Google Project Suncatcher: First Google TPU in Orbit
On October 1, 2026, a Planet Labs-built satellite carried a Google TPU into orbit on a SpaceX rideshare — Project Suncatcher. Why the chip only runs in 15-minute bursts, the 2027 two-sat laser demo, the 81-satellite endgame, and Google's Starship cost-curve framing (~1,800 launches), with TechCrunch linked throughout.
Google puts a TPU in orbit
On October 1, 2026, Google's prototype orbital compute satellite lifted off from California on a SpaceX rocket — the first time the company has sent one of its advanced chips into space, according to TechCrunch. The satellite was built by Planet Labs. The program is Project Suncatcher, led by Google's Travis Beals.
This page is the long-form briefing behind the BSH Technologies Instagram carousel.
What the prototype has to prove
The mission is meant to show that a Google Tensor Processing Unit (TPU) can function in space: supply about a kilowatt of continuous power, cool the chip, and run models to see what breaks. Beals told TechCrunch that ground testing helps, but nothing replaces the real environment.
- Builder: Planet Labs (standard platform for this first flight).
- Chip: Google TPU — Google's accelerator line competing with Nvidia GPUs for AI workloads.
- Launch: SpaceX rideshare from California with more than 100 payloads on the same rocket (including other space-AI missions).
- Program lead: Travis Beals, Project Suncatcher.
Why only 15-minute bursts
Once commissioned, the satellite will fire up its TPU in 15-minute bursts so it does not strain the satellite's power and thermal management systems. That constraint is hardware reality on a small sat, not a marketing detail — continuous high-power AI silicon in orbit needs both watts and heat rejection.
2027: two satellites, one laser link
Planet Labs and Google are working on a follow-on demo expected to fly next year (2027): two satellites more purpose-built for advanced compute that can run heavier workloads and attempt to collaborate via a laser communications link.
The 81-satellite moonshot
Beals frames Suncatcher as a long-term moonshot: an orbital data center as a network of 81 satellites flying in close formation, processing in parallel. Bandwidth and latency between TPUs "really, really matters" when you try to run a multi-rack workload — the team is looking ahead to workloads five years out, not only today's jobs.
Starship math (and a corrected headline)
TechCrunch reports that Google also released a peer-reviewed version of its orbital data centers white paper (to appear in Joule). The analysis is not pitched as a full economic feasibility study, but it sketches how launch prices might fall. Google looks to SpaceX (and is a major SpaceX investor). Researchers model a ~20% per year learning curve and launch prices near $200/kg by 2035, which — on their payload assumptions — implies Starship flying on the order of ~1,800 launches over about 10 years (~180/year if each flight delivers ~200 metric tons). TechCrunch notes a correction: the story headline originally said 1,600; the corrected framing is 1,800.
Radiation and what the chips can do
Updated radiation testing left Google still confident its chips can handle large inference workloads in orbit across a typical five-year satellite life (error rates described as very low for typical inference). Mega-scale training runs with thousands of chips for months remain the harder case.
Why it matters
- Hardware, not a deck. A Google TPU is physically on orbit — the first concrete step of Suncatcher.
- Launch cadence is the bottleneck. The same companies that ship phone-side AI are modeling Starship flight rates to make orbital compute affordable.
- Networking is the product. The 2027 laser link and the 81-sat formation are about treating space like a multi-rack cluster, not a single board.
Primary sources
- TechCrunch — Google thinks SpaceX's Starship has to launch ~1,800 times before space data centers get off the ground (Oct 1, 2026; headline corrected from 1,600)
How BSH can help
At BSH Technologies we help teams read frontier infrastructure bets — orbital compute, on-device AI, and the launch/cost curves behind them — and decide what actually belongs in a product roadmap. If you are tracking Project Suncatcher or planning AI systems that assume new power, cooling, or networking constraints, our Thrissur engineers can turn the briefing into an architecture checklist.
Frequently asked questions
What is Google Project Suncatcher?
Project Suncatcher is Google's program to develop large-scale compute clusters in Earth orbit. On October 1, 2026, a Planet Labs-built prototype carrying a Google TPU launched on a SpaceX rocket from California — the first time Google sent one of its advanced chips into space, per TechCrunch.
Why does the orbital TPU only run in 15-minute bursts?
Once commissioned, the satellite runs the TPU in 15-minute bursts to avoid straining power and thermal management. The payload is designed around about a kilowatt of continuous power.
Is the Starship figure 1,600 or 1,800 launches?
TechCrunch's article correction states the headline originally said 1,600; the corrected framing from Google's orbital compute analysis is about 1,800 Starship launches over roughly 10 years under the paper's payload assumptions.
From the blog
View all posts
Gemini 4 Argon: Google’s Frontier Model Goes to Cyber Defenders First
On September 30, 2026, Google announced Gemini 4 Argon — rolling out first to trusted cyber defenders via the Fairwind Program. Intro pricing, internal results, Ars Technica’s DeepSWE and Wiz notes, and why consumers wait, with Google and Ars linked throughout.
OpenAI Dots: Always-On AI Agents on GPT-6 Astra (DevDay 2026)
At DevDay on September 29, 2026, OpenAI introduced Dots — hosted, always-on agents powered by GPT-6 Astra, each with its own cloud computer and ~4,000 app connectors. What a Dot does, who gets it today, the usage fine print, and the new Pro tiers, with OpenAI and The Register linked throughout.