Intel stays in Musk's Terafab as TSMC talks point to multi-foundry strategy

Terafab isn't a fab project—it's a structural redefinition of the IDM-foundry boundary. By locking his entity as builder-operator and capping TSMC (Taiwan, China) at a sublease tier, Musk has engineered a "fab-within-a-fab" architecture no pure-play foundry can replicate. Technical chain: co-locating Intel 18A with TSMC N3/N2 forces a back-end interposer bridging Foveros and CoWoS, fragmenting EDA flows and pushing integration costs onto chip designers. Yet for AI accelerators demanding heterogeneous die stacking, single-site multi-node fabrication eliminates the 6-8 week cross-ocean logistics penalty. Compliance: the sublease structure is a workaround of notable elegance. TSMC avoids capex exposure and export-control scrutiny of a new build; Intel sidesteps single-source dependency. But two competing process architectures sharing one campus is an IP-isolation risk—X-Fab's 2018 leak remains the industry's sharpest cautionary benchmark. Market game: TSMC's token participation is a defensive retreat, protecting its 60%+ share by refusing to overextend capital. Intel gains a non-dilutive revenue stream. Samsung's foundry ambitions take the real hit—multi-fab complexes erode the single-source premium both TSMC and Samsung monetize. 18-24 month outlook: expect two or three similar complexes. The foundry model is fracturing into "process-IP licensing plus local fabrication"—a 2.0 version of the 1990s IDM era, but at 2nm.