SpaceX and Google launch AI chips into orbit for space data centers
The AI infrastructure shift you didn't see coming
For years, the AI industry has operated on an unspoken assumption: data centers need land, power grids, cooling systems, and physical security on Earth. The more computing we need, the more warehouses we build, and the more local communities push back.
Every few months, a story quietly rewrites an assumption everyone had stopped questioning.
On Thursday, SpaceX launched Google's tensor processing units into orbit aboard a Falcon 9 rocket from Vandenberg Space Force Base. Not as a science experiment. As a deliberate step toward making AI data centers in space commercially viable.
The launch at 11:15 am PT carried Planet Labs satellites equipped with Google's homegrown TPUs, solar-powered prototypes that represent the first real-world test of Alphabet's Project Suncatcher, a moonshot the company first revealed in November 2025 to build "reliable, solar-powered AI computing infrastructure that can operate continuously in space."
Why space makes sense for AI compute
The logic is surprisingly practical. In low Earth orbit, satellites can access near-constant sunlight. Alphabet says this generates up to eight times more solar power than what's available on Earth. Eventually, the company expects to link multiple constellations of satellites together, enabling them to manage larger AI workloads while in orbit.
SpaceX has been promoting orbital supercomputers as the natural next frontier for AI infrastructure. The company frames it as "infinite real estate, a way to escape the growing backlash against terrestrial data centers that consume enormous amounts of energy, water, and land in communities that increasingly don't want them.
"We could see orbital data centers be the cheapest way to train AI, and that will be true within two years, maybe three at the latest," Elon Musk said earlier this year.
The $82 billion connection
Alphabet holds a stake in SpaceX currently valued at over $82 billion, a position that became liquid when SpaceX went public in June in a record-breaking IPO. The two companies compete in AI (Google through DeepMind and Gemini, SpaceXAI through Grok) but remain close infrastructure partners.
SpaceX COO Gwynne Shotwell confirmed at a September event that the company plans to deploy "supercompute in space" in 2027. SpaceX is also developing its own orbital data centers in Redmond, Washington, using swarms of satellites equipped with GPUs and solar arrays produced in partnership with Tesla.
The reality check
Industry experts caution that orbital AI computing faces significant hurdles. Rocket launches remain capacity-constrained and expensive. Chips in space must withstand extreme temperatures and radiation that no terrestrial data center ever encounters. Cooling systems and orbital debris add another layer of complexity.
Google has tested its TPUs running AI workloads in a controlled facility at the University of California at Davis, but nobody yet knows how the chips will perform in the actual conditions of low Earth orbit. Thursday's launch is designed to answer that question.
The Transporter-18 mission also marked a second potentially historic launch for SpaceX on the same day: the company is also slated to carry astronauts to the International Space Station from Florida's Space Coast, beginning a six-month NASA mission.
What it means
The story here is not that a rocket carried computer chips. The story is that two of the most valuable companies on Earth are betting that the future of AI compute happens above our heads, not on the ground. If Project Suncatcher works, even partially, it reshapes every calculation the AI industry makes about where to build, how much energy to use, and who gets to decide.
The CUDA moat was supposed to be unbreakable. Then came space-based AI compute.
Sources
- CNBC: "SpaceX set to launch Google AI chips into orbit in push toward space-based data centers, Lora Kolodny, Oct 1, 2026
- Alphabet blog post on Project Suncatcher (via CNBC)