Musk Sets SpaceX Compute and Terafab Targets

Asked whether SpaceX had ordered the extreme ultraviolet lithography machines that advanced chipmaking depends on, Elon Musk did not say. Instead he gave analysts a timetable, and a warning to go with it. “I’ll tell you what our internal expectations are, which obviously I’d put big error bars on this because we don’t know if we will actually achieve this or not,” he said on SpaceX’s third-quarter earnings call on 20 October, according to the transcript.

Those expectations, for chips and for the data centres that would use them, are the clearest picture yet of how big Musk wants SpaceX’s AI business to get, and how fast. Every number below is a target he or his chief financial officer set out on the call, not a result. (TSN’s Musk Companies hub follows the wider story.)

Terafab’s first chip is meant to work by mid-2027

SpaceX is aiming “to make our first test chip functional in approximately May or June of next year”, Musk said. “That’s just a kind of a SRAM chip or something that I think basically tests the line. That’s at the research fab in Austin.” The goal after that is “to reach volume production around Q3 of 2028 at Terafab at Gibbons Creek, the main Terafab location”.

He immediately qualified it. “Now, these are very aggressive dates, and we’re very appreciative of ASML and all of our other suppliers in accelerating equipment deliveries,” he said, before loosening the target to volume production being “at least in the set of possible outcomes for Q3 or Q4 of 2028”. The thanks to ASML, the Dutch company that makes EUV tools, was as close as he came to Morgan Stanley analyst Adam Jonas’s question about orders and delivery dates.

Terafab is the chip complex Tesla and SpaceX say they will build and run themselves in Texas, with TSMC at most subleasing space, as Musk put it earlier this month; Intel has said it remains involved. Its stated ambition, Musk repeated, is to “make a terawatt of superintelligence chips per year”.

The data-centre plan is 13 gigawatts by the end of 2027

The nearer-term targets are about switching on bought-in chips. SpaceX is “tracking to be able to bring roughly 200,000 GB300s online per month” for the rest of this year and next, Musk said, “with next year being primarily Vera Rubin compute”, referring to Nvidia’s current and next server chips. Later in the call he put the current rate at 220,000 GB300s a month. He wants a training cluster of “440,000 Vera Rubin computers” running by the end of the second quarter of 2027.

In power terms, that adds up as follows:

Target	Musk’s words
About 3GW online by end of 2026	“roughly 3 gigawatts, maybe 2.5, 3”
10GW more during 2027	“fairly confident”, later “highly confident”
13GW or more by end of 2027	“I would be surprised if we brought on less”
Over 20GW of compute in 2027	“a non-zero possibility, maybe it’s a small possibility”

Source: SpaceX Q3 2026 earnings call transcript, 20 October 2026.

Power and cooling, not chips, are the limit, Musk said, so SpaceX will “aim for an incremental 20 gigawatts of power cooling infrastructure next year”. He was careful to add: “I’m not saying we’ll have 20 gigawatts of power.” One part of his plan is to cast turbine blades and vanes in-house using its rocket-engine foundry. Turbine suppliers “tell us that the blades and vanes are 12 to 18 months behind the rest of the turbine”, he said, which he claims will let SpaceX bring power online that much sooner than rivals. He also mentioned solar, wind and factory-built modular inference clusters.

The economics are what make the pace attractive to SpaceX. Chief financial officer Bret Johnsen said pricing for new capacity was “slightly above the high end of the $30 to $50 per watt range” discussed last quarter, and that new deployments were generating “a payback of less than a year”. Musk said the goal is to use roughly half the compute internally and rent the other half, a split he said could move to 60/40 either way. The quarter’s actual capacity and spending are in TSN’s results story.

None of these dates comes with a cost estimate or a confirmed tool order. The first one that can be checked is year-end: whether SpaceX reports something close to 3GW online when it publishes fourth-quarter results. For Terafab, the test is a working chip off the Austin line by June 2027.