AI demand turns foundries into platform suppliers

AI accelerator demand is pushing foundry beyond wafer starts into platform capacity, where customers need advanced packaging, memory integration, qualification, delivery timing and acceptable geography to align. Demand is estimated at 110-120% of available capacity, leaving AI accelerators, custom ASICs, mobile and CPUs competing for the same advanced-node supply.

TSMC remains the default because of its market share, profit share, yield, ecosystem depth and packaging scale. Packaging has become a second moat, with wafer access increasingly tied to the ability to secure slots for CoWoS, SoIC, HBM integration and chiplet assembly.

Samsung and Intel are positioned less as replacements than as targeted alternatives. Samsung offers a more integrated pitch across logic, HBM, packaging and Taylor, though advanced-node yield and external packaging capacity still trail TSMC. Intel’s near-term opening is EMIB-T, advanced packaging and US supply assurance, rather than broad wafer share or a flagship commitment to 18A.

The main swing factor is whether TSMC can scale CoWoS and SoIC quickly enough to ease the bottleneck. If scarcity fades, urgency around Samsung and Intel weakens; if constraints persist, customers have more reason to use them for specific technical and geographic needs.