Google VP Amin Vahdat predicts TPUs for agentic workloads

As demand for specialized AI workloads surges, hyperscalers and startups are increasingly turning to custom silicon and application-specific integrated circuits (ASICs) to improve efficiency and reduce costs. While companies like Amazon report massive revenue and performance gains from custom chips like Trainium and Graviton, the sector faces significant hurdles, including high manufacturing costs, global supply chain constraints, and the fast-moving pace of AI development. Open article →