Elon Musk announced plans to build Terafab, the largest semiconductor manufacturing facility ever.
A joint venture between Tesla, SpaceX and xAI, Terafab was announced via a livestream on Saturday as “the next step towards becoming a galactic civilization.” Yes, Musk’s plans for this one extend beyond Earth, and it will cost $20 to $25 billion to build.
This Tweet is currently unavailable. It might be loading or has been removed.
Initially, the Terafab will be a massive chip manufacturing plant located next to Tesla’s Giga Texas in Austin. Musk said they will be able to make a chip, test it, improve it, and keep iterating like that in a single building, something which “does not exist anywhere else in the world.” There’s lot of ramping up to do, but the goal (per Musk’s comments at the Tesla shareholder meeting in November last year) is for Terafab to eventually be capable of producing 1 million chip wafers per month using a 2-nanometer process tech.
For reference, the largest semiconductor manufacturing company in the world right now is Taiwan’s TSMC, and that company plans to reach a monthly output of 140,000 2-nanometer chip wafers per month by the end of 2026.
In terms of actual chips, Musk expects Terafab to produce between 100 and 200 billion AI and memory chips per year, which will mostly be used by Tesla itself, powering its cars and robots. According to Musk, Tesla needs the chips, and its current suppliers, including Samsung, TSMC, Micron, and others, have “a maximum rate at which they’re comfortable expanding.”
Elon Musk’s X teases new dislike button on replies
“That rate is much less than we would like. And so, we either build the Terafab, or we don’t have the chips,” he said during the presentation.
Mashable Light Speed
It’s a lofty goal for a company (well, three companies) that never manufactured semiconductors before, but it gets even more ambitious.
At full capacity, it will produce an annual 100 to 200 gigawatts of computing power on Earth, as well as 1 terawatt of computing power in Space (this means that the chips produced at Terafab will ultimately draw that much power when deployed).
For reference, Musk said that the U.S. electricity demand is 0.5 terawatts. This is why, Musk argued, 80% of Terafab’s compute output will actually reside in space, on SpaceX’s solar-powered AI satellites. Solar irradiance (the amount of power received from the Sun per unit area) is 5x higher in space than on Earth, while the heat rejection in the vacuum in space makes it easier to cool all those chips.
Again, for reference, the amount of compute running in Earth’s low orbit right now is negligible, and is mainly limited to onboard processing on satellites.
What Musk is describing does exist, with China’s “Three-Body Computing Constellation” being the most prominent example. It’s a 12-satellite constellation running AI models via chips capable of total computing capacity of 5 peta operations per second (POPS), with plans to expand to 2,800 satellites capable of 1,000 POPS. While I couldn’t find the total power consumption for that particular project, it is likely in the kilowatts – orders of magnitude lower than what Musk is planning.
Even for Musk’s standards, these goals are incredibly ambitious. TSMC is the primary reason why there’s so much political strife around Taiwan; if it were easy to build advanced foundries that produce chip wafers at the scale TSMC does, the U.S. wouldn’t be so worried about China controlling Taiwan. Building a massive data center in space is an equally complex problem, though here SpaceX is uniquely positioned as the most advanced aerospace company in the world.
It doesn’t end there, as Musk ultimately plans to one day build an “incredibly epic” “mass driver” on the moon which will push compute output from terawatts to petawatts, though details on how this happens are incredibly vague at this point.
Elon Musk found liable for defrauding Twitter investors
Musk’s timeline for all and any this is unclear. The company’s next-gen AI chip, the AI5, should reach volume production in 2027, but that certainly won’t be built at the Terafab, which will likely take years to build. For a reality check, it’s worth recalling Musk’s ambitious goals promises at Tesla Battery Day in 2020, almost none of which are true now, long after the original timelines have passed.
