Starcloud Raises $170M at $1.1B Valuation to Build AI Data Centers in Orbit

The AI industry has a power problem. Data centers consumed approximately 415 terawatt-hours of electricity globally in 2024 — about 1.5% of all electricity generated on Earth — and the International Energy Agency projects that figure will nearly triple to 1,100 TWh by 2026, rivaling the entire electrical consumption of Japan. In northern Virginia alone, data centers consume one in every five kilowatt-hours produced by the region's largest utility. Utilities across the United States are pausing new data center connections, fast-tracking emergency natural gas plants, and warning of grid shortfalls measured in gigawatts.

Starcloud, a company founded in January 2024, has raised $170 million in a Series A led by Benchmark and EQT Ventures to pursue an audacious alternative: move the data centers to space. The round values Starcloud at $1.1 billion, making it the fastest company in Y Combinator history to reach unicorn status — just 17 months after its accelerator demo day. The investment brings total funding to $200 million, with participation from NFX, Nebular, Y Combinator, Adjacent, 776 Ventures, FUSE, Manhattan West, and Monolith Power Systems.

The Thesis: Unlimited Solar Power in Orbit

The core thesis behind Starcloud is straightforward physics. In low Earth orbit, a satellite has nearly continuous access to unobstructed sunlight — no clouds, no night cycles that match terrestrial patterns, no weather, and no need for land, water, or grid connections. A solar array in LEO receives approximately 1,361 watts per square meter of solar irradiance, compared to an average of roughly 200–300 watts per square meter at ground level after atmospheric losses and day-night cycles. And in the vacuum of space, the primary challenge of data center cooling — removing waste heat from thousands of processors — is solved by radiative cooling, which requires no water or electricity.

Starcloud's CEO Philip Johnston told CNBC that the company's orbital data centers will achieve 10x lower energy costs than terrestrial facilities. The company's internal analysis estimates that a 40-megawatt orbital data center cluster operated for ten years would cost approximately $8.2 million — compared to $167 million for an equivalent terrestrial system when accounting for electricity costs, land, cooling infrastructure, and grid connections.

What Starcloud Has Already Demonstrated

What separates Starcloud from the typical deep-tech pitch is execution speed. With just $3 million in pre-seed funding, the company designed, built, and launched its first satellite — Starcloud-1 — in a record 21 months. The 60-kilogram satellite, roughly the size of a small refrigerator, launched in November 2025 carrying an NVIDIA H100 GPU with 80 GB of RAM — a chip 100 times more powerful than any GPU previously flown in space.

Starcloud-1 achieved a series of historic firsts in orbit. The team successfully trained NanoGPT — a model created by OpenAI founding member Andrej Karpathy — on the H100 using the complete works of Shakespeare, marking the first AI model ever trained in space. They ran Google's Gemma (based on the Gemini architecture) on the orbital GPU, demonstrating the first large language model inference on a high-performance GPU in orbit. And they performed model fine-tuning in space — adapting a pre-trained model to new data while the satellite circled the Earth at 28,000 kilometers per hour.

The Founding Team

Starcloud's founding team combines aerospace engineering, hyperscale computing, and strategy consulting. CEO Philip Johnston spent years at McKinsey advising national space agencies on satellite strategy before pursuing graduate degrees at Harvard, Wharton, and Columbia in applied math and physics. CTO Ezra Feilden comes from Airbus Defence & Space, where he worked on missions including NASA's Lunar Pathfinder, and holds a PhD in Materials Engineering from Imperial College London. Chief Engineer Adi Oltean spent over 20 years at Microsoft, where he deployed some of the first large language models on large GPU clusters, and later worked as a Principal Software Engineer at SpaceX on Starlink. Oltean holds more than 25 patents.

The combination of Oltean's SpaceX satellite manufacturing experience with his Microsoft-era hyperscale computing background is particularly relevant. Building data centers in space is not purely a spacecraft problem or purely a computing problem — it requires deep expertise in both. Having a founding engineer who has built satellite systems at SpaceX's scale and deployed GPU clusters at Microsoft's scale provides a credibility advantage that is difficult for competitors to replicate.

The Satellite Roadmap

Starcloud's roadmap scales aggressively from demonstration to commercial operations. Starcloud-2, planned for October 2026, will carry NVIDIA Blackwell B200 GPUs alongside multiple H100s and an AWS server blade, powered by a 7-kilowatt solar array — roughly 100 times the computing capacity of Starcloud-1. Starcloud-3, targeted for 2027, steps up to a 3-tonne satellite with 200-kilowatt power and optical inter-satellite terminals for high-bandwidth data transfer between orbital nodes.

The long-term vision is genuinely massive. Starcloud's roadmap envisions 40-megawatt orbital data center clusters in the early 2030s — the scale at which the company projects cost parity with terrestrial data centers. Beyond that, the company has outlined a concept for a 5-gigawatt mega-cluster comprising approximately 88,000 satellites with solar arrays stretching up to 4 square kilometers. Whether that ultimate vision is achievable or aspirational, the near-term milestones — Starcloud-2 in 2026, Starcloud-3 in 2027 — are concrete and funded.

The Economic Argument

Starcloud's economic thesis rests on the gap between terrestrial energy costs and orbital solar economics. On Earth, a data center's electricity bill typically represents 30–40% of total operating costs. For AI workloads that run GPUs at full power continuously — training runs that last weeks or months — energy costs dominate the total cost of ownership. And those costs are rising: utilities serving data center corridors are implementing emergency rate increases, and new natural gas generation assets require years of permitting and construction.

In orbit, the 'electricity' is effectively free after the capital cost of the solar arrays. There is no utility bill, no grid connection fee, no carbon tax, and no land cost. Cooling — the second-largest operating expense for terrestrial data centers — is achieved through passive radiation into the cold vacuum of space. The primary costs are launch (which continues to decline as SpaceX scales Starship), satellite manufacturing, and ground station bandwidth for data transfer. Starcloud's analysis suggests these costs cross below terrestrial data center economics at the 40-megawatt scale.

Why This Matters for the Space Economy

Orbital data centers represent a potential inversion of the traditional space business model. Instead of launching satellites to look down at Earth or provide communications, Starcloud is launching satellites to perform computation — selling processing cycles rather than imagery or bandwidth. If the economics work at scale, orbital computing could become one of the largest revenue-generating applications of space infrastructure, potentially exceeding the satellite communications and Earth observation markets combined.

The timing is driven by convergence: AI compute demand is growing exponentially, terrestrial power infrastructure cannot keep pace, launch costs are declining, and GPU efficiency continues to improve. Starcloud is betting that these trends intersect at a point where putting data centers in orbit transitions from science fiction to economic necessity. The $1.1 billion valuation suggests that Benchmark, EQT, and the broader investor community believe that intersection is approaching faster than most expect.

Frequently Asked Questions