TSMC's third-quarter revenue rises 51% to NT$1.49 trillion as September sales stay near record

TSMC's 55% YoY revenue surge is not a demand story—it is a capacity-bottleneck story. The real signal: advanced-node wafers and CoWoS packaging have become the single scarcest resource in the global AI compute race. ASML's EUV backlog and Lam Research's etch-tool orders now carry visibility into 2028, granting equipment makers unprecedented pricing leverage. The structural risk buried in that 54.6% growth rate is customer concentration. Demand is tethered to a handful of AI accelerator buyers. If inference workloads migrate from cloud training to edge deployment, the HBM and advanced-packaging capex cycle stretches, compressing ROI timelines. Meanwhile, Taiwan, China's power and water constraints, plus the yield-catch-up cost at the Arizona fab, are quietly eroding the gross-margin moat that defined TSMC's dominance. On competition: Samsung's 2nm GAA node, if it reaches volume production by Q1 2027, shatters TSMC's exclusive window at the leading edge. Intel's 18A foundry push, if it secures credible external customers, creates a pincer from both mature and leading nodes. The 18-month question is not who catches up—it is whether AI compute demand transitions from arms-race capex to application-level monetization. If it does not, H1 2027 will likely trigger a sharp overcapacity correction across the advanced-node supply chain.