Tesla is building toward its own "Terafab" — and, according to a DigiTimes report, that plan is exposing growing tensions with the two foundries it still depends on: Samsung and TSMC.

The report, surfaced by DigiTimes this week, frames the conflict plainly. Tesla wants in-house chipmaking capacity at terafab scale. Samsung Foundry and TSMC want Tesla's money on their advanced nodes. Those two goals are starting to collide.

The tensions aren't loud. They're structural.

What is Tesla actually doing?

Tesla has spent years designing its own silicon rather than buying merchant chips. Its in-house design teams already produce compute for vehicles and, more demandingly, for the AI training clusters behind its autonomy program. That workload is exactly the kind that has pushed other hyperscalers — Amazon, Google, Microsoft — toward custom accelerators and, in some cases, toward funding their own fabrication paths.

The "Terafab" concept, as DigiTimes describes it, pushes further: Tesla aims at fabrication capacity sized in terascale terms, not merely at custom design. Details in the report remain thin. Which node, which wafer size, which geography, and which timeline — none of that is confirmed, and DigiTimes doesn't pin those down.

What is clear is the friction. A customer that announces plans to make its own chips is a customer its foundries must treat differently.

Where do Samsung and TSMC fit?

Both foundries have courted Tesla. Samsung Foundry, in particular, has worked to land high-profile AI and automotive customers at its advanced nodes as it competes for a stronger share of a market TSMC dominates. TSMC, for its part, remains the default destination for leading-edge logic.

Tesla's in-house ambitions complicate both relationships, per the DigiTimes account. A foundry investing in capacity for a customer must weigh how long that customer stays a customer. A customer building its own fab must decide how much leading-edge volume it can still commit externally while its own capacity ramps — if it ramps.

That bargaining dynamic is the real story the report surfaces. It's not a contract dispute on the record. It's the quiet repricing of a relationship once one side starts building its own supply.

Why would an automaker want a fab?

The logic Tesla shares with the hyperscalers is supply security and margin. AI training silicon is scarce, expensive, and allocated by a handful of suppliers. Owning capacity — or the prospect of owning it — changes how a buyer negotiates for the capacity it still needs.

There's also the strategic argument. Chips are now the differentiating component in both vehicles and autonomy systems. Companies that design them increasingly ask whether fabricating them is the next vertical step, even though leading-edge fabs cost tens of billions of dollars and take years to yield.

Whether Tesla can execute at the terafab scale its label implies is unproven. No other automaker has. Even among the cloud giants, only some have gone beyond co-investment into full ownership of leading-edge capacity.

What happens next?

For now, Tesla still needs Samsung and TSMC, and they still want Tesla's volume. DigiTimes reports the relationship as strained, not broken — a customer hedging its suppliers while the suppliers hedge a customer that might become a competitor.

Watch for the concrete signals: external wafer commitments Tesla signs — or doesn't — over the coming quarters, and any disclosure of site, scale, or timing for the Terafab itself. Until those numbers land, the story stays where DigiTimes left it: an EV maker whose chip ambitions are making its foundry partners nervous.