New Delhi — OpenAI CEO Sam Altman dismissed orbital data centers as unfeasible, directly challenging visions from Elon Musk, Jeff Bezos and Google. Speaking Friday in a live interview here, Altman said basic math shows the idea won’t work at scale anytime soon.

“Do the very rough math of launch costs relative to the cost of power we can do on Earth,” Altman stated. He pointed out that fixing a broken GPU hundreds of miles up remains unsolved. “Orbital data centers are not something that’s going to matter at scale this decade,” he added.

Elon Musk’s SpaceX leads the most aggressive push. The company filed with the U.S. Federal Communications Commission to deploy up to one million satellites as orbital data centers. Each would stretch 31 miles long and orbit over 310 miles above Earth, according to the filing. Proponents argue space-based solar panels produce eight times more power than ground systems. They also avoid lengthy permitting for Earth-based facilities.

Altman countered that cooling poses a massive hurdle. Space’s vacuum insulates well but traps heat. The International Space Station relies on a 10-tonne ammonia system to manage just 70 kilowatts—the output of one GPU rack. A full data center would demand radiator panels covering a square kilometer, he implied.

Google presses ahead regardless. Project Suncatcher plans prototype satellites with Trillium AI chips by early 2027. Bezos co-leads Project Prometheus, a secretive effort for gigawatt-scale space data centers, sources say. Startups including Starcloud and Aetherflux chase similar goals. China already orbits 12 satellites from a planned 2,800-unit computing network.

Launch economics undermine the rush. Costs sit around $1,000 per kilogram. Google’s research pegs economic viability at one-fifth that level. Altman conceded the concept might work eventually. “Orbital data centers could make sense someday,” he said.

Musk and Altman share a history as rivals. Musk co-founded OpenAI but left amid disputes. SpaceX dominates launches, powering much of the AI boom with Starship development. Bezos’ Blue Origin lags but invests heavily in orbital infrastructure.

Terrestrial data centers face their own strains. Power demands from AI training explode. Nvidia GPUs guzzle electricity, sparking debates over nuclear restarts and grid upgrades. Space advocates promise boundless solar energy above clouds and weather.

Yet Altman’s math highlights trade-offs. A Falcon 9 launch costs $67 million for 22 tonnes to low Earth orbit—over $3,000 per kilogram in practice. Starship aims lower, but reliability stays unproven at scale. Repair missions add billions more.

Industry watchers split on timelines. Some predict space compute by 2030 if reusability drops costs. Others echo Altman: Earth power and fiber optics suffice for now. OpenAI itself expands ground facilities, including a Stargate supercomputer with Microsoft in Texas.

Altman’s remarks ripple through Silicon Valley. Investors poured billions into space startups last year. His skepticism may cool enthusiasm, even as prototypes loom.