TSMC reported a quarter that left little doubt about the strength of current AI infrastructure demand. Second-quarter revenue rose 33.7 percent from a year earlier to $40.2 billion, while net profit increased 77.4 percent to T$706.6 billion. Gross margin reached 67.7 percent, operating margin reached 60.3 percent, and the company guided third-quarter revenue to between $44.6 billion and $45.8 billion.
It also raised its 2026 capital-spending forecast to between $60 billion and $64 billion, up from a previous range of $52 billion to $56 billion, and said full-year dollar revenue should grow by slightly more than 40 percent. Another $100 billion of planned U.S. investment added to a report built around continued demand, expanding capacity and confidence that customers will need considerably more computing power.
Taken together, the results offer little support for the idea that AI demand is weakening. TSMC is selling more, earning wider margins and committing more capital because its customers are still asking it to produce more.
Yet the report landed after a sharp semiconductor selloff in Asia, and the first U.S. reaction offered little relief. South Korea’s KOSPI fell 6.2 percent, Samsung Electronics declined 6.6 percent, and SK Hynix lost 9 percent. Japan’s Nikkei fell 3 percent, while ASML had closed 0.4 percent lower in Europe on Wednesday despite raising its forecast and announcing additional production capacity.
At about 4:00 a.m. Eastern, TSMC’s U.S. shares were down roughly 0.3 percent, the VanEck Semiconductor ETF was down 1.6 percent, and the iShares Semiconductor ETF was down 2.2 percent.
The timing limits how much can be inferred from those moves. TSMC’s Taiwan-listed shares had closed before the earnings release, and most Asian markets were already falling, so the report didn’t cause the regional selloff. Nor was the early U.S. reaction uniformly negative. Nvidia was slightly higher, and ASML’s U.S. shares were up about 2.2 percent.
The weakness also remained concentrated. S&P 500 exposure was slightly positive in early trading, while credit ETFs and long-duration Treasurys were modestly firmer. Nasdaq and semiconductor exposure were weaker, but there was little evidence of a broader growth scare or funding shock spreading across the market.
A positioning unwind therefore remains a plausible explanation. Semiconductor shares have risen sharply, South Korea’s market has become heavily dependent on Samsung and SK Hynix, and leveraged retail products can amplify a reversal. TSMC’s modest premarket decline could also disappear once regular trading begins.
But positioning alone doesn’t explain why results this strong produced so little immediate enthusiasm across the group.
For much of the AI buildout, evidence of scarce chips, memory and manufacturing capacity was enough to support nearly every company positioned near the bottleneck. Investors were still trying to determine whether the demand would last and whether suppliers could convert it into profit.
TSMC’s quarter answers both questions as clearly as almost any supplier could. Demand remains strong, and the company is converting it into rapid profit growth and a gross margin of 67.7 percent.
Once those answers become widely assumed, however, another strong quarter adds less new information. Investors begin looking beyond the current order book and asking whether future returns can keep pace with the amount of capital being committed, the capacity being added and the growth already reflected in share prices.
That shifts the focus from whether the AI buildout is real to what the next round of investment will earn.
TSMC isn’t sacrificing current profitability to fund its expansion, but a capital budget approaching $64 billion, together with another $100 billion of planned U.S. investment, raises the amount of future revenue and profit needed to make each new project as productive as the last one. The present quarter can be excellent even as the standard for future quarters continues to rise.
None of this means AI capital spending has become unprofitable, and it doesn’t suggest the semiconductor cycle has peaked. It means the shares can become harder to satisfy before the businesses themselves become less attractive.
The U.S. opening will provide a better test, although retail sales, jobless claims and the Philadelphia Fed survey at 8:30 a.m. Eastern could complicate the first move. If TSMC rallies and semiconductor breadth improves, the overnight weakness will look more like a regional positioning unwind after an extraordinary run.
But if TSMC remains flat or falls while SMH, SOXX and the large memory suppliers stay weak, investors will be signaling that confirmation of strong AI demand is no longer enough by itself.
The buildout can remain strong while the trade becomes more demanding. Once record revenue, wider margins and larger investment plans are already expected, semiconductor companies must keep proving that returns can rise faster than the capital and expectations already committed to them.