SpaceX is preparing to send Google's AI chips into space to power data centers in orbit, The News International reports, moving one of the semiconductor industry's most sought-after accelerator families off the planet for the first time.

The plan, as described in the report, calls for Google AI chips to fly on SpaceX launches and operate as part of data center infrastructure in space. The pairing pairs two companies that already sit at the center of the AI supply chain: Google, which designs its tensor processing units in-house and buys heavily from Nvidia, and SpaceX, whose Falcon 9 and Starship launch cadence has driven down per-kilogram launch costs.

That cost curve matters commercially. Orbital data centers have long been dismissed as uneconomical because launching racks of servers cost more than building them on the ground. Reusable rockets change the arithmetic. If SpaceX can lift compute hardware at the prices it advertises internally, cooling — which consumes a large share of terrestrial data center power — becomes easier in the vacuum of space, where radiators can reject heat continuously.

The reported deal also fits each company's competitive position. Google is racing Amazon, Microsoft and Meta to secure AI compute, and an orbital option would add capacity outside the constraints of grid interconnects and land permitting that currently bottleneck new US data center construction. For SpaceX, carrying third-party AI hardware would create a launch customer category beyond its own Starlink constellation, filling manifest capacity on flights that already exist.

The report does not specify which Google chip family would fly, how many units are involved, or a launch date. Google's custom accelerator line spans multiple TPU generations, and the company has never disclosed shipment volumes for any of them, so any capacity figure for an orbital deployment would be an estimate rather than a confirmed number.

Power is the other open question. Terrestrial AI accelerators draw hundreds of watts each, and a rack-scale orbital deployment would need a generation source — most plausibly solar panels paired with batteries — sized to the compute load. Neither company has publicly detailed an architecture for this, according to the report.

The commercial logic nonetheless tracks a broader shift. AI infrastructure spending has pushed cloud providers toward unconventional siting, from nuclear-adjacent campuses to undersea cables repurposed for power delivery. Orbital compute is the most extreme version of that search, and SpaceX is the only launch operator with a reusable fleet flying at the frequency such a buildout would demand.

For the semiconductor supply chain, the development signals that demand for AI silicon now extends beyond terrestrial constraints. If the deployment proceeds, packaging and qualification standards for accelerators would need to survive launch vibration and radiation exposure — requirements that today's data center parts are not built to meet.

Neither SpaceX nor Google has confirmed chip counts, revenue terms or a timeline, and the report attributes the plan to sources familiar with the effort rather than company statements. Whether orbital data centers remain a demonstration or become a meaningful share of AI compute will depend on launch costs staying on their current downward path and on radiation-hardened versions of commercial accelerators proving economical at scale.