TSMC posts record revenue in second quarter on AI demand

TSMC posted Q2 revenue of T$1.27T (US$39.62B), up 36% YoY and at a record high, driven by surging demand for artificial intelligence applications. The result was slightly above the T$1.264T LSEG SmartEstimate (20 analysts). The beat alongside strong AI-led growth is likely to support positive sentiment toward the chip supply chain.

Analysis

This is less a TSM-specific print than a read-through on the bottlenecked AI stack: when the leading foundry is still accelerating, the scarce resources are advanced-node wafers, packaging, and tool time, not end-demand. That favors the highest-leverage picks-and-shovels names first — AMAT, LRCX, KLAC, and advanced packaging/substrate suppliers — because incremental AI demand usually shows up there with a 1-3 month lag before it fully reflects in customer capex plans.

The market should also infer that share is continuing to migrate toward the most capable manufacturing ecosystem, which is negative for legacy foundry capacity and for any CPU/ASIC vendor that cannot secure wafer starts. The catch is that the top line is not the same as cash generation: ongoing capex intensity and overseas buildout can keep free-cash-flow conversion muted, so the stock can still de-rate if investors decide the growth is being reinvested away rather than harvested.

Contrarian view: consensus may be too linear on AI demand. The first warning sign would be not revenue, but a slowdown in equipment bookings or any commentary implying customer digestion after the current build cycle; that would hit semis before it shows up in the foundry print. Over 6-18 months the structural trend is still positive, but the multiple ceiling remains geopolitics plus cyclicality — if export restrictions tighten or hyperscaler capex pauses, the tape can reverse quickly even with strong reported demand.

Key Decisions for Investors

• Buy TSM on a 2-5 day post-print consolidation; keep it as a 1-3 month momentum long, but cut if management signals softer H2 demand or margin compression on the next update.
• Prefer a basket long in AMAT/LRCX/KLAC over a standalone TSM add for the next 1-3 months; this is the higher-beta way to express continued AI capacity buildout with better operating leverage.
• Pair long AMAT or LRCX vs short INTC for 1-3 months to express foundry-share migration and capex intensity; thesis breaks if Intel’s foundry narrative improves or TSM indicates a capex slowdown.
• Set an alert on equipment-order commentary and hyperscaler capex guidance over the next earnings cycle; if either rolls over, reduce semiconductor beta aggressively because the AI demand story would be being pulled forward, not expanded.