Taiwan Semiconductor Manufacturing Company reported first-quarter revenue of roughly $35.7 billion, a 35% jump year over year and comfortably ahead of analyst forecasts. The driver, as it has been for five consecutive quarters now, is relentless demand for AI-grade chips — the N3 and N2 nodes that Nvidia, AMD, Apple, and an expanding list of hyperscaler-custom silicon programs all depend on.
What's striking about this print is how it sits against an otherwise uneven semiconductor cycle. Memory is soft. Automotive is mixed. Consumer is flat. The only demand curve still pointing vertical is the one tied to AI training and inference, and TSMC sits at the bottom of that funnel. Every hyperscaler capex announcement ultimately translates into a wafer start on the company's fab floor.
Management raised full-year revenue guidance and hinted at further price discipline on leading-edge nodes. The subtext: with only one foundry in the world capable of delivering volume 2nm this year, TSMC has pricing power it has never had before, and it intends to use it.
The open question is capacity. Arizona, Kumamoto and Dresden are ramping but years from taking a meaningful share of leading-edge output. If AI demand stays on its current trajectory through 2027, the world's bottleneck will not be model quality — it will be wafer allocation meetings in Hsinchu.