# TSMC beat every earnings record and Wall Street sold the stock anyway > Source: > Published: 2026-07-26 23:08:38+00:00 *The world's most important chipmaker delivered a record quarter on July 16. Investors still sold the stock because the AI buildout is starting to look less like easy profit and more like a very expensive construction project.* Records are supposed to mean something. TSMC reported second-quarter 2026 revenue of $40.2 billion, up 36% year-over-year, with a gross margin of 67.7% and an operating margin of 60.3%. Net profit rose 77.4% to NT$706.56 billion, about $22 billion. Management raised its full-year revenue growth outlook to "slightly above 40%." By any normal standard, this was the report every shareholder wanted. The stock fell anyway. That's the story. Not weak demand. Not a bad quarter. Not some hidden collapse in AI chips. According to Reuters, the Philadelphia Semiconductor Index fell 3.8% on July 16, while U.S.-listed TSMC shares were down 2.5% during the session and Micron fell 4.8%. ABC News reported the next day that TSMC closed down more than 3% and Nvidia lost 2.2% as the chip selloff continued. The market looked at excellent numbers and still decided to take money off the table. The honest answer is that TSMC handed investors two things at once: the best backward-looking results in the company's history, and a forward-looking capex reset that changed the math on every free-cash-flow model in the room. The company lifted its 2026 capital expenditure guidance to $60 billion to $64 billion, up from a prior range of $52 billion to $56 billion. That's a big move. It also announced an additional $100 billion investment in Arizona, taking its planned U.S. commitment to $265 billion. Focus Taiwan reported that chairman and CEO C.C. Wei said the new Arizona spending is expected to include four additional wafer fabs and advanced packaging facilities, with the new fabs aimed at 2-nanometer and more advanced processes. TSMC also plans to build 13 leading-edge and advanced packaging fabs in Taiwan over the next several years. The ambitions are widening, not narrowing. When a company trading at a rich multiple resets its spending envelope that aggressively, investors don't just applaud the confidence. They pull out a spreadsheet. Higher spending means lower near-term free cash flow, heavier depreciation loads in later years, and a longer wait before capital can come back to shareholders. The earnings beat was real. The free-cash-flow reckoning is real too. The market chose the second one. There's also the 2nm problem. TSMC said 2-nanometer shipments made up just 3% of second-quarter wafer revenue, while 3nm was 30%, 5nm was 33% and 7nm was 11%. Together, 7nm and more advanced processes made up 77% of wafer revenue. That is exactly where you want TSMC to be if you're betting on AI chips. It is also where the cost curve gets unforgiving. Management guided third-quarter gross margin to 65% to 67%, down from the 67.7% just posted. In the earnings call transcript, TSMC said the steep 2nm ramp would dilute gross margin by about 3 to 4 percentage points in the second half of 2026. Overseas fabs add another drag. The company said overseas expansion could create gross-margin dilution of 2 to 3 percentage points in early stages and 3 to 4 points later as the scale grows. Investors heard that clearly: best-ever margins may not be the new floor. ## The AI demand story is still intact Here's the thing: this is not a demand story. Wei was asked on the call whether TSMC was updating its five-year AI semiconductor growth view. He did not give a new number. He said AI demand was "stronger and stronger" than before. That's not cautious language. You can see it in the revenue mix. High-performance computing drove the quarter, and the leading-edge nodes are where Nvidia, Apple, AMD, Broadcom and other advanced-chip customers keep turning when they need capacity. TSMC is not spending $60 billion-plus this year because the order book is thin. It is spending because customers keep asking for more wafers, more advanced packaging and more capacity in more places. But the question investors are asking now is harder than "is AI real?" Everyone has answered that one. The question is how long it takes $60 billion a year in capex, a $265 billion Arizona plan, and 13 additional Taiwan fabs to turn into free cash flow at the scale shareholders expect. You don't get that answer from one record quarter. You get it over years of utilization, pricing, yields and customer demand that has to stay strong after the concrete is poured. Frankly, that is a better question than the one the market was asking six months ago. Then, too much of the AI trade treated scarcity as permanent and margins as almost automatic. TSMC's report says demand is still running hot, but it also says the supply response is getting enormous. Fabs take years. Equipment costs are rising. Arizona is strategically valuable, but it is not Hsinchu with a different zip code. TSMC didn't do anything wrong. It executed as well as any company in the semiconductor industry has ever done. The market's reaction isn't a verdict on the quarter. It is a verdict on what comes next, and on whether the gap between spending and returns closes fast enough to justify the multiples the AI complex has been carrying. That gap is now the central question in technology investing. TSMC just made it impossible to ignore. **Also read:** [Waymo tells Uber it's going solo in Austin and Atlanta when their contract expires in 2028](https://startupfortune.com/waymo-tells-uber-its-going-solo-in-austin-and-atlanta-when-their-contract-expires-in-2028/) • [SanDisk was the S&P 500's biggest winner in 2026 until the memory chip market broke](https://startupfortune.com/sandisk-was-the-sp-500s-biggest-winner-in-2026-until-the-memory-chip-market-broke/) • [Sberbank is building a crypto exchange faster than Russia can pass the law that permits one](https://startupfortune.com/sberbank-is-building-a-crypto-exchange-faster-than-russia-can-pass-the-law-that-permits-one/)