Broadcom Warns TSMC Has Hit Capacity Limits as AI Chip Demand Outstrips Supply
Broadcom flagged on Monday that manufacturing partner TSMC is running at full stretch, with AI chip demand now roughly three times above available supply. The constraint extends well beyond silicon – lead times for PCBs used in optical transceivers have ballooned from six weeks to as long as six months, and laser component suppliers are similarly squeezed. Customers across the semiconductor supply chain are locking in three-to-five-year agreements to secure future capacity, with Samsung confirming it has shifted to longer-term contracts with major buyers. The bottleneck is expected to persist into 2027 at the earliest.
Why it matters for Asia
Taiwan sits at the dead centre of this crunch. TSMC fabricates the vast majority of advanced AI chips powering everything from hyperscaler data centres to enterprise inference workloads, and any capacity ceiling directly constrains how fast companies across Asia Pacific can scale their AI infrastructure. For enterprise buyers in the region planning [GPU](https://www.amazon.sg/s?k=nvidia+gpu&tag=aiinasia-22)-heavy deployments, the message is blunt – secure your supply now or wait.