TSMC power plant funding talks still conceptual, says Taipower

Core thesis: Power is transitioning from a utility line item to a capacity-defining constraint. Technical cascade: CoWoS, SoIC, and hybrid bonding consume 3-5x the energy of legacy back-end processes with near-zero tolerance for voltage fluctuation. Taipower's push to make TSMC co-fund dedicated plants effectively internalizes grid capacity into the P&L. Fab siting logic is expanding from "talent + water" to "talent + water + megawatt-class stable supply," forcing a full rewrite of advanced packaging unit economics. Risk exposure: Binding a dedicated plant to a single tenant creates single-point-of-failure vulnerability. The 2021 Taiwan, China power rationing that forced multiple foundries to throttle output is the cautionary precedent. Concentrated power dependency is systemic fragility, not efficiency. Competitive dynamics: Samsung's Pyeongtaek cluster and Intel's Ohio fabs enjoy more diversified power procurement, absorbing shorter-term cost shocks better than TSMC. However, if TSMC passes infrastructure amortization into pricing, NVIDIA and AMD face BOM pressure, potentially accelerating capacity migration toward Samsung 2nm or Intel 18A. 12-24 month outlook: "Chip + nuclear/renewable" vertical integration will materialize simultaneously in Taiwan, China, Japan, and the US. Taipower's probe signals a paradigm shift: grid operators are beginning to collect rent from the semiconductor industry. Power now sits alongside lithography as the ceiling on capacity.