Investors: the valuation reflects strategic backing (Nvidia, Cisco, and EQT, which owns more than 70 data centers) and a claim on scarce launch slots.
- Near-term buyers: the realistic early customers are processing data that is already in orbit, such as Earth observation, SAR and defense.
- Developers: the 2027 Starcloud-2 flights are the first test of whether a multi-kilowatt orbital inference service works.
Read the full section
- Investors: the valuation reflects strategic backing (Nvidia, Cisco, and EQT, which owns more than 70 data centers) and a claim on scarce launch slots. Demonstrated compute economics do not support it yet.
- Near-term buyers: the realistic early customers are processing data that is already in orbit, such as Earth observation, SAR and defense. Those workloads avoid downlink bottlenecks.
- Developers: the 2027 Starcloud-2 flights are the first test of whether a multi-kilowatt orbital inference service works. Starship's progress and launch prices are the variables to track.