TSMC's Net Profit Surged 77% — Here's What the Headline Isn't Telling You

The near-term picture is defined by a tension that is unusual even for TSMC. Revenue guidance for Q3 of $44.6–$45.8 billion represents an 11–14% sequential jump from Q2's $40.2 billion — an aggressive and confidence-signaling range. The exchange rate assumption embedded in guidance is 1 USD = 32 NTD, reflecting slightly more NT dollar weakness than the Q2 actual rate of 31.60. Yet despite this outsized revenue jump, margins are expected to step down. Gross margin guidance of 65–67% and operating margin guidance of 56–58% both fall below Q2's records of 67.7% and 60.3%. CFO Wendell Huang identified the direct cause: N2 ramp-up costs, which he said will dilute second-half gross margin by approximately 3–4 percentage points. A new node's early production quarters almost always show margin compression as yields mature, and this pattern has played out predictably with every major TSMC node transition in the past decade.

The key interpretive question for the next six months is whether this margin step-down is structural or temporary. I believe it is temporary, but the timing of recovery matters. The precedent from N3 and N5 ramp-ups is encouraging: both nodes compressed margins in their early production quarters, and both recovered as yields matured and volumes scaled. The logic should apply to N2 as well. As N2 scales from its Q2 debut of 3% revenue contribution to a larger share of the mix, the higher per-wafer ASP of N2 should more than offset the early-stage yield dilution — but in my estimation, meaningful margin recovery from this is more likely a 2027 first-half story than a Q3 or Q4 story. The market will likely price in "peak margin" concern in the near term, which creates a narrative that is technically correct in the very short run and likely wrong over a 12-month horizon. How quickly N2 yield ramps will be the single most important leading indicator.

Short-term demand momentum remains strong on the volume side. Annual revenue guidance of "slightly above 40% year-over-year in U.S. dollar terms" is an unusually confident public commitment, and the trajectory supports it. H1 2026 revenue of NT$2,404.48 billion was already up 35.6% year-over-year. June monthly revenue of NT$442.68 billion came in 67.9% above the prior year, confirming no demand softening heading into Q3. TrendForce forecasts TSMC's 2026 revenue growth at approximately 32% versus the global foundry market's growth of +24.8%, implying that TSMC's market share expansion — from 70.4% in Q4 2025 to 72.3% in Q1 2026 — continues through the year. The dividend signal also reinforces management's confidence: NT$24 per share for 2026 is a 33% increase year-over-year, demonstrating the company believes cash returns to shareholders remain sustainable even alongside record-high CapEx. Given H1 momentum and the Q3 revenue guidance, hitting the annual target looks achievable, with gross margins as the primary uncertainty.

The medium-term picture is dominated by the CapEx cycle. Wendell Huang's $60–64 billion annual CapEx guidance for 2026 is already extraordinary in absolute terms, and C.C. Wei's statement that "the CapEx in the next three years will be even more significantly higher than the past three years" signals this is a sustained elevation, not a temporary spike. The Arizona program's current phase includes Arizona Fab 1 (N4, in production since early 2025), Arizona Fab 2 (N3/N2, targeted for 2028), Arizona Fab 3 (advanced process, targeted for late 2020s), plus four additional advanced 2nm-and-beyond fabs announced this quarter — all before accounting for further phases. CHIPS Act support of $6.6 billion in grants, $5 billion in loans, and up to 25% investment tax credits on qualifying spend helps, but represents a modest fraction of the $265 billion total commitment. During this window, the CHIPS Act's five-year buyback restriction limits shareholder return flexibility to dividends alone. Free cash flow will remain compressed for an extended period, and markets will need to price the long-dated return on capital rather than near-term FCF yield. The core thesis that justifies accepting this sustained FCF pressure rests entirely on AI demand remaining structurally elevated — and that remains an assumption, not a guarantee.

The competitive landscape looks favorable to TSMC over the medium term. TrendForce data shows Samsung's foundry share contracting from 7.1% in Q4 2025 to 6.5% in Q1 2026 — the gap versus TSMC is widening, not narrowing, despite Samsung's public commitments to advanced node competitiveness. Industry reports suggest Samsung's SF2 (2nm) yields may be approaching 55–60%, but initial production yields and mature high-volume manufacturing yields are fundamentally different metrics. Intel's 18A process has reportedly resolved earlier yield challenges, but high-volume manufacturing for external customers at scale remains ahead. C.C. Wei's comment that choosing and ramping a process technology "is not buying a milk from 7-Eleven" captures the customer migration barrier precisely: transferring a critical leading-edge design from TSMC to a competitor involves two to four years of re-qualification work and substantial yield risk. In an environment where AI chip demand is as time-sensitive as it currently is, almost no major fabless customer will accept that disruption. TSMC's lead is probably wider today than it was 18 months ago.

