TSMC breaks silence on Musk's Terafab: Talks underway, but fab control non-negotiable

TSMC's choice to publicly acknowledge talks rather than stay silent signals Terafab's scale has outgrown commercial confidentiality. The real question isn't whether they're talking—it's who holds the lithography recipe. Terafab almost certainly targets AI inference and data-center accelerator volume. TSMC's ironclad stance on fab control locks down three assets: EUV recipe optimization, CoWoS yield data, and the equipment-vendor calibration loop. Once the process black box becomes transparent to an end customer, node microarchitecture turns reverse-engineerable. This is sovereignty-of-technology, not a commercial term. On compliance, a US-landed line triggers a dual noose of CHIPS Act subsidies and export controls. The "non-negotiable" clause is a firewall: it blocks process know-how from leaking through a JV structure and preempts Washington from mandating a line split under a security pretext. Intel 4's yield collapse already proved that process decoupled from its original team degrades exponentially. Intel 18A and Samsung's 2nm GAA will immediately smell a de-TSMC window. TSMC's counter is compressing the 2nm ramp timeline—locking customers with schedule, not price. Over 12–24 months, the industry bifurcates: customer-owned fabs (IDM 2.0) versus TSMC-owned fabs with customer-owned IP. One burns capital; the other burns time. Capital will vote.