The race to scale artificial intelligence is colliding with a hard constraint on Earth: energy. As demand for AI compute surges and terrestrial data centers face years-long delays tied to power limitations, water-intensive cooling requirements, land availability, and permitting, some of the world’s most influential tech and space leaders are looking beyond the planet itself. Jeff Bezos, Elon Musk, and a growing group of startups and enterprises are exploring whether AI data centers can operate in orbit—powered by the sun and freed from Earth’s energy grid.

The remaining question is no longer just whether orbital data centers are appealing—but whether they can actually function as part of AI infrastructure.

The idea of space-based AI compute is no longer purely theoretical. In recent AiNews.com coverage, an NVIDIA-backed startup, Starcloud, successfully trained an AI model in space, demonstrating that AI workloads can run beyond Earth-based infrastructure. While still experimental, the project offered early proof that orbital computing is technically feasible—even as questions around scale, cost, and long-term performance remain unresolved.

That early technical validation is now drawing the attention of companies with the capital, launch capacity, and long-term ambition to test whether orbital AI compute can scale.

With early experiments demonstrating that AI workloads can operate in space, attention is now shifting to whether orbital compute can scale at an industrial level. Elon Musk and Jeff Bezos, who have spent years competing to build rockets and deploy satellite networks, are now turning that rivalry toward a new frontier: taking the trillion-dollar AI data-center boom into orbit.

According to people familiar with the matter, Blue Origin, the space company founded by Bezos, has had a dedicated team working for more than a year on technologies required for orbital AI data centers, underscoring that the effort extends beyond early-stage exploration. On the SpaceX side, the concept of hosting AI compute on upgraded Starlink satellites has been included in discussions with investors, positioning orbital AI infrastructure as part of the company’s long-term growth narrative rather than a standalone experiment.

The concept has reportedly been pitched to investors as part of discussions that could value SpaceX at up to $800 billion, highlighting how closely orbital AI compute is being tied to the company’s future business model.

Interest in orbital data centers extends beyond private discussions. In October, Jeff Bezos said during an event in Italy that shifting data centers to orbit made sense given the abundance of solar power in space. While he acknowledged that space-based infrastructure would take time to compete with terrestrial AI data centers on cost, Bezos predicted that orbital systems could become economically competitive within 20 years or sooner.

That long-term view aligns with progress at Blue Origin, which has made major strides this year toward demonstrating its New Glenn rocket. The partially reusable vehicle features a large payload fairing designed to carry significant numbers of satellites into orbit—capabilities that could prove critical for deploying AI compute at scale.

Proponents argue that moving data centers to orbit could bypass some of the most stubborn obstacles to powering and cooling AI data centers on Earth. At the same time, deploying satellites capable of meaningful AI compute introduces a new set of engineering and economic challenges, including whether large constellations can be built and launched at a cost that makes orbital data centers viable.

Those challenges are being weighed against mounting constraints on Earth, including:

By operating above the planet, satellites could rely on near-continuous solar power to run AI workloads and transmit results back to Earth, reducing dependence on terrestrial energy infrastructure. “Taking resource-intensive infrastructure off Earth has been an idea for years, but it has required launch and satellite costs to come down,” said Will Marshall, CEO of Planet Labs. “We are nearing that point.”

Among the most concrete efforts underway is a collaboration between Google and Planet Labs, which aims to launch two test satellites in early 2027 equipped with Google’s tensor processing units (TPUs). Google has described the project as a “moonshot,” acknowledging the difficulty of scaling orbital data centers to match terrestrial capacity.

The scale challenge is immense. Travis Beals, a Google executive involved in the project, said it would take roughly 10,000 satellites—each delivering around 100 kilowatts of power—to replicate the compute capacity of a single gigawatt-scale terrestrial data center. Google has emphasized that its 2027 test mission is focused on validating coordination, reliability, and performance, not on achieving cost or capacity parity with Earth-based facilities.

While Big Tech experiments cautiously, startups are moving aggressively. Aetherflux, which is building an American power grid in space, announced plans to launch its first orbital data center satellite, dubbed “Galactic Brain,” with a target date of Q1 2027. The company positions its approach as a way to bypass the five-to-eight-year timelines typically required to build new data centers on Earth.

“The race for artificial general intelligence is fundamentally a race for compute capacity, and by extension, energy,” said Baiju Bhatt, founder and CEO of Aetherflux. “Galactic Brain puts the sunlight next to the silicon and skips the power grid entirely.”

Aetherflux plans to scale capacity by launching a constellation of satellites over time. In 2026, the company also expects to launch its first satellite to test wireless energy transmission from low Earth orbit to Earth using lasers, building on the company’s existing work developing space-solar satellites designed to beam power to contested environments.

A growing group of companies and executives are exploring whether orbital data centers can be made viable, extending well beyond SpaceX, Blue Origin, and Google.

