Google is engaging in discussions with SpaceX and other potential launch providers to advance Project Suncatcher, their research initiative to deploy artificial intelligence infrastructure in orbit. The information, initially reported by The Wall Street Journal and later covered by Reuters and TechCrunch, positions Elon Musk’s company as a natural candidate to launch the first prototypes of this experiment: satellites powered by solar energy and equipped with Google TPU chips.
This news aligns with a trend that once sounded almost absurd but is now gaining mainstream attention: building data centers in space. The energy demands of AI, limited land availability, electrical power shortages in key regions, and social opposition to new data center campuses are pushing Google, SpaceX, NVIDIA, Starcloud, and Anthropic to explore an extreme idea: shifting some computing capacity off the planet.
Project Suncatcher: TPUs, Satellites, and Nearly Continuous Solar Power
Google unveiled Project Suncatcher in November 2025 as a “moonshot” research initiative. Its vision is to create compact constellations of solar-powered satellites, equipped with TPUs and connected via free-space optical links. The initial goal isn’t to rapidly build an orbital hyperscale data center, but to validate whether AI hardware can operate reliably in space, communicate with low latency among satellites, and harness solar energy more continuously than on Earth.
The next step outlined by Google involves a joint learning mission with Planet to launch two prototype satellites early in 2027. According to the design described by Google Research, the long-term concept includes clusters of satellites flying in close formation, each equipped with TPUs and linked by optical connections. The associated Suncatcher paper illustrates a potential cluster of 81 satellites within a 1-kilometer radius, though this is still in the research phase, not a commercial cloud product ready for clients.
A collaboration with SpaceX appears logical. SpaceX dominates much of the commercial launch market, has experience operating constellations like Starlink, and maintains a launch cadence that few competitors can match. Reuters emphasizes that developing orbital data centers is emerging as a key technological narrative around a possible future IPO for SpaceX—an endeavor requiring enormous investments and drastically reduced launch costs.
Not Just Google: NVIDIA, Starcloud, SpaceX, and Anthropic Looking Upward
The enthusiasm for space-based data centers isn’t limited to Google. NVIDIA announced its “space computing” initiative in March 2026, with platforms such as Space-1 Vera Rubin, IGX Thor, and Jetson Orin to bring accelerated computing to space missions, satellites, orbital vehicles, and future in-orbit analysis systems. Partners like Aetherflux, Axiom Space, Kepler Communications, Planet, Sophia Space, and Starcloud are involved in these efforts.
Starcloud stands out as one of the most ambitious names. The startup, part of NVIDIA’s Inception program, envisions orbital data centers specifically designed to run cloud and AI workloads in space. NVIDIA reports that Starcloud’s concept involves a 5 GW orbital data center with large solar panels and cooling structures about 4 kilometers wide and long. The company argues space could reduce energy costs and eliminate the need for water-based cooling systems used on land, though these estimates depend on launch and maintenance costs decreasing sufficiently.
SpaceX is also shifting gears from a different angle. Its SpaceXAI division announced a deal with Anthropic to provide access to Colossus 1, a supercomputer featuring over 220,000 NVIDIA GPUs, including H100, H200, and GB200 models. In the same announcement, Anthropic expressed interest in collaborating to develop several gigawatts of orbital computing capacity. While this doesn’t turn the project into an immediate reality, it signals that the idea has transitioned from science fiction to strategic conversations among AI companies.
Underlying all this is a simple truth: AI is reaching physical limits. New models demand more chips, more power, more cooling, and larger data centers. On Earth, this involves negotiations with utilities, land purchases, permitting, water management, grid reinforcement, and community engagement—challenges many find difficult to surmount. In orbit, the Sun offers a plentiful and almost continuous energy source along certain trajectories, but introduces other problems.
The Tough Part: Launch, Cooling, Repair, and Connectivity
The primary obstacle remains cost. Putting hardware into orbit is expensive—even with reusable rockets. Additionally, a data center isn’t just racks of chips; it requires structural support, energy supply, radiators, communications, radiation shielding, thermal management, redundancy, maintenance, and end-of-life strategies to prevent contributing to space debris.
It’s also important to nuance the idea that “space cools for free.” In the vacuum, there’s no air to remove heat via convection. Heat must be dissipated through radiation, which demands large radiative surfaces and carefully engineered thermal designs. Higher computing densities make heat evacuation increasingly challenging. On land, cooling consumes water and electricity; in orbit, it requires designing lightweight, deployable, and highly reliable structures.
Connectivity is another challenge. For an orbital data center to be useful for terrestrial AI, it needs to transfer vast quantities of data between satellites and ground stations. Optical links can offer high bandwidth but require precise alignment, continuous availability, and complex constellation management. Such setups make sense for tasks like Earth observation, in-orbit data analysis, or satellite-based inference. However, replacing a terrestrial data hub with one in space faces much higher demands in terms of latency, reliability, and coverage.
Environmental and regulatory risks can’t be ignored. Deploying thousands of additional satellites increases orbital congestion, hampers astronomical observations, and prompts stricter rules on coordination, deorbiting, and liability. If the industry tries to solve one environmental problem on Earth and creates another in space, public acceptance will be difficult to achieve.
Most cautious experts believe that space data centers won’t replace terrestrial ones in the coming years. The initial phase will likely focus on in-orbit processing, satellite image analysis, defense, meteorology, communications, autonomous navigation, and less latency-sensitive AI workloads. Later, if launch costs fall enough and systems like Starship meet expectations, more ambitious clusters could emerge.
Nevertheless, Google’s movement shifts the tone of the debate. When a company with Google’s experience in chips, cloud, networking, and long-term projects experiments with TPUs in space, it’s no longer just science fiction. The involvement of SpaceX, NVIDIA, Starcloud, and Anthropic indicates that AI infrastructure is seeking new frontiers as terrestrial options become saturated.
The question is no longer if someone will deploy AI servers in space. That’s already underway on a small scale. The real question is whether this can become a competitive, secure, maintainable, and cost-effective infrastructure. The outcome will determine if orbital data centers remain a spectacular demonstration or evolve into a new layer of the global cloud.
FAQs
What is Project Suncatcher?
Project Suncatcher is Google’s research initiative to explore whether it’s feasible to deploy AI computing in space using solar-powered satellites equipped with TPUs and connected via optical links.
Why is Google talking to SpaceX?
According to Reuters and other outlets, Google is in discussions with SpaceX and other launch providers for future missions related to Project Suncatcher. SpaceX has an advantage due to its high launch frequency and experience running satellite constellations.
Which other companies are working on space data centers?
NVIDIA is developing space computing platforms with partners like Starcloud, Planet, Axiom Space, and Kepler. SpaceXAI and Anthropic have also expressed interest in developing gigawatt-scale orbital AI capacity.
Will space-based data centers replace terrestrial ones?
Not in the near term. They will first be useful for specific tasks such as processing in-orbit data, Earth observation, or local inference in satellites. Large orbital AI centers need to overcome issues related to cost, cooling, maintenance, communication, and regulation.