Google Sets Oct. 1 Launch for Its First Orbital TPU Test

Google plans to send four of its sixth-generation Trillium TPUs into orbit on Oct. 1 aboard a SpaceX Transporter-18 rideshare mission, the first orbital test of Project Suncatcher, its research effort to build AI computing infrastructure in space. iTechPost and Gizmodo both reported the Oct. 1 date; Google's own blog post on the mission says only "late September 2026."

The satellite carrying the chips was built with Planet, the Earth-imaging company, according to Gizmodo and iTechPost. It carries no customer workloads. Its only job is to report back whether the chips, the cooling system and the radiation shielding survive contact with orbit long enough to justify a second launch.

Why the chips have to prove themselves first

A Falcon 9's roughly 10-minute ascent subjects payloads to forces up to 10 times Earth's gravity, and individual TPU chips inside can see 50 to 100 times that stress, according to Google's blog post confirming the mission. The company said ground testing showed the chips withstand those loads, and the Oct. 1 flight is meant to confirm it in orbit rather than on a test stand.

Heat is the harder problem. Space has no air to carry waste heat away, so Suncatcher's satellite relies only on radiators and heat pipes, a cooling approach Google's Nov. 4, 2025 research post on the project called fundamentally different from anything used in a terrestrial data center.

The radiation numbers behind the confidence

Google said it fired a 67 MeV proton beam at a Trillium chip at UC Davis's Crocker Nuclear Laboratory while the chip ran AI workloads. A satellite shielded for a five-year mission would absorb roughly 750 rad(Si) of radiation over that time, according to the research post; Google's test chip showed no hard failures up to a dose of 15,000 rad(Si), 20 times that expected total.

1. Expected 5-year mission dose (shielded)750 rad(Si)
2. Max dose tested, no hard failures15K rad(Si)

Source: Google Research, 'Exploring a space-based, scalable AI infrastructure system design,' accessed 2026-09-25

Google has not said what the Oct. 1 mission itself cost, or what a single Suncatcher satellite would cost to build at production scale.

Google is not alone in orbit

A second Suncatcher mission, carrying two prototype satellites linked by free-space optical connections, is planned for early 2027, according to Google's research post. The company's illustrative long-term design is a cluster of 81 satellites arranged within a one-kilometer radius, sharing power and compute over laser links rated at tens of terabits per second.

Google is not the first company to try flying AI hardware. Starcloud launched an Nvidia H100 GPU into orbit in November 2025, and SpaceX is separately exploring orbital data centers, Gizmodo reported. Google has projected launch costs could fall below $200 per kilogram by the mid-2030s, a figure the company says would make space-based compute cost-competitive with data centers on the ground, though it is Google's own estimate rather than an independently verified one.

The launch lands in an active month for Google's AI infrastructure elsewhere: the company rebuilt its AX agent orchestrator around Redis instead of Kubernetes on Sept. 21, and its WeatherNext 3 model started issuing hourly forecasts on Sept. 15.

Milestone	Date
Ground radiation test, Trillium chip, UC Davis Crocker Nuclear Laboratory	Disclosed Nov. 4, 2025
First orbital test, four TPUs on one satellite with Planet	Oct. 1, 2026 (planned)
Second mission, two satellites with optical inter-satellite links	Early 2027 (planned)