By rssfeeds-admin That’s the unglamorous reality driving one of the most audacious infrastructure bets in tech history. If you can’t reliably power a data center on Earth, the theory goes, why not build one in space? It sounds like the premise of a science fiction film. But in 2026, it’s a sector attracting billion-dollar bets from Google, SpaceX, Nvidia, and a wave of well-funded startups. The U.S. GAO just published a formal technology spotlight on orbital data centers. The FCC has received three applications for space computing constellations since January alone. And Starcloud, a YC-backed startup, became the fastest unicorn in YC history in March 2026, just 17 months after demo day, after raising a $170M Series A led by Benchmark.
So, are data centers in space actually possible? The answer is more nuanced and more imminent than most people realize.
Why Space? The Case Is Stronger Than It Sounds
To understand why serious engineers and serious capital are chasing this idea, you have to understand what’s broken about the terrestrial model.
AI training workloads are extraordinarily power-hungry, and that demand is growing faster than grids can keep pace. Hyperscalers like Alphabet, Amazon, Microsoft, and Meta are projected to spend $400 billion on terrestrial data centers in 2026 alone. And that spending is increasingly constrained not by capital, but by where you can physically plug in.
Space solves the power problem fundamentally. Orbital data centers can access solar energy continuously, with no clouds, no night (in certain orbits), no grid connections, and no permitting battles. Some sun-synchronous orbit planes have constant sunlight, providing effectively uninterrupted solar energy. There’s also the cooling angle. On Earth, data centers are notorious water consumers. In orbit, excess heat can theoretically be radiated directly into the cold of space, eliminating water-based cooling systems. And building a hyperscale campus on Earth means land acquisition, environmental review, transmission infrastructure, and local politics. In orbit, none of those constraints apply. Space-based data centers could reduce the land, electricity, and water demands that make terrestrial facilities so contentious.
What’s Already Happened
This isn’t purely theoretical anymore. The proof-of-concept phase is underway, and some results are genuinely impressive.
In November 2025, Starcloud launched its first satellite carrying an Nvidia H100 GPU, roughly 100 times more powerful than any GPU previously operated in space, becoming the first entity to train a large language model in orbit and the first to run a version of Google’s Gemini from space. At Nvidia’s GTC conference in March 2026, CEO Jensen Huang announced the company would build data centers in space and is developing a new version of its Vera Rubin chip platform specifically for orbital deployment, with a Space-1 module expected in 2027. Google is pursuing its own initiative through Project Suncatcher, an 81-satellite cluster being built in partnership with Planet, with two prototype satellites set to launch in early 2027.
The commercial services layer is also starting to emerge. Lonestar Data Holdings announced StarVault in April 2026, describing it as the world’s first commercially operational space-based sovereign data storage platform, launching in October aboard LizzieSat-4. Separately, Atomic-6 launched ODC.space, a marketplace where enterprises can order orbital data center capacity, sovereign satellites, or colocated nodes, the way they’d order rack space from a colo provider, with sovereign racks starting at $3.5 million a month and delivery running two to three years.
The Engineering Obstacles Are Real
Before anyone moves their workloads off-planet, there are some stubborn problems to solve.
Cooling is harder than it looks. Space is cold, but it’s also a vacuum, and heat doesn’t dissipate in a vacuum the way it does in air. The extreme heat generated by AI chips will probably not easily dissipate on its own in orbit, and while heat radiator technologies like those on the ISS exist, they are likely too heavy and too expensive to launch at a data center scale. Launch costs remain prohibitive too: it currently costs around $1,000 per kilogram to reach orbit, and Google’s own analysis concludes that cost must fall by at least a factor of five, to around $200 per kilogram, before space data centers start to make economic sense, a threshold Google projects may not arrive until around 2035. Latency is a genuine constraint as well, since data processed in orbit still has to travel back to Earth. Google’s Project Suncatcher proposes flying satellite groupings in extremely tight clusters to reduce this round-trip delay, but it remains an open engineering challenge for real-time applications. And at the infrastructure level, large orbital data centers would require solar arrays bigger than any yet assembled in space, and cooling solutions at that scale remain entirely unproven.
The Geopolitical Dimension
If space data centers seem like a story about tech companies chasing lower electricity bills, there’s another layer: sovereignty. China launched 12 satellites in May 2025 for a space-based computing constellation, the first of a proposed 2,800-satellite fleet, signaling that Beijing views orbital computing as strategic national infrastructure, not just a commercial opportunity. For enterprises and governments worried about where their data lives and who can access it, a sovereign orbital data center offers something no terrestrial option can: jurisdiction that doesn’t sit within any nation’s borders. That’s a genuinely novel proposition, and it’s one reason governments and financial institutions are reportedly already lining up for products like Lonestar’s StarVault.
The Timeline That Actually Matters
The GAO’s assessment, published just days ago, is probably the most sober read on timing: deployment of some data center satellites is planned by the mid-2030s, with several engineering and economic barriers still to clear before large-scale operations are viable. But “mid-2030s for large-scale” doesn’t mean nothing is happening now. The more accurate picture is a tiered rollout: small-scale demonstration missions and niche commercial services through 2028, growing constellations and improving economics into the early 2030s, and genuine hyperscale orbital infrastructure as a real possibility by mid-decade, contingent on Starship’s commercial maturity and continued progress on cooling and power systems. Starcloud’s own forecast is more aggressive: the company expects that within 10 years, most new data centers will be built in space. That may be optimistic. But six months ago, training an LLM from orbit also seemed like it was decades away.
What Cloud Teams Should Be Watching
For infrastructure and cloud professionals, the takeaway isn’t to move your Kubernetes cluster to low Earth orbit next quarter. It’s to recognize that the underlying economics of computing are about to be disrupted from an entirely unexpected direction.
The companies that win the next decade of AI infrastructure may not be the ones with the best land deals or the best grid connections. They may be the ones who figured out how to build a power plant that never needs fuel, a cooling system that doesn’t touch water, and a data center that no government can regulate, all while traveling at 17,000 miles per hour, 340 miles above your head.
Possible? The evidence says: increasingly, yes. Practical at scale? That’s the question the next five years will answer.
The post Are Data Centers in Space Really Possible? appeared first on CloudWedge.
Discover more from RSS Feeds Cloud
Subscribe to get the latest posts sent to your email.