Google Sends TPUs to Orbit as Space Data Center Race Lifts Off
By @sharedot · · 8 pages
- Tech
- Space
- Project Suncatcher
- AI Hardware
Google's Project Suncatcher satellite carrying four AI chips launches Oct 1 on a SpaceX rocket, the first in-orbit test of space AI computing.
What happened: Google puts AI chips in orbit
On Thursday, October 1, 2026, Google will load an experimental satellite named MVP onto a SpaceX rocket and launch it as part of Project Suncatcher, the company's initiative to explore whether AI data centers can operate in space.
The refrigerator-sized satellite, built with satellite company Planet Labs, carries four of Google's Trillium Tensor Processing Units — specialized chips with roughly the computing power of one Earth-based server.
In orbit, the chips will be able to process short queries for up to 15 minutes at a time so Google's Gemini AI can respond from space, according to Mashable SEA. The satellite is expected to operate for one year before drifting for six more and burning up on reentry.
Why it matters: the energy math behind the moonshot
The appeal of orbital computing is power. Google says satellites in low Earth orbit can access near-constant sunlight and generate up to eight times more solar power than equivalent panels on the ground — a tantalizing answer to AI's exploding electricity demands.
Elon Musk has pushed the same vision, saying at the time SpaceX acquired xAI that "global electricity demand for AI simply cannot be met with terrestrial solutions," and predicting a working space data center could exist as soon as 2027.
The National reports Nvidia announced its Space-1 Vera Rubin Module in March to help bring AI compute to orbital data centers, suggesting industry-wide momentum behind the idea.
The evidence: chips survived Earth-bound torture tests
Google ran extensive ground testing before committing hardware to orbit, Network World reports. Engineers shook the satellite on all three axes to simulate launch vibration and acceleration of up to 10 times gravity, with individual chips facing up to 100 times gravity.
Trillium TPUs were also blasted with a proton beam at UC Davis's Crocker Nuclear Laboratory while running AI workloads, and Google says they tolerated a total ionizing radiation dose greater than a five-year mission would deliver.
A cooling system combining heat pipes and radiators was validated in a thermal vacuum chamber. But as Google's Travis Beals put it, "some things can only be tested in space."
The hard part: cooling chips in a vacuum
Cooling is the central engineering challenge. TPUs generate enormous heat in a small area, but in space there is no airflow — fans are useless in a vacuum, so heat can only be diffused through radiators.
That constraint is why the MVP satellite's chips must shut down after 15 minutes of processing queries to cool off before running again, Mashable SEA explains. Skeptics remain unconvinced the approach scales.
Ariel Ekblaw of the Aurelia Institute told The National that space is far colder than people realize, making it a poor home for "a really hot little volcano of a bunch of computers," and noted that swapping upgraded chips every six months would require launching all the way to orbit.
The stakes: a crowded orbital data center race
Google is not alone. Mashable notes the project literally realizes Musk's dream of space-based AI data centers — with Google's hardware riding a Musk rocket.
The National reports SpaceX, Nvidia, Starcloud and Blue Origin all have stated ambitions for orbital computing infrastructure.
The timing is notable on Earth, too: Entrepreneur reports a UMass Amherst poll found only 1 in 9 Americans support a data center in their community, giving space a certain NIMBY-proof appeal.
Separately, Brave New Coin connects the launch to Pope Leo XIV's warnings in France that a "paradise of machines" could erode human agency, framing the debate over how this expanding compute capacity should be governed.
What comes next: laser links in 2027
Nobody should expect an orbital data center soon — experts told Entrepreneur such systems remain years from commercial viability, and Google itself frames Suncatcher as long-term research.
The next milestone arrives in 2027, when Google plans two more satellites to test high-bandwidth laser links between computing clusters, a precision challenge Tomorrow's World Today compares to hitting a coin-sized target from miles away while both points are moving.
The MVP mission's real job is generating failure data: what works, what breaks, and whether radiation, thermal extremes and launch forces leave the hardware viable. If the answers hold, constellations of solar-powered satellites could one day manage AI workloads far above the nearest NIMBY.
Sources
- sea.mashable.com › Liftoff: Google is realizing Elon Musk's dream of AI data centers in space
- computerworld.com › Google is set to launch a small AI data center into space
- thenationalnews.com › Google wants AI data centres in space, but experts warn of obstacles
- networkworld.com › Google's first prototype satellite is going up
- tomorrowsworldtoday.com › Google is Sending Chips into Orbit for Space Data Centers Testing
- entrepreneur.com › Google's Next AI Data Center Might Be in Outer Space
- bravenewcoin.com › Google's Space AI Data Centres Raise New Questions as Pope Leo Warns of a Threat to Humanity