TSMC has struggled to keep up with the demand for AI chips, and orders have started to go to competitors

The AI boom keeps wrenching the semiconductor supply chain. Demand for advanced CoWoS packaging has outstripped TSMC's capacity, and orders are leaking to Intel and other fabs — some clients are even reserving capacity that hasn't been built yet (yes, literally booking future slots). Accelerator and HPC makers are the ones pushing hardest; NVIDIA, AMD, Amazon AWS and similar players are driving much of the pressure.

NVIDIA is reportedly considering Intel’s EMIB for upcoming Feynman GPUs, a move that would be notable vs. the usual TSMC-centric flow. Taiwanese OSATs — ASE, SPIL, Powertech, KYEC — are also seeing extra work as TSMC strains under load. TSMC is not sitting still: it currently runs 5 chip packaging/testing sites in Taiwan, is adding more facilities domestically, and has 2 fabs in Arizona in the pipeline.

That said, expansion takes time. If the shortfall continues, competitors might entrench themselves sooner than anyone expected — messy, fast, and disruptive rather than neat and predictable.