Google and SpaceX explore orbital data centers to power the next wave of AI

Home » Python » Google and SpaceX explore orbital data centers to power the next wave of AI

Google is quietly holding advanced discussions with SpaceX to send data center hardware into orbit, betting that space-based infrastructure could one day relieve the pressure that artificial intelligence is putting on the planet’s energy and land resources. What only a few years ago sounded like pure science fiction is now being treated as a serious strategic option inside two of the world’s most influential tech companies.

Under an internal initiative known as Project Suncatcher, Google wants to test small racks of AI servers aboard satellites before the end of this decade. SpaceX, already the dominant provider of commercial launches, sees orbital data centers as the next big pillar of its business and a key storyline for a blockbuster stock market debut. Between them, they are trying to work out whether moving compute off the planet can make sense technically and economically.

Google’s Project Suncatcher: taking AI compute off the planet

According to multiple reports, including coverage from The Wall Street Journal, Bloomberg and Reuters, Google has been developing Project Suncatcher for months as a long‑term research and infrastructure effort. The program’s central idea is straightforward: build a mesh of solar‑powered satellites equipped with Google’s custom Tensor Processing Units (TPUs) and link them together into an orbital cloud for AI workloads.

Google first publicly acknowledged Suncatcher in November last year, describing it as an experiment in running machine‑learning infrastructure beyond Earth. CEO Sundar Pichai later told Fox News Sunday that the plan is to “send small racks of machines on satellites, test them and then start scaling from there,” adding that he expects space‑based data centers to become a relatively normal way of building infrastructure “within about a decade.”

Under the current roadmap, Google and satellite operator Planet Labs intend to launch at least two prototype spacecraft carrying TPUs into low Earth orbit by early 2027. These early missions are framed as “learning” exercises: validate the hardware in space, measure performance, and understand the operational challenges before committing to a broader roll‑out.

Behind that cautious approach sits a very down‑to‑earth problem. The International Energy Agency projects that global data center electricity consumption could more than double by 2030, reaching around 945 TWh, with AI as the main driver. In the United States alone, data centers could account for nearly half of the expected growth in power demand by the end of the decade, forcing cloud providers to look for new types of generation and new locations.

Google has already been securing nuclear, geothermal and other low‑carbon energy sources for its terrestrial campuses, but the company now views orbit as a more radical option. In the right orbits, a solar panel can harvest up to eight times more energy per year than a comparable panel in mid‑latitude locations on the ground, and certain sun‑synchronous trajectories can keep satellites in near‑continuous sunlight, reducing the need for heavy batteries.

Why SpaceX is central to Google’s orbital ambitions

To make Suncatcher more than a paper exercise, Google needs affordable access to space, and that is where SpaceX enters the picture as the leading supplier of reusable launches. The Wall Street Journal reports that Google is in active talks with Elon Musk’s company over a potential launch agreement that would help loft its first orbital data center prototypes.

At the same time, Google is not limiting itself to a single launch provider. The company has been speaking with other rocket firms as well, keeping options open with players such as Blue Origin, Rocket Lab or United Launch Alliance. This multi‑vendor stance reflects how early and experimental the orbital compute market still is; no one wants to be tied to a single partner before the economics are clearer.

There is also a longstanding financial link between the two firms. Google invested about $900 million in SpaceX back in 2015, a stake that has been reported at roughly 6.1% of the space company’s equity. Don Harrison, a senior Google executive, sits on SpaceX’s board, giving Alphabet direct visibility into Musk’s ambitions around rockets, satellites and AI infrastructure.

Despite this connection, the relationship is not straightforwardly cozy. Google and SpaceX are simultaneously prospective partners and future rivals in orbital computing, each sketching its own vision of how AI data centers might be deployed beyond Earth. Any launch deal would effectively pair two competitors in order to test a technology that neither side can yet prove at industrial scale.

From SpaceX’s perspective, the interest is not just about selling rocket rides. The company has formally applied to U.S. regulators for permission to deploy a “system of orbital data centers” that, on paper, could include up to one million satellites dedicated to on‑orbit computation. For comparison, there are currently estimated to be fewer than 17,000 satellites of all types in orbit.

