Chip Supply Shifts From TSMC: Google's Reported Intel TPU Order Lifts Foundry, but Doubts Remain - Tech Times

Recent reports suggest that Google has placed an order with Intel's foundry for over three million Tensor Processing Units (TPUs) to be manufactured by 2028, while NVIDIA is reportedly evaluating Inte... Read original →

If Google truly shifts over three million TPUs to Intel’s foundry, the immediate technical ripple will hit advanced packaging: TSMC’s CoWoS capacity is already stretched thin, while Intel’s EMIB—though cheaper—lacks proven yield stability under large-scale AI workloads. From a compliance standpoint, U.S.-driven supply chain diversification aims to mitigate geographic concentration risk, yet Intel remains a TSMC customer, revealing that its 18A node isn’t yet self-sufficient. Strategically, NVIDIA’s rumored evaluation of 18A likely pressures TSMC on pricing and capacity rather than signals real migration. TSMC will probably accelerate CoWoS expansion in Arizona and Japan to retain North American clients. Over the next 12–24 months, the lasting impact won’t be market-share upheaval but the institutionalization of dual-sourcing among AI chip designers—forcing foundries to concede on EUV allocation, packaging lead times, and IP terms, thereby redefining the foundry business model itself.