TSMC's Revenue "Miss" Is a Mirage: The Real Print Is a Margin and Capex Escalation

Taiwan Semiconductor ADR reported Q2 FY2026 EPS of 4.31 against an estimate of 3.82, a +12.8% surprise, while revenue of 39.36 billion on the street-comparison basis came in 1.2% below the 39.83 billion consensus. Those two figures, taken at face value, describe a company that missed on demand and made it up on cost. That reading is wrong, and it is the variant perception this essay argues: the revenue comparison basis flattered consensus rather than measuring demand, because CEO C.C. Wei stated on the call that the quarter closed at USD 40.2 billion, "at the high end of our guidance in U.S. dollar terms, driven by strong demand for our leading-edge process technologies." The company's own reporting basis and the street-comparison basis are different animals, and on the company's own basis there was no miss at all. What the print actually delivered was a profitability inflection, and the guide extends it.

The margin sequence is where the argument lives. Gross margin has climbed from 53.0% in Q4 FY2023 to 66.2% in Q1 FY2026, a thirteen-point expansion over nine quarters, and CFO Wendell Huang disclosed that Q2 added another 150 basis points sequentially to reach 67.7%, "slightly ahead of our guidance, primarily due to cost improvement efforts and a slightly higher overall capacity utilization rate, partially offset by dilution from our overseas fabs." That attribution matters twice over. First, utilization and cost work, not pricing heroics, drove the beat, which means it is repeatable as long as wafer demand holds. Second, the overseas-fab dilution is now visible in the bridge yet the margin still rose, which tells you how much operating leverage the Taiwan leading-edge fabs are generating. Against the peer set the gap is no longer a gap but a different industry: UMC's latest quarter printed a 29.2% gross margin on +5.5% revenue growth, and the rest of the foundry comp table sits between 8.1% and 29.3%. TSMC is earning roughly double the gross margin of its nearest listed rival while growing revenue +35.1% year-over-year in Q1, roughly seven times UMC's pace.

The mix data explains the durability. Per Wendell Huang, "2 nm process technology contributed 3% of wafer revenue in the second quarter. 3 nm, 5 nm, and 7 nm accounted for 30%, 33%, and 11% respectively," with advanced technology at 77% of wafer revenue. A node in its first meaningful revenue quarter already contributing 3% while N3 carries 30% is the steepest leading-edge ramp profile the company has shown, and it validates the pricing power embedded in the margin trajectory. Smartphone revenue fell 4% to 22% of the total, which confirms that HPC and AI accelerators, not handsets, are setting the utilization rate that drove the gross margin beat. The consumer softness the bears keep forecasting is already in the numbers and the margin still expanded.

The forward guide extends the thesis rather than capping it. Huang guided Q3 revenue to a range of USD 44.6 billion to USD 45.8 billion, a 12% sequential increase and 37% year-over-year at the midpoint, with gross margin between 65% and 67% and operating margin between 56% and 58%. He framed the modest margin step-down explicitly: "We have just guided our third quarter gross margin to decrease by 1.7 percentage point to 66% at the midpoint." Read that against the trajectory: management is guiding a level roughly in line with the Q1 print of 66.2% and calling it a decline, which is a conservative framing of what is structurally a mid-60s margin business, a level the foundry industry has never sustained at this scale.

The capex decision is the loudest signal in the print, and it is where the second-order read-throughs start. Management raised the full-year 2026 capital budget to USD 60 billion to USD 64 billion, up from the USD 52 billion to USD 56 billion guided in January, with Huang citing "continued strong structural demand from our customers, including the newly emerging agentic AI market." Q2 capex alone was USD 15.7 billion in dollar terms, TWD 496 billion, against TWD 783 billion of operating cash flow, so the raise is funded from operations, not leverage; cash and marketable securities ended at TWD 3.5 trillion. For suppliers, this is a direct order-book event: ASML is the primary beneficiary of a leading-edge capacity add of this size, and ASM International sees the ALD and epitaxy content rise with every N2 and A16-layer build. Advanced-packaging chemical suppliers Daxin Materials, already shipping CMP slurry into N2 and TSMC Arizona, and San Fu Chemical, with its CoWoS stripper and developer franchise, get a volume tailwind from both the wafer and packaging legs of the raise.

On the customer side, the read is bifurcated by node. Leading-edge buyers are effectively pre-paying for capacity through TSMC's pricing, and the 77% advanced-node share says they are doing so willingly; the AI CAGR question on the call referenced the January guidance of a mid-to-high 50% five-year CAGR, which the capex raise implicitly reaffirms. Mature-node customers such as Microchip at JASM Kumamoto, Novatek Microelectronics in display drivers, and Rohm Semiconductor in GaN get no pricing relief, but they do get supply assurance, and for Analog Devices and Lattice Semiconductor the JASM and fine-pitch capacity continues to de-risk their own roadmaps. Nobody in that list is share-gaining away from TSMC; the capex raise widens the moat they all pay to sit behind.

The one genuinely cautionary datapoint in the pack is not financial but tonal. Call-over-call, overall sentiment slipped 0.06 to 0.31, QA sentiment fell 0.07 to 0.32, and the AI-optimism score dropped 0.14 to 0.23, while uncertainty rose 6.0 points and QA evasiveness climbed 6.1 points to 49.2, the second-highest evasiveness reading in the six-call history. Management raised capex and guided 37% growth while sounding measurably more guarded doing it, a divergence the tone history makes visible across the full series.

That divergence deserves interpretation rather than dismissal. The evasiveness spike is concentrated in the Q&A, which is where the US capex commitment, an additional USD 100 billion per the question from Sunny Lin, and the AI CAGR durability came under pressure. One plausible reading is that management is managing geopolitical disclosure, not demand uncertainty: the prepared-remarks sentiment stayed flat while the improvised answers got cagey. But a PM should hold both facts at once. The numbers say accelerate; the voice says there is something, most likely the tariff-and-geography negotiation embedded in that USD 100 billion US commitment, that management is not ready to discuss. The technology roadmap commentary was, by contrast, unusually specific: C.C. Wei put A14 at a 10-15% speed gain or 25-30% power gain versus N2 with close to 20% density improvement, and disclosed internal vehicles showing near-90% SRAM yield, language that commits the roadmap to measurable targets competitors must now answer.

The balance sheet and capital return close the quality argument. TSMC generated TWD 783 billion of operating cash flow in the quarter, lifted its cash balance by TWD 99 billion to TWD 3.1 trillion, and confirmed the dividend steps from TWD 18 per share in 2025 to TWD 24 per share in 2026, a 33% increase on top of the prior year's 28.6%. A company raising capex by USD 10 billion and the dividend by a third in the same year is not a company managing for a downturn.

What to watch next quarter: first, whether Q3 revenue lands above the USD 45.2 billion midpoint of the USD 44.6 to 45.8 billion guide, because a beat there confirms the 37% year-over-year demand claim and discredits the street-basis miss narrative; second, whether gross margin holds the top half of the 65% to 67% band, since overseas-fab dilution is now a disclosed drag and a print below 66% would signal the dilution is accelerating faster than utilization gains; third, 2nm's share of wafer revenue, which must move meaningfully off 3% for the N2 pricing story to carry the 2027 margin; and fourth, the tone metrics, specifically whether QA evasiveness falls back from 49.2 toward the 33.8 to 43.1 range of the prior three calls. If the numbers keep accelerating while the evasiveness stays elevated, treat the discount as geopolitical and buy it; if evasiveness rises again alongside any margin slip below 65%, the divergence stops being noise and starts being the thesis.