Big Tech companies are building their own AI chips to meet AI compute demand: Google, Amazon, Meta, Microsoft and Apple all have custom silicon.
Elon Musk’s xAI, SpaceX and Tesla just raised the stakes. This past weekend, Musk announced the formation of Terafab, a chipmaking facility in Austin, Tex., during a livestream on X.
Billed by Tesla as the “largest chip manufacturing facility ever,” Terafab will combine logic, memory, packaging and testing in one place. Musk said such an integrated plant “doesn’t exist anywhere in the world” as far as he knows.
Other Big Tech companies design their chips but don’t physically make them; Terafab will both design and manufacture chips. The plant will make two types of chips – one for AI inference and edge computing for vehicles and robots and the other designed for space.
Musk said 20 gigawatts per year is the current output of AI compute. Expected compute demand from Tesla and SpaceX is 1 terawatt (or 1,000 gigawatts) per year for his vehicles, rockets and robots.
Musk said he is willing to “push the limits of physics in compute and do some wild and crazy things.”
He told his current suppliers including Samsung, Micron and TSMC that he will buy all of their chips. But their manufacturing expansion plans are “much less than we would like,” he said. That’s why “we either build the Terafab or we won’t have the chips.”
Demand for AI compute has created a frenzy of data center construction and propelled AI chipmaker Nvidia into becoming the most valuable company in the world. To lessen their reliance on Nvidia, which has the lion’s share of the AI chip market, Big Tech has been ramping up its own chipmaking efforts to meet demand.
Musk couches his chipmaking venture as one that was necessary for his vision to make humanity not just multi-planetary but a space-faring society.
Invoking Star Trek, Asimov and other science-fiction references, Musk said he envisions galactic travel for humanity one day, and “turn science fiction to science fact.”