Musk plays down TSMC presence at AI fab facility - Mobile World Live

Musk's deliberate minimization of TSMC's role in his AI chip manufacturing is not PR spin—it is a supply-chain power-shift signal. Technically, advanced packaging (CoWoS/SoIC), not front-end wafer fab, is the true bottleneck for AI compute. If xAI is structurally reducing reliance on TSMC's packaging lines, the likely pivot is toward Intel 18A or Samsung 2nm integrated foundry-plus-packaging, or even in-house OSAT. That directly threatens TSMC's 70%+ moat in 2.5D/3D packaging. On compliance, with US export controls on China tightening, any narrative reducing structural dependence on Taiwan, China's advanced nodes earns Washington policy goodwill, materially cutting friction costs in export-license reviews for AI hardware. In market dynamics, Intel Foundry and Samsung Foundry are the most direct beneficiaries. More critically, if top AI labs begin de-TSMC-ifying their stacks, NVIDIA's and AMD's exclusive foundry bargaining power faces a repricing window. 12-to-24-month outlook: advanced packaging will evolve from a TSMC monopoly into multipolar infrastructure, echoing the 2019 foundry shift from Samsung dominance to a three-way split. But packaging yield-ramp cycles of 18 to 24 months mean TSMC remains irreplaceable near-term; genuine structural shift will not materialize before 2027.