Inside the Suncatcher architecture: a modular orbital cloud

Beneath the branding, Project Suncatcher is essentially a proposal for a highly modular, distributed data center made up of many small satellites. Instead of a single monolithic facility, Google envisions constellations of compact spacecraft flying in tight formations, each one hosting TPUs and memory for AI training and inference tasks.

In a technical paper, Google sketches an example formation of 81 satellites confined within a radius of about one kilometer. To operate as a coherent data center, those spacecraft would need to share data at extremely high throughput, with the company citing target link speeds of up to 10 Tbps between neighbors using dense wavelength‑division multiplexing optical links similar to those already used in terrestrial data centers.

Achieving that bandwidth requires the satellites to fly relatively close to one another, which adds a whole separate layer of complexity. The company expects to rely on AI‑driven control systems to continuously adjust each spacecraft’s position and avoid collisions, accounting for gravity, atmospheric drag in low orbits and other perturbations that can nudge satellites out of place.

On top of orbital dynamics, thermal management and radiation protection loom as critical engineering issues. Google’s TPUs are high‑power chips that generate significant heat, and in a vacuum there is no air to carry that heat away. Proposed solutions include elaborate heat pipes and radiators, but designing and validating such systems for dense AI workloads in orbit remains an open challenge.

The memory side is no easier. Modern high‑bandwidth memory (HBM) used in AI accelerators is sensitive to radiation, which can cause data errors and corruption if not properly mitigated. Hardening these components or adding robust error‑correction schemes is likely to increase cost, mass and design complexity—exactly the parameters that orbital operators try to minimize.

Launch costs: the make‑or‑break variable

Beneath all the futuristic diagrams, the economic bottleneck for orbital data centers is still the price per kilogram to low Earth orbit. Google has estimated that if launch costs fall to around $200 per kilogram, the annualized energy cost of running a Suncatcher‑style constellation could begin to look comparable to that of a modern ground‑based data center.

Right now, industry estimates still put typical launch prices in the range of about $1,500 to $2,000 per kilogram for many missions, even with partially reusable rockets. SpaceX and other players say they are working toward much lower prices later in the 2030s, especially as fully reusable vehicles such as Starship mature.

In one of the more optimistic scenarios discussed around SpaceX’s roadmap, advanced reusability and high‑cadence launches could bring the cost down to roughly $60 per kilogram. If those numbers were achieved and sustained, the economics of shipping compute hardware into orbit would look very different from today, potentially tipping the balance toward large‑scale deployment.

For now, that remains a projection. Analysts and industry observers, including outlets like TechCrunch, caution that terrestrial data centers continue to be markedly cheaper when the full costs of satellite design, launch, on‑orbit operations and eventual replacement are factored in. The gap between long‑term vision and present‑day financial reality is one of the main reasons why many experts still treat orbital compute as a speculative bet.

SpaceX is using that long‑term narrative as a selling point in presentations to potential investors ahead of a planned IPO. Reports suggest the company is targeting a public listing this summer, aiming for a valuation as high as $1.75 trillion, with orbital data centers playing a central role in the story it tells about future growth.

Musk’s orbital data center push and the wider AI ecosystem

For Elon Musk, putting AI data centers into orbit has become a personal and strategic obsession. After learning of Google’s Suncatcher plans, he reportedly elevated orbital computing to a top priority inside SpaceX, presenting it as the only credible way to scale AI compute far beyond the constraints of terrestrial power grids and land availability.

Musk has argued that the real bottleneck for advanced AI is not chips but electricity, and that the “cheapest place to put AI” over the long run will be space, where satellites can draw essentially continuous solar power without night, clouds or atmospheric interference. In some of his more upbeat predictions, he has suggested that within two or three years, space‑based compute could become the most economical way to run large‑scale AI systems.

Even SpaceX’s own documentation for regulators and investors strikes a more cautious note, however. The company acknowledges that orbital data centers depend on multiple unproven technologies and may ultimately fail to be commercially viable. Cooling, radiation, latency and the difficulty of on‑orbit servicing all pose significant risks.

Meanwhile, SpaceX is weaving orbital compute into a broader AI infrastructure ecosystem that spans both space and ground. Recent filings and reports describe a merger of SpaceX and xAI into a single entity valued at around $1.25 trillion, effectively combining rockets, satellites and AI compute under one corporate umbrella.

