Google is testing AI chips for possible data centers in space, according to a report from KGW. The effort, still at an experimental stage, signals that the company is investigating whether compute infrastructure for artificial intelligence workloads can operate outside terrestrial data centers entirely.

The report offers few engineering details. It does not specify which chips Google is testing, whether they belong to its Tensor Processing Unit (TPU) family, or whether third-party silicon is involved. It also leaves open whether the tests involve hardware already flown on orbital platforms or ground-based qualification for future deployment.

The distinction matters. Radiation-hardened or radiation-tolerant chip designs differ materially from standard data-center parts, and space deployment imposes constraints on power delivery, thermal management and interconnects that shape any commercial case. Until Google identifies the silicon, the program sits firmly in the research-and-testing category rather than on a deployment roadmap.

KGW characterizes the initiative as testing aimed at "possible data centers in space" — language that stops well short of confirmed plans. No launch schedule, orbital facility partner, capacity target or budget figure appears in the report.

The commercial logic behind orbital compute is straightforward even if the economics remain unproven. AI training and inference demand is straining terrestrial power grids, and space offers essentially unlimited solar generation alongside radiative cooling. The obstacles are launch cost, hardware longevity under radiation exposure, and the difficulty of servicing or replacing failed components in orbit.

Google's interest places it among a small set of companies exploring whether those obstacles can be overcome at scale. The report does not say whether Google is coordinating with launch providers, satellite operators or government agencies on the tests.

For the semiconductor supply chain, the development is notable mostly as a signal of demand-side experimentation. If orbital data centers move from concept to procurement, they would open a new buyer category for processors, power electronics and memory qualified for space environments — a market currently dominated by specialized, low-volume suppliers rather than hyperscale chip programs.

That remains speculative. What is confirmed today is narrower: Google is testing AI chips with an eye toward data centers in space, per KGW, and has disclosed no timeline, partner or hardware specification. Whether the tests graduate into an orbital deployment program will depend on results the company has not yet made public.