TSMC's June revenue jumped 68% from a year earlier, and the world's largest contract chipmaker now heads into Thursday's earnings call with very little evidence that the AI buildout is cooling.
Taiwan Semiconductor Manufacturing Co. just posted the kind of number that should quiet anyone betting on an immediate AI spending pause. Not forever. For now. Revenue for the second quarter, the three months through June, hit NT$1.27 trillion, roughly $39.6 billion, up 36% from a year earlier. According to Bloomberg, the figure matched the average analyst estimate almost exactly. That is its own signal. Companies that beat by a hair get accused of sandbagging. Companies that miss get punished. TSMC delivered what the Street had modeled, while June alone told the sharper story: monthly sales rose 68% year over year, well ahead of May's 30% gain, which TSMC had also disclosed.
TSMC isn't a niche supplier riding one customer's spending spree. It's the foundry underneath Nvidia's GPUs, Apple's iPhone chips and AMD's server processors. If you're looking for a clean read on global AI hardware demand, this is about as close as you get. When TSMC's monthly numbers move, they're describing orders placed by nearly every major chip designer at once. The signal is cleaner.
The bottleneck is packaging
TSMC reports full second-quarter results and management guidance on July 16, and the number investors actually care about isn't June's revenue. It's what the company says about capacity for 2026 and beyond. GF Securities analyst Jeff Pu expects TSMC to raise both its full-year revenue outlook and its capital budget at that call. The company has already guided to $52 billion to $56 billion in 2026 capital spending, about 37% above what it spent in 2025, and Pu has said he expects TSMC to land at the top of that range, near $56 billion this year and roughly $73 billion in 2027.
Most of that money is chasing one bottleneck: CoWoS, the advanced packaging process used to bring logic chips and high-bandwidth memory together for AI systems such as Nvidia's Blackwell platform. The bottleneck is uglier. TSMC's CoWoS capacity has gone from about 35,000 wafers a month at the end of 2024 to roughly 75,000 by the end of 2025, and the company is targeting 125,000 to 130,000 wafers a month by the end of this year. Industry estimates suggest that expansion could shrink the gap between packaging supply and demand to around 10% by December. That gap, not basic wafer output, is the ceiling on how many AI chips actually reach customers.
The stock argument is different
The market's reaction isn't uniformly bullish. MarketWatch recently reported that Morgan Stanley strategists led by Michael Wilson see chip momentum cooling and prefer a rotation toward hyperscalers, consumer discretionary names, biotech and transports. Other recent coverage of Morgan Stanley's Lisa Shalett pointed to the same caution around semiconductor valuations after a huge run in the Philadelphia Semiconductor Index. You don't have to dismiss that warning. You just have to separate it from the operating data.
That's a valuation argument, not a demand argument. Frankly, a stock can be expensive and a business can still be growing hard. June's 68% jump suggests the underlying AI infrastructure buildout hasn't hit the wall that price watchers are worried about. What it hasn't answered is whether AI demand is broadening beyond Nvidia's ecosystem or concentrating harder around it. TSMC serves Apple and AMD too, but neither company's chip volumes carry the same data center force that Nvidia has right now.
That distinction matters. If AI spending keeps flowing mainly through Nvidia, TSMC still wins, but the market becomes more exposed to one customer's product cycle. If demand spreads further across AMD, custom silicon from cloud companies and new accelerator designs, then TSMC's packaging shortage starts to look less like a temporary squeeze and more like the central constraint in computing. You want to know which one it is before calling the top.
Thursday's call should test it. If TSMC raises capex guidance again, as Pu expects, it will be a direct bet that advanced packaging, not chip design, is now the binding constraint on the AI industry. If it doesn't, the overbought warnings will start to look less like caution and more like an early read on where this cycle actually stops.
Also read: China's Export Data Will Show How Much of Its Growth Now Runs on AI Chips • JPMorgan Cuts MiniMax's Price Target a Second Time as Zhipu Pulls Ahead • Google Delays Gemini 3.5 Pro Launch to July 17 After Scrapping Its Base Model
