Elon Musk’s next moonshot is not a car or a robot. It is a chip factory. A really, really big chip factory.
On March 14, he posted that the Terafab project would launch in seven days, pointing to Saturday, March 21. That is a date for a project launch, not a promise that finished chips will start rolling out that weekend.
So what is Terafab, why does it matter for AI, and how much can we actually know before the launch?
Designing a Chip Is Not the Same as Making One
Tesla already designs specialised AI chips. Terafab would take it further into fabrication: physically manufacturing the silicon, rather than sending a design to an outside foundry.
Those are distinct businesses. Designing the floor plan does not mean you own a concrete plant. And in this case the concrete plant needs equipment, materials and manufacturing precision that make a normal construction project look pleasantly uncomplicated.
The March 14 Reuters report traces Musk’s argument to concerns that suppliers will not provide enough chips for Tesla’s ambitions. It also recalls his suggestion of possible discussions with Intel, accompanied by the explicit point that no deal had been signed. Possibility is not partnership.
What Has Actually Been Specified?
Ahead of the launch, reports have circulated figures for a 2nm process, enormous monthly wafer capacity and a multibillion-dollar budget. They should not be presented as a settled production specification on the strength of a short social-media announcement.
The useful questions are still basic: where would the plant be, who would provide its manufacturing technology, what would it make first, and when could it produce working chips at an economic yield? A wafer-start target measures wafers entering production, not the number of good processors emerging at the other end.
One comparison can be cleared up now: Apple’s M4 is not a 2nm chip. Apple’s M4 announcement identifies second-generation 3nm technology. Node names are not interchangeable badges meaning “very small and expensive”.
AI5 Has Its Own Roadmap
Tesla’s January 28 investor update places planned AI5 production in 2027 and AI6 production in 2028. It describes them as custom inference chips for autonomy. That roadmap does not establish that AI5 will be manufactured in a new Terafab facility.
Inference means running a trained model to produce a result; training is the work of developing the model in the first place. Tesla’s document discusses its training infrastructure separately from the AI5 inference programme. A chip designed for a car’s onboard workload is therefore not automatically a substitute for every GPU in a data center.
The Apple Silicon analogy only goes so far, too. Taking control of a processor’s design and operating a leading-edge foundry are different steps. Tesla’s progress on one would not, by itself, demonstrate success at the other.
What “Launch” Could Mean on March 21
The date falls on a Saturday, and the wording leaves plenty of room. A presentation, a location announcement or a construction plan would all fit a project launch. None requires a functioning production line to appear in seven days.
It is reasonable to expect more details, but we should not pre-write them. A groundbreaking, an Intel agreement or a financing package would each need its own confirmation. The chip itself is unlikely to pop out of the ground next week, no matter how confidently someone points at the ground.
Would This Break NVIDIA’s Grip?
Potentially, owning more manufacturing capacity could give Tesla more control over supply. That is the strategic appeal. Complete self-sufficiency is a much larger claim: a fab still depends on equipment, materials and a wider industrial ecosystem.
The impact on NVIDIA would depend on which workloads Tesla’s chips serve and which purchases they replace. More specialised inference hardware does not necessarily eliminate demand for training GPUs. Nor is building a foundry the same competitive move as designing another accelerator.
If you have followed the recent wave of AI coding tools, this is a reminder that the software boom has a very physical underside. Someone still has to make the chips—and that job cannot be completed by typing “build me a semiconductor factory” into a prompt box.
The Musk Factor: Ambition Meets Manufacturing
Musk is good at making a timeline sound like a dare. Terafab deserves scrutiny on its own terms: a working design, access to a process, reliable production and the money to sustain it. Previous cars, robots or computing projects do not answer all those questions for a new fab.
The announcement is interesting because it raises the stakes of Tesla’s hardware ambitions. It does not yet settle whether the project will arrive on time, what it will cost or whether the resulting chips will be competitive.
For now, March 21 is the next date to watch. The exciting version is a serious manufacturing plan; the less exciting version is an announcement about a future announcement. Either way, the cat is watching closely. 🐱
