TSMC delivered its fifth straight quarter of record earnings on Thursday, and the beat was not a close call. April-June net profit rose 77% from a year earlier to NT$706.6B, about $21.99B, sailing past the NT$632.6B consensus. Revenue printed a record NT$1.27T, up 36% year over year, powered by AI processors the company cannot make fast enough. Management backed the momentum with its wallet, raising 2026 capital spending guidance to $60B-$64B from $52B-$56B. The print lands in the middle of a market debate about whether AI capex has run ahead of returns. This article walks through the numbers, the node economics behind them, the case for the streak continuing, and what could finally break it.
The Read
- Record Q2 net profit of $21.99B, up 77% and well above consensus
- 3nm sold out, N2 priced at a 10-20% premium, capex raised to $60B-$64B
- Fifth record quarter in a row resets expectations for the rest of 2026
A 77% Profit Jump Caps a Five-Quarter Record Streak
The headline numbers left analysts little to argue with. Net profit reached NT$706.6B ($21.99B), up 77% year over year, against an LSEG SmartEstimate of NT$632.6B weighted toward the most accurate forecasters. That is a beat of roughly NT$74B on the bottom line alone.
Revenue told the same story. The quarter came in at a record NT$1.27T, up 36% from a year ago, extending a run in which every quarter since early 2025 has set a new high. The full quarterly package is laid out on the investor relations page TSMC published alongside the release.
The beat had a warning shot. June revenue had already surged 68% year over year, telegraphing the quarter a week before the release and giving the market time to position for a strong print.
The streak matters as much as the level. Five consecutive record quarters is the kind of consistency that turns a cyclical foundry into a compounding story in investors’ minds. The tension is that consistency at records breeds records in expectations, and the bar moves up with every print.
Sold-Out 3nm and a 10-20% Premium on N2
Underneath the totals, the mix is doing the heavy lifting. TSMC’s leading-edge 3-nanometer node is sold out, backed by demand from Nvidia, Apple and AMD, while its CoWoS advanced packaging capacity remains under strain from AI accelerator orders.
The next leg is already priced. The N2 node, in volume production since Q4 2025, carries a 10% to 20% premium over 3nm, and the company is adding capacity to meet what it calls solid demand. Premium pricing on a supply-constrained node is how foundry margins defy gravity.
Margins were framed generously going in. Management had guided second-quarter gross margin to 65.5% to 67.5% on high fab utilization and ongoing cost work, partly offset by the ramp of overseas facilities. A foundry printing those levels at this scale has pricing power, not just volume.
The earnings season around it makes the contrast sharper. Chip peers have printed a much bumpier picture this cycle, as Samsung’s own Q2 report earlier this month showed on the memory side. TSMC’s advantage is that it sells shovels to every AI miner at once, whichever model wins.
An $8B Capex Raise Reads as a Demand Signal
The bull case starts with the guide, not the quarter. Raising 2026 capex to $60B-$64B from $52B-$56B is management telling the market its order book justifies billions in extra cleanroom. Foundries do not build capacity for demand they doubt.
The read-through goes beyond one company. The print is the latest sign that AI demand remains healthy, and it may be exactly what the market needed to calm mounting worries about AI overspending. It slots into the multi-trillion AI spending path Nvidia has sketched through 2030, with TSMC positioned as the toll collector on that road.
The customer list is the moat. When Nvidia, Apple and AMD all depend on the same sold-out node, the foundry captures the AI race without having to pick its winner, and switching away is measured in years, not quarters.
If 3nm stays sold out and N2 ramps at premium pricing, the earnings power compounds into 2027. In that scenario the streak extends, and the stock rerates on visibility, not just growth.
What Could Break the Streak After Five Records
The bear case is not about this quarter, which was close to flawless. It is about the setup it creates. After five straight records and a raised capex guide, any wobble in AI orders lands on a cost base that was sized for acceleration, and operating leverage cuts both ways.
The demand-shock scenario has a fresh template. Last week, the Samsung and DeepSeek signals that rocked Nvidia, Micron and AMD showed how quickly the market reprices the whole chip complex when one data point questions AI demand. A $64B capex plan is a commitment made before those questions get answered.
The asymmetry to price: the beat was worth 77% profit growth, but the guide now assumes the boom holds through 2026. Expectations, not execution, are the risk from here.
More to come.
About the Author
