TSMC July sales jump 44.7% to record NT$467.58B on AI chips

The monthly print from Taiwan's biggest chipmaker keeps arriving on schedule and keeps setting records, which is either reassuring or slightly alarming depending on where you sit. TSMC said July 2026 consolidated revenue was NT$467.58 billion, roughly US$14.51 billion, up 44.7% from a year earlier and 5.6% above June. June had itself been the previous record at NT$442.68 billion, so July is not just a good month, it is a new high-water mark. Bloomberg framed the release as evidence AI hardware demand is holding up despite recent jitters in AI-adjacent equities, and the cumulative number backs that read: the first seven months of 2026 totaled NT$2.87 trillion, about 37% ahead of the same period last year.

The interesting part, per Focus Taiwan's summary of analyst reaction, is what is doing the work inside the beat. Higher shipments on the company's most advanced nodes, particularly its 2-nanometer process, which has entered commercial production and carries a higher price tag, are being singled out as the biggest driver. That is consistent with what TSMC told investors when it lifted its full-year outlook in July: management now expects 2026 sales to grow slightly more than 40% in US dollar terms, the second upward revision of the year.

For anyone watching the AI infrastructure trade from the outside, the signal is straightforward. If the fab making most of the interesting silicon at the leading edge is still shipping into a stronger-than-expected order book two-thirds of the way through the year, whatever wobble is showing up in AI equity prices is not yet showing up in the wafers. Our own chips coverage has picked up the same story from the other end of the supply chain this week, including a fresh Sony and TSMC Kumamoto commitment for next-gen image sensors.

The caveats are worth naming rather than hand-waving. The retrieved reporting does not break the July number down by customer, so it does not say how much of the 44.7% jump is Nvidia versus Apple versus everyone else, and concentration in a small handful of hyperscalers means the same print can reverse quickly if a single order book slips. It also does not quantify how much of the beat is 2nm price mix versus unit volume, which matters for how much of this pace is repeatable. And a monthly revenue release, by design, says nothing about what happens to gross margin as the Arizona, Kumamoto and Dresden fabs ramp.

Still, if the question of the summer was whether AI capex was quietly rolling over, one of the least ambiguous places to look is the top of the foundry stack, and this month the answer is not yet.