Google and SpaceX Are Talking About Putting AI Data Centers in Space

The Wall Street Journal reported Tuesday that Google and SpaceX are in active talks to build data centers in orbit—and the conversation has gotten serious enough to warrant the word "partnership." If a deal comes together, it would mark the beginning of a new era for AI infrastructure: one where the compute powering your AI assistant might be thousands of kilometers above your head.

This isn't science fiction. Both companies have concrete plans, and the economics—while still challenging today—are moving in the right direction.

Google's Project Suncatcher

Google isn't approaching this as a speculative bet. CEO Sundar Pichai formally announced Project Suncatcher in late 2025: a roadmap to deploy a constellation of solar-powered satellites equipped with Google's custom Tensor Processing Units (TPUs—the specialized chips Google uses to run AI workloads).

The plan calls for prototype satellites by 2027. Each satellite in the proposed constellation would carry TPUs, draw power from solar panels, and communicate back to Earth to serve AI inference requests—running the calculations behind AI outputs like text generation, image analysis, and complex reasoning.

The economic case hinges on launch cost. Google's internal analysis puts the financial break-even point for space data centers at around $200 per kilogram to orbit. We're not there yet, but SpaceX's Falcon 9 has already driven launch costs down dramatically over the past decade, and Starship—designed for even cheaper launches—is progressing steadily.

SpaceX's Parallel Play

SpaceX isn't a passive participant in this story. The company has already filed an application with the FCC (Federal Communications Commission) to launch a constellation of up to one million satellites specifically dedicated to orbital data centers. The target altitude is between 500 and 2,000 kilometers—the same range as many existing satellite constellations.

SpaceX projects that this orbital network could eventually provide 100 gigawatts of AI compute capacity. To put that in perspective: the entire global data center industry today consumes roughly 200 gigawatts of power. SpaceX is pitching space as not just a supplement but a future peer to ground-based AI infrastructure.

The timing is also tied to SpaceX's long-anticipated IPO. The company is building investor excitement around the space data center thesis, and a high-profile partnership with Google—one of the world's largest AI compute buyers—would be a significant narrative anchor.

Why Space for AI Compute?

The question worth asking: why go to space at all? Ground-based data centers are expensive to build and harder to scale each year—driven by the AI boom's insatiable demand for power and cooling. The best locations for data centers (cheap land, cold climates, reliable power) are increasingly strained.

Space offers some genuine advantages. Solar energy is abundant and constant in orbit—no weather, no night, no grid dependency. Cooling in the cold vacuum of space is efficient. And satellite constellations can be scaled by simply launching more satellites rather than constructing new facilities on land.

The honest catch: the infrastructure required to beam compute output back to Earth at low latency is still evolving. You can run a model in orbit, but getting the result to a user in milliseconds requires high-bandwidth satellite-to-ground links. Companies like SpaceX (with Starlink) are already building this infrastructure for internet connectivity, which is exactly why a Google-SpaceX pairing makes sense—SpaceX has the launch capability and the downlink network.

What It Means for AI Infrastructure

The AI compute market is already enormous and growing fast. Microsoft, Google, Amazon, and Meta are collectively spending hundreds of billions on new data centers. The constraint isn't ambition—it's physical: finding enough land, power, water, and equipment to build at the required speed.

Space data centers don't solve that problem overnight, but they introduce a new vector for scaling. If orbital compute becomes cost-competitive with ground-based compute within the next five to seven years—a plausible scenario as launch costs fall—it changes the long-term infrastructure calculus for every major AI company.

For now, ground-based infrastructure will dominate. But the fact that Google and SpaceX are in serious discussions suggests both companies believe the timeline to viable orbital AI compute is real.

What to Watch

The immediate question is whether Google and SpaceX formalize their talks into a signed agreement. Beyond that, watch for progress on Starship's launch economics (each successful mission makes the math better), the FCC's review of SpaceX's orbital data center filing, and Google's timeline for Project Suncatcher prototype satellites in 2027.

For the broader AI industry, this partnership—if it closes—represents the clearest signal yet that the infrastructure race has a new frontier.

The Bottom Line

Google and SpaceX putting AI data centers in orbit sounds like tomorrow's news—but the conversations are happening today, the filings are in with regulators, and the prototypes are planned for 2027. The economics aren't there yet, but they're getting closer. When they arrive, AI compute will have a new address: somewhere above the atmosphere.