Google Suncatcher: orbital TPU runs, but space may cost 100x more
Key takeaways
- Google's energy chief says orbital data centres could cost up to 100 times more than terrestrial ones
- The Suncatcher MVP has four TPUs but only about 1kW of solar power, so it runs one chip at a time for roughly 20 minutes
- Corio says permitting, not power, is the real bottleneck on Earth, so IT teams should plan around grid and planning constraints
- SpaceX's own IPO paperwork says its orbital data centres might never become commercially viable
Google's first Project Suncatcher satellite has run an AI model on a TPU in orbit, but the executive who powers Google's ground-based data centres says the economics of doing this at scale are brutal. Amanda Peterson Corio, Google's global head of data centre energy, told The Information's WTF Summit in Napa Valley that space data centres could cost up to 100 times as much as terrestrial ones. Her advice: "Let's explore all the terrestrial options before we go extraterrestrial."
This assessment looks at what the prototype does, what it cannot yet do, and how it stacks up against rivals.
What it does well
The Suncatcher MVP, built with satellite operator Planet, launched on SpaceX's Transporter-18 rideshare mission on 1 October. It carries four TPUs (Tensor Processing Units, Google's custom AI accelerator chips) powered by solar panels.
After about three weeks testing the satellite's subsystems, Google ran its "Hello World" milestone. An onboard Gemma 4 model, running on a TPU, wrote a haiku about being in space, and the satellite downlinked it as it passed over Svalbard, Norway. The poem read "Blue marble below / Data streams in silent grace / Starlight feeds my core."
The haiku is trivial, but a data centre-class chip in orbit ran a model and returned a result to Earth. Google bills it as the first AI response from a data centre-class TPU in orbit. That proves the basic loop works, which is the first thing any orbital compute plan has to show.
Where it falls short
Scale is the problem. The satellite generates only about 1kW of solar power, so it can run just one of its four chips at a time, for roughly 20 minutes per session, according to The Next Web. A single 1 GW AI campus, the kind now under construction on the ground, draws about a million times that much power.
Then there is cost. Corio's 100x ceiling sits alongside a sober timeline from Travis Beals, the Suncatcher project lead, who told NPR on launch day that he did not expect orbital compute to become cheaper than the terrestrial kind within the next five years. Google has called Suncatcher a "moonshot" since unveiling it a year ago.
Corio also pointed at where the real constraint sits on Earth: permitting rather than power. That is the same politics that backers of orbital compute say they can escape. If the bottleneck is planning approval and grid connection, not electrons, then moving hardware to orbit solves a problem that may not be the binding one.
The claim of a first is also narrow. Starcloud put an Nvidia H100 into orbit a year ago and was running AI models on it within weeks.
How it compares
| Approach | Hardware | Status per the reporting |
|---|---|---|
| Google Suncatcher MVP | Four TPUs, about 1kW solar | Launched 1 October; ran Gemma 4 haiku; one chip at a time for about 20 minutes |
| Starcloud | Nvidia H100 | In orbit a year ago; running AI models within weeks |
| SpaceX orbital compute | Not specified | FCC request for up to a million data-centre satellites; own IPO paperwork warns it may never be commercially viable |
| Terrestrial 1 GW campus | Conventional | Under construction now; limited by permitting, per Corio |
Nvidia has lined up a Space-1 Vera Rubin module aimed at orbital data centres, while OpenAI's Sam Altman called the idea "ridiculous" under current conditions earlier this year.
Pricing and value
No customer price exists, but the investor arithmetic is revealing. Short seller Jim Chanos read Corio's statement as $5 trillion per gigawatt of compute, saying the gap between her view and the estimates of SpaceX management and bulls "should give pause to anyone trying to model this business out."
SpaceX's CFO told investors last month that new compute pays for itself in under a year at current rental rates. TIKR calculated that if orbital capacity costs up to 100 times as much and rents at the same price, payback stretches to as long as a century, or about a decade even at a tenth of Corio's ceiling. Orbital compute underpins part of SpaceX's roughly $1.8 trillion market value, and Musk has posted that monthly orbital deployments will probably overtake terrestrial compute around the end of 2028.
There is a conflict of interest worth noting. Alphabet held a 6.11% SpaceX stake at the end of 2025, according to Bloomberg, and Google has reportedly discussed launching future Suncatcher hardware with SpaceX. Its energy chief has potentially poured cold water on a thesis that props up an asset Alphabet owns.
Who should use it
Nobody can buy Suncatcher capacity, so this is a watching brief. Infrastructure planners should take Corio's view as the practical one: the near-term job is winning planning permission and grid connections for ground sites. In the US, the picture is already tough; utilities are making AI data centres prove they exist before granting power connections, with 50% of planned buildouts delayed, stalled or cancelled.
Investors in SpaceX or Alphabet should model orbital revenue with Corio's cost ceiling and SpaceX's own risk language in mind, not the bull case alone.
Verdict
Suncatcher is a credible engineering demonstration and a weak business case today. Treat it as research, not a capacity option. IT and infrastructure teams should keep investing in terrestrial planning, power and permitting, and watch for the next Suncatcher hardware, SpaceX's FCC decision and any independent cost data. Our analysis of Nvidia's chip shipment commitments shows how much ground-side constraints still shape AI supply, and SpaceX's next Starship flight is the launch-cost story that Suncatcher's economics depend on.
Frequently asked questions
What is Google's Project Suncatcher?
It is Google's moonshot to test TPU AI chips in orbit, with the long-term aim of running gigawatts of compute in space. The first prototype, MVP, launched on 1 October with four TPUs.
How much more would a space data centre cost?
Amanda Peterson Corio, Google's global head of data centre energy, said up to 100 times as much as one built on the ground.