TSMC has confirmed it is in talks with Elon Musk over his "Terafab" concept — and the world's largest contract chipmaker has already named its price for any deal: TSMC keeps control of the factory. That single condition, stated publicly, is the whole negotiation in miniature.

The disclosure came from TSMC leadership directly, according to DigiTimes, which reports that the company broke its silence on the matter after weeks of speculation. Musk has floated the Terafab idea — a wafer fab of unprecedented scale, implied by the name to target trillions of dollars of output or trillion-component ambition — as part of his broader push into AI and robotics hardware.

Talks are underway. That much is confirmed. Everything beyond it is a line in the sand.

Why control is the sticking point

For TSMC, the math is straightforward. The company runs the most advanced logic manufacturing on earth, and it does so from Hsinchu, not from a customer's headquarters. Apple, Nvidia, AMD — none of them tell TSMC how to run a fab. They book capacity. TSMC builds the wafers.

Musk, by contrast, has a documented history of demanding vertical integration. He owns his car plants, his battery lines, and, in the form of xAI, a growing compute estate. The Terafab concept fits that instinct: a fab so large and so dedicated that it functions as an extension of his own product roadmap.

Those two philosophies don't share a production line easily. TSMC's position, as reported, is that it will talk about scale, siting, investment and output. It will not hand over the keys.

What a Terafab would actually mean

No capacity figure, process node or site has been confirmed by either party. The project remains a concept with a name attached to it — which puts it firmly in roadmap territory, not in the capex column.

But the scale Musk gestures at is not absurd on its face. Leading-edge fab costs now run into the tens of billions of dollars per site; TSMC's Arizona complex alone carries a price tag north of $65 billion across its planned phases. A fab an order of magnitude larger would require commitments no single customer has ever made — and a level of demand concentration that TSMC has spent decades avoiding.

The quiet leverage on both sides

Musk needs leading-edge silicon for AI training and inference at volumes that strain the current supply of Nvidia-class accelerators. TSMC makes the overwhelming majority of those chips. Musk cannot build a Terafab without TSMC-class process expertise, and that expertise sits with a short list of companies — effectively one, at the leading edge.

TSMC, meanwhile, watches its customer base argue over capacity it controls. A dedicated mega-fab, funded largely by one customer, would shift construction risk off its own balance sheet. The catch is exactly the control question now on the table.

Neither side has put a number on the talks — no investment figure, no wafer target, no timeline. DigiTimes reports only that discussions are live and that TSMC's condition stands.

If Musk wants TSMC's process engineers, the negotiation ahead will decide whether Terafab becomes a funded project or stays a name.