Elon Musk has confirmed he is in discussions with TSMC about a chipmaking collaboration under the Terafab concept, making the world's largest contract foundry the second company publicly named in the plan after Intel.

Musk disclosed the talks himself, according to Tom's Hardware. The statement significantly expands the commercial footprint of the Terafab idea, which until now had Intel as its only named partner.

The scope is unusually narrow and unusually broad at the same time. Under the arrangement as described, a Terafab would supply chips exclusively to three Musk-controlled companies: Tesla, the electric vehicle maker; SpaceX, the launch and satellite operator; and xAI, Musk's artificial intelligence venture. That exclusivity clause is the structural core of the proposal. It converts a foundry relationship into a dedicated-capacity arrangement, closer in spirit to a captive fab than to standard merchant foundry contracting.

For TSMC, the calculus is delicate. The company dominates advanced logic manufacturing and counts Apple, Nvidia, AMD and Qualcomm among its largest customers. Committing dedicated leading-edge capacity to three customers — however large their combined demand — would sit uneasily beside its merchant foundry model, which allocates wafers across many buyers. TSMC has not publicly detailed terms, capacity volumes, process nodes or timelines for any Terafab arrangement, and Musk's confirmation covers only that discussions are underway, not that a deal is signed.

Intel's position in the picture has not changed. It remains the only other company Musk has named as a Terafab partner. That pairing carried its own logic: Intel Foundry Services needs anchor customers to fill its fab buildout, and a multi-year dedicated-supply commitment from Musk's companies would be exactly the kind of order book Intel has chased since launching its foundry business. Adding TSMC to the conversation gives Musk leverage on both price and security of supply, pitting the two leading-edge manufacturers against each other for a customer whose appetite spans automotive silicon, spacecraft-grade chips and AI training accelerators.

The demand profile of the three named customers is heterogeneous but complementary. Tesla needs high-volume automotive-grade processors for vehicle control and its Dojo-class training infrastructure. SpaceX requires radiation-tolerant components for satellites and launch systems at scale, exemplified by the Starlink constellation. xAI consumes the same class of AI accelerators as its rivals — large, power-hungry devices typically built on the most advanced nodes TSMC offers. No specific node, wafer size or capacity figure has been disclosed for the Terafab concept by any party.

Geopolitically, the talks land at a moment when both Washington and Taipei are reassessing how advanced capacity is allocated and located. A dedicated-supply arrangement spanning US and Taiwan manufacturing would touch exactly the questions regulators on both sides have raised about chokepoints, though no government involvement in the discussions has been reported.

Whether TSMC accepts exclusivity is the open question. Its foundry model has historically rejected captive arrangements, and its existing customers would scrutinize any capacity carved out for a single corporate family. If the talks progress, expect the first hard numbers — wafer starts per month, node commitments, contract duration — to surface in TSMC's earnings calls rather than in Musk's posts.