SpaceX has also inked a high‑profile deal with AI company Anthropic. Under that agreement, Anthropic will get access to SpaceX’s Colossus 1 data center in Memphis, backed by more than 220,000 Nvidia GPUs and roughly 300 megawatts of power. The two companies have also expressed interest in collaborating on “several gigawatts” of orbital AI compute capacity over the longer term, further blurring the line between terrestrial and space‑based infrastructure.

Environmental pressure and the search for new compute frontiers

The push toward orbital data centers is emerging against a backdrop of mounting concern about the environmental footprint of AI. Sally Radwan, digital director at the United Nations Environment Programme, has warned that while the full impact is still not well understood, the early signals are worrying enough to justify careful scrutiny before deploying AI at massive scale.

UN analyses highlight that AI’s environmental cost extends beyond electricity use. It includes the extraction of rare materials, the growing volume of electronic waste from hardware upgrades, the water needed to cool dense server farms and the greenhouse‑gas emissions tied to power generation. Those factors are pushing policymakers and local communities to resist large new data center complexes in some regions.

For proponents of orbital compute, space offers a way to sidestep at least some of these constraints. Satellites powered by solar panels do not compete for land, do not draw on local water supplies for cooling and do not directly stress terrestrial grids. Supporters argue that, over time, shifting a portion of AI workloads off‑planet could ease pressure on communities that are already hosting multiple hyperscale facilities.

Skeptics respond that space is not a free environmental lunch. Manufacturing, launching and eventually de‑orbiting thousands or even millions of satellites has its own material and energy costs, and the risk of worsening orbital debris is a growing concern for space agencies and regulators. The trade‑offs between ground‑based and orbital approaches remain the subject of active research rather than settled science.

What is clear is that AI’s power demands are forcing cloud providers and space companies into closer collaboration than at any previous point. It was not long ago that Elon Musk co‑founded OpenAI partly as a check on Google’s AI ambitions, following disagreements with Larry Page over safety. Today, Musk’s SpaceX is negotiating with Google over launches, even as both firms angle to dominate a new, still hypothetical layer of space infrastructure.

A nascent orbital compute race, with rivals and open questions

Google and SpaceX are not the only ones testing the waters. Other startups and investors are beginning to frame orbital data centers as a distinct market, suggesting that the idea has gained at least some traction beyond a handful of tech giants.

Baiju Bhatt, co‑founder of Robinhood, recently rebranded his space startup as Cowboy Space Corporation and raised about $275 million to pursue orbital data centers powered by in‑house rockets. His argument is that launch capacity will remain constrained for years, so owning the rockets is the only way to guarantee regular access to orbit for compute payloads.

Cowboy Space is reportedly targeting its first launch before the end of 2028. While that timeline trails Google’s 2027 prototype plans, it underlines the belief among some entrepreneurs that orbital compute could mature into a meaningful business line in the medium term if launch economics and satellite technology continue to improve.

Established aerospace players could also find themselves drawn into the field. Blue Origin, United Launch Alliance and Rocket Lab are all plausible launch partners for orbital compute experiments if they can offer competitive prices and mission profiles. So far, Google has not named specific alternatives to SpaceX, but industry watchers assume these names are part of any broader vendor review.

For now, much of the activity lives in regulatory filings, investor decks and early‑stage design documents rather than production‑grade systems. Even optimistic backers concede that challenges like in‑orbit servicing, hardware replacement and long‑term reliability under radiation have no proven playbook at the scale required for modern AI workloads.

Despite that uncertainty, the very fact that Google, SpaceX and others are openly negotiating orbital data center projects marks a shift. What once sounded like an outlandish concept has moved into serious corporate planning, complete with timelines, prototypes and multi‑billion‑dollar investment narratives.

Taken together, the moves by Google and SpaceX suggest that the next chapter of AI infrastructure may stretch far beyond new terrestrial campuses. If Project Suncatcher, SpaceX’s own orbital compute plans and the efforts of emerging competitors bear fruit, the most advanced data centers of the 2030s might not have street addresses at all, but orbital coordinates circling high above the Earth.