SpaceX Plans to Build AI Data Centers in Orbit as Terrestrial Power Runs Short

Ten days after pulling off the largest initial public offering in history, SpaceX is asking investors to believe in something that still sounds like science fiction: artificial-intelligence data centers that orbit the Earth, powered by the sun and cooled by the vacuum of space.

The company unveiled the design of its AI1 Compute Satellite on June 8, 2026 — a 70-meter-wingspan orbital supercomputer that radiates waste heat directly into space and draws on solar power unbroken by clouds or nightfall. The pitch is straightforward and, to a power-starved industry, seductive: launch the compute to where the energy is free and the land is no one's, instead of fighting for grid capacity and water rights in Memphis, Virginia or Arizona.

The timing is not coincidental. SpaceX closed its IPO on June 12, raising roughly $75 billion by selling 555,555,555 shares at $135 apiece. The stock popped 19% on its Nasdaq debut and pushed the company's market capitalization above $2.1 trillion, briefly touching $2.21 trillion at the session high — worth more than Elon Musk's Tesla. Investors, as one analysis put it, are no longer betting on rockets but on an entire orbital ecosystem, with orbital data centers among the most ambitious bets riding the wave.

That ambition is grounded in a very terrestrial cash flow. After folding xAI into SpaceX in February, the company found itself sitting on underused compute at its Colossus complex in Memphis, Tennessee. In May, Anthropic agreed to pay SpaceX about $1.25 billion per month — roughly $45 billion over three years, through May 2029 — for capacity across Colossus and Colossus II, taking more than 300 megawatts of compute. Anthropic also "expressed interest," per the companies' announcements, in helping SpaceX develop multiple gigawatts of compute capacity in space. In January, SpaceX had already filed with the FCC for authorization to operate as many as one million solar-powered satellite data centers in low Earth orbit.

The engineering reality

The physics that makes orbit attractive also makes it brutal. The two engineering professors who reviewed SpaceX's plans for The Conversation — Sven Bilén and Wangda Zuo, of Penn State and the University of Colorado — are blunt about the gap between the brochure and the laws of thermodynamics.

"There is a vast difference between launching satellites and operating an industrial-scale computing infrastructure in orbit," Bilén and Zuo wrote. "Space is unforgiving."

Cooling is the central paradox. Space is cold — about minus 270 degrees Celsius — but there is no air to carry heat away, so waste heat must leave as infrared radiation, a comparatively slow process. The authors estimate that shedding just 10 megawatts of waste heat would require radiator surfaces "comparable to the size of two football fields," on top of the acreage of solar panels needed to generate the power in the first place. Radiation degrades electronics, micrometeorites and orbital debris can puncture a spacecraft, and a single collision could destroy a facility and spawn yet more debris.

Then there is the refresh problem. Earth-based operators swap out servers every three to five years as chips improve; in orbit, that hardware "may be difficult or too expensive to upgrade," the professors warn, risking obsolescence long before the surrounding structure wears out. Their verdict on AI1 is sobering: the satellite is "100 to 1,000 times less capable than current Earth-based data centers." Two prototypes are slated to launch in early 2027, with a commercial constellation to follow.

Why It Matters

The orbital gambit is best read as a symptom of a grid crisis, not a triumph of space engineering. AI's appetite for power is colliding with the limits of US infrastructure, and the spending is staggering: Amazon, Microsoft, Google and Meta are collectively guiding to roughly $750 billion in AI capital expenditure in 2026. Data centers are drawing public backlash over land, water and electricity in community after community. When a company would rather assemble servers in a near-vacuum than win another zoning fight, it signals how binding the terrestrial constraints have become.

But the economics remain unproven, and the most credible near-term use cases are narrow. Bilén and Zuo argue the first viable orbital data centers will likely serve space itself — processing Earth-observation imagery, military and intelligence data, or computing tied to satellite operations — rather than competing with cloud providers on latency-sensitive consumer AI. The Anthropic deal also reveals the awkward truth beneath the hype: SpaceX is monetizing capacity that, by one report, was running near 11% utilization, far below the roughly 40% rivals achieve. The orbital story helps justify a $2 trillion valuation; the Memphis story pays the bills today.

What to Watch

The first real test comes in early 2027, when the two AI1 prototypes are scheduled to fly and reveal whether on-orbit thermal management and high-speed downlink work outside a slide deck. Watch the FCC's response to SpaceX's million-satellite filing, which will draw scrutiny over spectrum and orbital congestion. Watch whether Anthropic converts its stated interest in gigawatt-scale space compute into a binding commitment as its $1.25 billion-a-month terrestrial contract runs toward May 2029. And watch the capex line at the four hyperscalers: if their $750 billion fails to relieve grid pressure, the case for going to orbit — however punishing the physics — only gets stronger.

---

Sources: [ScienceDaily / The Conversation](https://www.sciencedaily.com/releases/2026/06/260618041501.htm), [NPR](https://www.npr.org/2026/06/11/nx-s1-5853199/spacex-ipo-price-elon-musk), [CNBC](https://www.cnbc.com/2026/06/12/spacex-ipo-spcx-live-updates.html), [TechCrunch](https://techcrunch.com/2026/05/20/anthropic-will-pay-xai-1-25-billion-per-month-for-compute/), [Data Center Dynamics](https://www.datacenterdynamics.com/en/news/anthropic-to-use-all-of-spacex-xais-colossus-1-data-center-compute/).