Taiwan Semiconductor Manufacturing: Record Profits Chart the Future of AI (NYSE: TSM)

TSMC reported record Q2 results: revenue of NT$1.27T (~$40.2B), up 36% YoY, with gross margin up 910 bps to 67.7% and net profit margin up 1290 bps to 55.6%. Diluted EPS rose to NT$27.25 (+77% YoY, +23% QoQ), driven by strong demand for leading-edge nodes (3nm/5nm/7nm) and HPC chips (66% of revenue). Despite shares “falling” amid capex concerns, TSMC raised capex to $60B–$64B (from $52B–$56B) and forecast Q3 revenue of $44.6B–$45.8B (+37% at the midpoint), implying margins of 66% gross and 57% operating.

Analysis

The key signal is not the earnings beat itself; it is that AI demand is still forcing the ecosystem to absorb more capacity, not less. That means the bottleneck remains manufacturing and advanced packaging, which is constructive for downstream accelerators like NVDA and for the tool vendors that monetize every incremental wafer-capacity dollar. The market may underappreciate that this turns the AI trade from a pure unit-growth story into a supply-chain scarcity story, which is usually stickier for 1-3 months after results.

The near-term loser is TSMC’s own free-cash-flow profile: higher capex means the P&L can stay strong while cash conversion and ROIC flatten, so a “cheap” forward multiple can stay cheap if depreciation and ramp costs rise faster than margin expansion. Over 6-18 months, the Arizona spend is strategically bullish because it lowers geopolitical discount rate and should improve customer stickiness with U.S. hyperscalers, but that benefit likely accrues to the multiple before it shows up in earnings. Intel and other lagging foundry alternatives remain structurally pressured because TSMC is using scale to widen—not narrow—the process lead.

Contrarian view: the consensus is treating this as a clean confirmation that AI demand is intact, but the more important question is whether the industry is entering a capital-intensity phase that suppresses equity returns even as revenue grows. If next quarter’s gross margin guide slips or capex keeps ratcheting up without a matching improvement in advanced-node mix, the stock can underperform even in a bullish AI tape. The thesis is falsified if AI/HPC revenue mix rolls over, gross margin trends back toward the low-60s, or U.S./China policy introduces a new export shock.

Key Decisions for Investors

• Buy TSM on a 3%-5% post-print pullback, or use a 1-3 month call spread if you want upside with defined risk; the trade works if the market starts pricing a scarcer AI supply chain rather than just a higher capex bill. Falsify below: next-quarter gross margin guide under 65.5% or a fresh capex step-up.
• Long AMAT or LRCX against INTC for 1-3 months to express the capex-upcycle/lagging-foundry divergence; tool vendors monetize the buildout with less execution risk than TSMC’s own margin path. Take profits if order commentary decelerates or TSMC pauses its capex ramp.
• Add to NVDA on any TSM weakness rather than strength: the print reduces near-term supply anxiety for accelerators and supports hyperscaler deployment plans. This is a cleaner relative long for the next 1-3 months than chasing TSM at the current valuation.
• For risk control, consider a hedge in semis via SOXX or SMH if you are long TSM outright; the main risk is multiple compression from capex intensity, not a demand collapse. If TSM trades through the post-earnings high but free-cash-flow commentary worsens, trim 25%-30% of the position.