Space-Based Data Centers Sound Crazy. Here's Why Companies Are Considering Them

Putting a data center in orbit sounds like the kind of idea that gets pitched, laughed at, and forgotten. In 2026, it's instead attracting hundreds of millions of dollars in real funding, Nvidia's direct involvement, and a serious technical case — not because it's easy, but because the alternative on the ground is running into limits of its own.

The company furthest along

Starcloud, a Seattle-area startup, has raised a $250 million Series A extension at a $2.3 billion valuation, led by Manhattan West with participation from Nvidia, Benchmark, and Cisco Investments, bringing its total funding to $450 million. The company is building production lines for its Starcloud-3 satellites at a new 100,000-square-foot manufacturing facility in Woodinville, Washington, and has filed with the FCC to eventually operate as many as 88,000 satellites configured as orbital data centers. This isn't purely theoretical: in November 2025, Starcloud successfully flew an Nvidia H100 GPU to orbit and used it to train a small AI model, NanoGPT, in space — a proof of concept that space-based AI compute works at all.

Nvidia's stake in making this real

Nvidia isn't just an investor here; the company is providing its Space-1 Vera Rubin Module, which it says will deliver 25 times as much in-space compute capability as the H100 already tested, with Starcloud's satellites serving as early flight platforms for that space-rated hardware. That's a meaningful bet from the world's dominant AI chipmaker that orbital compute is a real category, not a novelty.

The actual case for going to orbit

The logic isn't about speed or convenience — it's about the terrestrial bottlenecks that are increasingly constraining data center growth on the ground: available land near power and cooling infrastructure, the electricity itself, and enormous water demand for cooling. Orbital data centers sidestep all three at once: solar power is effectively constant and unfiltered by atmosphere, there's no competing land use, and the vacuum of space offers a fundamentally different cooling environment than a terrestrial facility fighting heat with water and air conditioning. Whether that trade-off nets out favorably once you account for launch costs and satellite lifespan is the real open question — but it's the same terrestrial power and land crunch driving hyperscalers toward nuclear deals and factory-scale data center buildouts that makes the orbital pitch worth taking seriously at all.

Solar power in orbit is effectively constant. Land isn't contested. Cooling works completely differently in a vacuum. None of that makes space-based compute easy — it just makes the terrestrial alternative look more constrained by comparison.

Starcloud isn't the only one chasing this

SpaceX, through a program reported as Starmind, has filed its own plans for as many as a million data-center satellites — a scale that dwarfs even Starcloud's ambitions and leverages SpaceX's existing launch cadence and Starlink manufacturing experience. That two well-funded, differently positioned players are both pursuing this at serious scale is itself a signal that the idea has moved past the novelty stage, even if neither has anything close to a fully operational orbital data center running yet.

What nobody's solved yet

The unresolved problems are not small. Radiation in orbit degrades electronics faster than on the ground, and there's no equivalent of sending a technician to swap out a failed GPU on a satellite the way you would in a terrestrial facility. Launch costs, even with reusable rockets, remain a real line item that a ground-based data center simply doesn't have. And getting data to and from orbit fast enough for AI workloads that expect low latency is its own unsolved engineering problem, distinct from the compute and power questions. None of this makes the idea absurd — it makes it a genuine bet on hard, unresolved engineering, backed by real money because the terrestrial constraints it's trying to route around are just as real.