Starcloud

Starcloud is a United States startup developing satellites that perform AI inference in orbit, based in Washington state and backed by Nvidia. It launched a satellite carrying an Nvidia H100 data center GPU in early November 2025 and reported training an AI model in orbit the following month. In August 2026 the company added a $250 million extension to its March 2026 Series A round, valuing it at $2.3 billion, and its stated constraint has shifted from technical feasibility to securing launch capacity (Source: https://techcrunch.com/2026/08/21/starcloud-raises-200-million-for-orbital-data-centers-as-launch-options-dry-up/).

Snapshot

Funding

Operations

Premise and economics

Starcloud's case for orbital compute rests on power and cooling rather than on latency. In orbit a satellite receives continuous solar illumination and radiates waste heat directly to space, which the company argues removes the electricity cost and the water and grid constraints that bind terrestrial facilities. Chief executive Philip Johnston told CNBC that orbital data centers would carry energy costs ten times lower than terrestrial ones; the company's own marketing states electricity costs roughly 90% lower (Source: https://www.cnbc.com/2025/12/10/nvidia-backed-starcloud-trains-first-ai-model-in-space-orbital-data-centers.html; Source: https://www.starcloud.com/). These are the company's own projections rather than measured operating figures.

The long-run design Starcloud has described to Nvidia is a 5-gigawatt orbital data center with solar and cooling panels approximately four kilometers on a side (Source: https://blogs.nvidia.com/blog/starcloud/). The economics of the category are contested: TechCrunch has reported separately on why the economics of orbital AI are difficult, and at least one competitor has concluded that launch costs are severe enough to justify building its own rockets (Source: https://techcrunch.com/2026/02/11/why-the-economics-of-orbital-ai-are-so-brutal/; Source: https://techcrunch.com/2026/05/11/there-arent-enough-rockets-for-space-data-centers-cowboy-space-raised-275-million-to-build-them/).

Hardware and missions

Starcloud-1, launched in early November 2025, carried an Nvidia H100 — a terrestrial data center GPU rather than a radiation-hardened edge processor, which distinguishes it from most other GPUs flown in space, which are designed for edge processing. Johnston has said the H100 has not required a single restart since reaching orbit (Source: https://sequoiacap.com/podcast/greetings-earthlings-philip-johnston-of-starcloud-on-data-centers-in-space). In December 2025 the company reported training an AI model on it, which it described as the first model trained in space (Source: https://www.cnbc.com/2025/12/10/nvidia-backed-starcloud-trains-first-ai-model-in-space-orbital-data-centers.html).

The next generation, Starcloud-2, comprises 8-kilowatt compute satellites. Two are planned for rideshare launches in 2027 and are intended to perform orbital inference for customers including U.S. government agencies. The largest planned spacecraft, Starcloud-3, is intended to fly on SpaceX's Starship (Source: https://techcrunch.com/2026/08/21/starcloud-raises-200-million-for-orbital-data-centers-as-launch-options-dry-up/).

Nvidia is developing a purpose-built space GPU, the Vera Rubin Space-1, and Starcloud is supplying operational data from Starcloud-1 to inform it. Johnston has said Starcloud hopes to fly the chip in late 2028, and that its engineers are tracking three design variables: the relationship between chip operating temperature and the radiator area needed to dissipate heat, the placement of radiation shielding, and the ruggedizing needed to survive launch (Source: https://techcrunch.com/2026/08/21/starcloud-raises-200-million-for-orbital-data-centers-as-launch-options-dry-up/).

Launch access as the binding constraint

The August 2026 raise was framed by the company around launch capacity rather than technology. Johnston told TechCrunch, "We can see what's coming — we're going to need to book an enormous amount of launch," and identified securing launch capacity as now one of the largest costs, citing the scheduled end of SpaceX's Falcon 9 program in 2028. The company is considering buying a dedicated Falcon 9 launch and contracting with other providers. Johnston noted that competing vehicles — Blue Origin's New Glenn, ULA's Vulcan, Rocket Lab's Neutron — are not yet flying regularly or at all, and said, "Obviously if we can't book any SpaceX launch capacity in 2029, that will be challenging for us." He said he remains confident in Starship's rapid reusability; SpaceX chief executive Elon Musk said in the week of August 17, 2026 that the company would delay an attempt to catch a returning Starship by a few months and would attempt a first re-flight at the end of 2026 or early 2027 (Source: https://techcrunch.com/2026/08/21/starcloud-raises-200-million-for-orbital-data-centers-as-launch-options-dry-up/).

Starcloud has asked the Federal Communications Commission (Federal Communications Commission (FCC)) for permission to operate 88,000 spacecraft, a constellation-scale authorization request that puts the company's expansion plans in front of a U.S. regulator (Source: https://techcrunch.com/2026/08/21/starcloud-raises-200-million-for-orbital-data-centers-as-launch-options-dry-up/).

Nvidia relationship

Nvidia's involvement runs through three channels: membership in the Nvidia Inception startup program, direct equity, and the Vera Rubin Space-1 development relationship. Johnston has characterized Nvidia's participation in the August round as a technical validation, saying the investment followed the data returned from Starcloud-1 and that Nvidia "more than any other VC, did way more technical duty on this than anybody else" (Source: https://techcrunch.com/2026/08/21/starcloud-raises-200-million-for-orbital-data-centers-as-launch-options-dry-up/). That is the founder's characterization of his own investor. Dion Harris, Nvidia's senior director of AI infrastructure, said of the November 2025 launch: "From one small data center, we've taken a giant leap toward a future where orbital computing harnesses the infinite power of the sun" (Source: https://www.cnbc.com/2025/12/10/nvidia-backed-starcloud-trains-first-ai-model-in-space-orbital-data-centers.html).

Open questions

• Whether the FCC acts on the 88,000-spacecraft request, and on what terms.
• Whether Starship demonstrates the launch cadence and cost the business case assumes before Falcon 9 retires in 2028.
• Whether the projected 10× energy-cost advantage survives contact with launch, replacement and radiation-degradation costs, which the published figures do not appear to net out.

Relationships

• related: Data Centers in Space: A Pipe Dream, or AI's Next Big Thing? — Micah Maidenberg, Merrill Sherman (WSJ, May 12 2026) — reporting on the orbital data center category
• deploys-in: Energy and Electric Power Sector — the business case is an energy-cost argument
• depends-on: Data Center Siting / AI Power Politics — orbital siting is advanced as an answer to terrestrial siting and power constraints
• regulated-by: Federal Communications Commission (FCC) — spacecraft licensing
• related: Nvidia & TSMC — AI Compute Infrastructure — investor, hardware supplier and co-development partner