Good Monday,
TSMC's record profits signal a demand ceiling forming, where constrained fab capacity becomes the binding constraint.
Preliminary earnings from TSMC: $35.7B Q1 revenue, up 35% year-over-year, nine consecutive quarters of record profits, March alone coming in at 45% growth year-over-year with the company selling out advanced capacity through 2028. The consensus read is: AI demand is real, TSMC captures it, downstream chip designers absorb the cost.
The harder thing to price is what happens at 2nm. For the first time in a major node transition, cost-per-transistor is going up, not down. The 2nm wafer is landing around $30,000, a 10-20% step-up over 3nm's $25-27K range. The era of Moore's Law cost deflation backstopping every fabless designer's margin model is just.. over.
Watch for whether Qualcomm or MediaTek start guiding for chiplet-heavy architectures as a cost management signal in upcoming earnings. That would tell you the pressure is already biting the decisions being made today.
TSMC full earnings releases on 16 Apr Thursday - we’ll be covering this and sharing our views pre- and post-earnings call.
Know someone allocating to AI infra? They probably saw the TSMC headline without the analysis. Forward this email.
Stories That Matter
Elice's 1GW power target on brownfield steel mill sites cuts land costs but adds remediation risk. Delivering AI-grade density and cooling at that scale remains unproven.
Japan's JPY 631.5B bet on Rapidus is 1/3 of Samsung's annual fab spend. Without yield jumps, Rapidus can't close TSMC's cost gap by 2026, tightening advanced node supply further.
Oil above $100/bbl hits datacenter operating costs directly via fuel surcharges. Smaller operators face margin squeeze while hyperscalers with locked power contracts stay hedged, widening the competitive gap.
Burry's Nvidia put bet signals a 2027 capex cliff, not immediate collapse. If right, foundry and chip equipment demand faces a sharp drop in 2027-2028.
Regime Snapshot
Compute (CRS): 65, scarcity. More buyers than available capacity. Lead times extending.
Memory (MRS): 87, Shortage. Demand outstripping supply. No relief until new capacity comes online.
Memory tighter than compute this cycle. Watch for HBM allocation pressure on new model deployments.
The Daily Chain gives you the headline. The terminal shows you which companies are exposed, how the constraint map is shifting, and what the regime history says about timing.
See you tomorrow,
Teng