The numbers
TSMC manufactures the chips that Nvidia, Apple and virtually every AI accelerator vendor design. When its revenue jumps 36 percent, that is not sentiment, it is purchase orders: the clearest hard-data signal available that AI infrastructure spending is still accelerating rather than plateauing.
Why this metric matters more than announcements
Model launches and partnership press releases are cheap. Fab capacity is not. Companies book TSMC capacity quarters in advance and pay for it, which makes this earnings line one of the most honest indicators of where the industry actually believes demand is going.
What it means for founders
Compute stays scarce and expensive for the foreseeable future. Practically: lock in inference pricing where you can, design your product so that heavier models are a tier rather than a default, and treat GPU cost per customer as a first-class metric next to CAC and churn. The companies that survived the last squeeze were the ones that knew their unit economics per token.