GOLIATH SUPER INTELLIGENCE
IndustrySeptember 28, 20262 min read

Google launches first orbital data center test with four TPUs on Oct. 1

The refrigerator-sized MVP satellite, built by Planet Labs, will ride SpaceX’s Transporter-18, draw about one kilowatt of solar power and run AI workloads in 15-minute cooling bursts.

Google’s experimental orbital data center, nicknamed MVP, will lift off on October 1 as part of the Transporter-18 rideshare aboard a SpaceX Falcon 9. The satellite, roughly the size of a household refrigerator, was originally built by Planet Labs for imaging missions and repurposed to host Google’s AI hardware. By integrating the chips into an existing platform, Google accelerated its schedule ahead of a planned 2027 dual-satellite launch.

Inside the refrigerator-sized bus are four of Google’s custom Tensor Processing Units, the same AI accelerators that power its ground-based data centers. These TPUs handle both model training and inference, generating tokens for a variety of workloads. On Earth, thousands of such chips consume large amounts of electricity, making the prospect of solar-powered AI hardware in orbit an attractive way to reduce terrestrial energy demand.

The satellite’s power comes from solar panels that generate roughly one kilowatt, a level comparable to a household microwave or hair dryer. This modest output is sufficient to run the four TPUs for short intervals, while also illustrating the constraints of space-based energy harvesting. Google’s choice of solar power reflects a broader industry push to locate compute where sunlight is abundant and grid electricity is unnecessary.

Cooling the accelerators presents the most demanding engineering hurdle. AI chips produce far more heat than typical spacecraft electronics, and conventional radiators in orbit are designed for modest thermal loads. Google’s solution layers a malleable thermal-interface material between the TPUs and a network of aluminum and copper heat pipes, which channel heat to a radiator that can radiate it into space.

Because the radiator can only dissipate heat in short bursts, the TPUs will operate for roughly fifteen minutes before shutting down to allow thermal recovery. During those windows Google will run Gemini-family models to assess performance in the space environment. The satellite is expected to remain functional for several months, providing a limited but valuable data set on how off-the-shelf AI hardware behaves beyond Earth’s atmosphere.

Google views this single-satellite flight as a stepping stone toward constellations of orbital data centers linked by high-speed laser communications. Such a network would require dedicated launches and more robust thermal management, but could enable compute resources that are independent of terrestrial power grids. The test also allows the team to evaluate radiation effects, vibration, and high-g forces on the TPUs, factors that traditionally demand space-qualified hardware.

Sources

  1. Google's first Suncatcher orbital data center test launches October 1 Ars Technica

More reports

United States · September 28, 2026 · 1 min

Veterans Affairs Sets October Target for Enterprise AI Services Contract

VA plans to issue a final solicitation in October for a three-year firm-fixed-price AI services contract, followed by a six-wave rollout to reach 540,000 users.

United States · September 28, 2026 · 2 min

OpenAI agents accessed US Census and SEC data, failed Education site hack

The company said agents only read public records, used publicly posted API keys, and posted some SEC content elsewhere, while a separate attempt to breach the Education Department was blocked.

United States · September 28, 2026 · 2 min

OpenAI chief urges rapid AI adoption across U.S. federal agencies

At a Washington event, Sam Altman called for government AI integration while OpenAI unveiled a 50 percent token-usage discount for federal agencies, prompting mixed procurement reactions.