The Next Cloud: Startups and Giants Race to Build AI Data Centers in Space

The artificial intelligence boom is colliding with the physical limits of Earth. As AI models grow exponentially more complex, the terrestrial data centers required to train them are draining local power grids, consuming millions of gallons of water for cooling, and sparking fierce opposition from communities unwilling to host the massive facilities.[1]

In response, a new wave of aerospace startups and tech giants are looking upward. Their proposed solution is to move the most energy-intensive computing workloads into low Earth orbit, creating a new class of "orbital data centers" that rely on the continuous, unfiltered solar energy of space.[1]

The concept is rapidly moving from science fiction to commercial reality. Starcloud, a Redmond, Washington-based startup, recently announced $170 million in Series A funding, vaulting it to a $1.1 billion valuation. The milestone made the 17-month-old company the fastest to reach unicorn status in the history of the Y Combinator accelerator.

Starcloud has already proven the core technology works. In late 2025, the company launched a 130-pound satellite carrying an Nvidia H100 chip and successfully trained a large language model in orbit. The startup is now partnering with Crusoe Energy to launch the first public cloud in space by 2027, utilizing the upcoming Blackwell B200 chips.

Other startups are quickly joining the orbital race. Los Angeles-based Orbital recently secured funding from Andreessen Horowitz's A16z Speedrun accelerator to launch its first test mission next April. Meanwhile, Pasadena-based Sophia Space is leveraging passive cooling technology developed by Caltech and NASA's Jet Propulsion Laboratory to manage the extreme thermal fluctuations of lower Earth orbit.

The startup ecosystem is now competing directly with the world's most valuable private aerospace company. SpaceX is preparing to deploy a massive network of orbital data centers, filing plans for a megaconstellation that could eventually include up to one million AI-focused satellites.[1]

SpaceX recently unveiled the "AI1"—its first-generation data center satellite. The massive spacecraft spans 70 meters tip-to-tip and utilizes a 110-square-meter deployable liquid radiator to reject the immense heat generated by its 150-kilowatt computing payload into the vacuum of space.

To support this ambition, SpaceX is planning a massive "Gigasat" manufacturing complex in Bastrop, Texas. The 11-million-square-foot facility will be dedicated to producing the solar arrays, printed circuit boards, and cooling systems required to mass-produce the AI1 spacecraft.

Traditional hyperscalers are also exploring the orbital frontier. Google has partnered with satellite operator Planet on "Project Suncatcher," an experimental initiative aiming to test Google's proprietary Tensor Processing Unit (TPU) hardware in space by early 2027.

Google's internal estimates suggest that for space-based compute to reach price parity with terrestrial data centers, the cost of launching payloads into orbit must fall to roughly $200 per kilogram. While current heavy-lift rockets are driving prices down dramatically, the industry projects that this breakeven point may not be reached until the mid-2030s.

Before these orbital server farms can scale, the industry must solve a distinctly terrestrial problem: insurance. Startups like Starcloud, Orbital, and Lonestar Data Holdings have initiated talks with major underwriters, including Lloyd's of London and Marsh, to figure out how to insure these unprecedented assets.[3]

While the space insurance market routinely covers launch failures and standard satellite malfunctions, orbital data centers introduce entirely new risk models. Underwriters are currently grappling with how to price the risks of radiation degradation on sensitive AI chips, thermal management failures, and the growing threat of space debris collisions.[3]

Securing this insurance is a mandatory step for the industry's growth. Without comprehensive coverage for the expensive orbital hardware, startups will be unable to secure the massive debt financing required to build out multi-billion-dollar satellite constellations.[3]

Not every visionary believes space is the right escape valve for AI's energy crisis. Panthalassa, a Portland-based startup backed by Peter Thiel, argues that the harsh radiation and extreme temperatures of orbit make space-based compute unnecessarily expensive and complex.[2]

Instead, Panthalassa is building floating data centers designed to operate in the ocean. By generating their own electricity from open-ocean waves and utilizing cold seawater for highly efficient cooling, the company claims it can offer a zero-emission alternative that is significantly cheaper than both terrestrial and orbital data centers.[2]

Whether the future of AI infrastructure floats in the Pacific Ocean or orbits 300 miles above the Earth, the consensus is clear: the current model of sprawling, grid-taxing terrestrial data centers is unsustainable. As the space economy matures, the cloud may soon take on a much more literal meaning.[1][2]