According to reporting by The Wall Street Journal, Sam Altman, CEO of OpenAI, has investigated whether his company could acquire a rocket operator as a way to deploy AI compute directly into space. Eric Schmidt, the former CEO of Google who now leads Relativity Space, has also spoken publicly about the potential for orbital data centers.

Meanwhile, IBM’s Red Hat software division and Axiom Space launched a space-based data-computing prototype earlier this year. Venture-backed startups including Aetherflux and Starcloud are advancing their own approaches, aiming to compete alongside — or partner with — much larger players.

Beyond launch costs, operating AI hardware in orbit introduces complex engineering constraints. High-performance chips must be kept within tight temperature ranges despite the absence of atmospheric cooling, while also being shielded from radiation that can degrade sensitive electronics. At the same time, AI workloads require reliable, low-latency data transmission back to Earth—requirements that become increasingly difficult as satellite constellations scale.

“There’s a bunch of engineering challenges, but I think those engineering challenges are all solvable,” said Jonny Dyer, CEO of Muon Space. “It ultimately comes back to launch.”

Launch economics may determine whether orbital data centers ever move beyond niche use cases. SpaceX has already reshaped launch frequency with its reusable Falcon 9 fleet, supporting both its Starlink satellite network and third-party payloads. That high-cadence approach has helped SpaceX reduce costs while spreading development expenses across frequent missions.

The scale required for orbital data centers could significantly reshape the aerospace supply chain. Launching thousands of satellites-as-data centers would drive sustained demand for rockets, components, and launch services. Industry executives note that while rocket development is expensive and complex, frequent launches—rather than occasional missions—are essential to making orbital data centers economically viable, allowing operators to amortize fixed costs and improve margins over time.

SpaceX is aiming to push launch costs even lower with Starship, the massive next-generation rocket it is developing. According to people familiar with the matter, AI-compute hardware would be installed on upgraded satellites designed specifically to fit Starship’s payload capacity. While the vehicle is not yet operational, SpaceX plans to debut an upgraded version early next year. In a recent post on X, Elon Musk said Starship could eventually deliver “around 300 gigawatts (GW) per year of solar-powered AI satellites to orbit, maybe 500 GW.”

For other AI companies interested in orbital data centers, access to affordable, reliable launch capacity remains a fundamental question.

Q: Why are AI companies considering space-based data centers?
A: The primary driver is energy. Training and running advanced AI models requires massive, continuous power, and new terrestrial data centers can take five to eight years to build due to grid, land, and permitting constraints. Space offers near-constant solar energy without relying on Earth’s power infrastructure.

Q: Are orbital data centers replacing Earth-based facilities anytime soon?
A: No. Current efforts are experimental and complementary. Even optimistic projections suggest orbital data centers will serve specialized or incremental workloads rather than fully replacing ground-based infrastructure in the near term.

Q: What companies are actively working on orbital AI compute?
A: Blue Origin, SpaceX, Google, Planet Labs, Aetherflux, Starcloud, and others. IBM’s Red Hat and Axiom Space have also tested space-based computing prototypes.

Q: What are the biggest technical challenges?
A: Managing heat for AI chips in vacuum, protecting hardware from radiation, ensuring reliable data transfer to Earth, and launching enough satellites at sustainable cost.

Q: Has AI actually been run in space yet?
A: Yes. A startup backed by NVIDIA, Starcloud, has successfully trained and run an AI model in orbit, offering the clearest proof so far that space-based AI compute can work under real conditions.

Orbital data centers are no longer just a thought experiment. While the economics and scalability remain uncertain, recent progress suggests the technical barriers may be more solvable than once assumed.

A startup backed by NVIDIA, Starcloud, has successfully trained and run an AI model in orbit—marking the first real proof that space-based data centers could become a viable part of future AI infrastructure. By relying on solar power in space and innovative approaches to thermal management and cooling, early projects are beginning to address the energy and heat challenges that constrain AI compute on Earth.

This does not mean the AI industry is abandoning terrestrial data centers. Instead, orbital compute is emerging as a potential new layer of infrastructure—one that could supplement Earth-based systems as demand continues to outpace energy availability.

For companies like Bezos’ Blue Origin, Musk’s SpaceX, Google, and energy-focused startups such as Aetherflux, the long-term bet is that launch costs will decline faster than Earth’s energy and infrastructure constraints can be resolved. If that happens, space may shift from being a symbolic frontier for AI to a practical extension of its infrastructure.

The next decade will determine whether orbital data centers remain niche experiments—or become a strategic pressure valve for the AI industry’s growing energy problem.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing, image, and idea-generation support from ChatGPT, an AI assistant. However, the final perspective and editorial choices are solely Alicia Shapiro’s. Special thanks to ChatGPT for assistance with research and editorial support in crafting this article.