The long-term picture is shaped by two variables that dwarf all others: geopolitical restructuring and the structural durability of AI demand. On geopolitics, Bloomberg Economics' model estimate of a $10.6 trillion (9.6% of global GDP) first-year impact from a Taiwan conflict scenario quantifies the systemic risk in the current concentration structure. TSMC's response — the $265 billion Arizona buildout — is the most direct evidence that TSMC itself takes this risk seriously. North America generates 74% of TSMC's revenue, making U.S. manufacturing a natural extension of the customer relationship as well as a risk management tool. The open question is cost structure: U.S. manufacturing runs at a structurally higher cost than Taiwan, and whether this ultimately translates into margin compression or is absorbed through premium pricing remains unresolved until Arizona Fab 1 generates a meaningful operational cost dataset. The dual-geography production system will almost certainly change TSMC's cost structure permanently; the direction of the change on margins depends on which force — higher costs or higher pricing power — dominates.

On AI demand durability, C.C. Wei made a remark during the conference call that I think deserves more attention than it received. He observed that the emergence of agentic AI is reviving the CPU's role in AI data centers — meaning demand for compute silicon is expanding from AI accelerators to include CPUs and custom ASICs simultaneously. This multi-architecture demand base makes the AI chip market structurally broader than a single GPU product cycle. The combined $700 billion-plus in AI CapEx from major hyperscalers in 2026 supports this structural breadth: Amazon at roughly $200 billion, Alphabet at roughly $190 billion, Meta at $125–145 billion, and Microsoft at over $120 billion are each committing multi-year capital to AI infrastructure. TrendForce forecasts the global foundry market reaching $218.8 billion in 2026, up 24.8% year-over-year — and TSMC is growing faster than the market. If agentic AI applications drive a sustained multi-year silicon demand cycle across accelerators, CPUs, and custom ASICs, the investment TSMC is making today becomes increasingly rational in hindsight.

Here is how I frame the scenario analysis, stated as conditions rather than probabilities. The constructive scenario requires that hyperscaler AI CapEx remains at or above current levels into 2027, that neither Samsung nor Intel achieves credible high-volume advanced-node external manufacturing before 2029, and that the Taiwan Strait security environment remains roughly stable. Under those conditions, TSMC's revenue growth continues to exceed the annual guidance of "slightly above 40% year-over-year in U.S. dollar terms," N2 yields mature to allow margin recovery in the first half of 2027, and the CapEx cycle's returns begin to become visible in FCF recovery. The more challenging scenario requires that hyperscalers pull back AI investment materially after failing to generate expected returns on their 2025–2026 spending, or that U.S.-China tensions escalate into technology sanctions that directly affect TSMC's manufacturing operations or major customers. In that scenario, inventory days at 87, annual CapEx commitments of $60–64 billion, and FCF pressure all become simultaneous headwinds rather than manageable trade-offs.

The most important single condition under which my constructive medium-term view could be wrong is a faster-than-expected inflection in AI ROI disappointment. The current hyperscaler investment pattern has characteristics of an arms race — driven partly by competitive fear of falling behind rather than purely by measured return-on-capital calculus. Arms races of this type can decelerate rapidly and simultaneously once the competitive dynamic shifts. If that deceleration arrives in 2027 rather than 2029, TSMC's inventory position, its CapEx pipeline, and its margin trajectory all come under pressure at the same time. Samsung delivering credible 2nm high-volume manufacturing sooner than expected is also a scenario I wouldn't rule out entirely — industry reports suggest SF2 yields are progressing, and while the gap to TSMC's manufacturing maturity remains wide, it is not infinite or permanent. Yield and volume manufacturing stability are different capabilities, but the distance between them can close faster than markets expect if Samsung sustains investment focus.

Three metrics will tell the TSMC story over the next two to three quarters more accurately than any headline number. First, whether Q3 gross margin lands in the middle or at the upper end of the 65–67% guidance range — the upper end would suggest N2 yield is maturing faster than feared. Second, how quickly N2 revenue share expands from its Q2 debut of 3% — acceleration toward 8–10% within two quarters would be a strong positive signal on both demand and manufacturing ramp quality. Third, whether inventory days of 87 stabilize or continue rising — further increase would shift the interpretation from proactive demand positioning toward early demand softening. These three data points from the Q3 report will either validate or challenge the thesis that Q2's margin compression is a temporary ramp-up cost on the way to a structurally more profitable next phase, rather than the beginning of a sustained downturn in TSMC's exceptional margin profile.