Here is the question that actually matters this morning: if the AI infrastructure build were beginning to slow, who would know first? The answer is TSMC. Every Nvidia GPU, every AMD accelerator, every Broadcom custom AI chip runs through its fabs. So when TSMC reports a monthly revenue number that is not just large but accelerating, that is not a press release. It is a production ledger.
TSMC reported a 53.3% year-on-year rise in August revenue to NT$514.806 billion, roughly $16.3 billion, with monthly revenue also increasing 10.1% from July. It marks a new monthly record. The sequential jump is the detail worth sitting with. August revenue was up from NT$467.58 billion in July, when TSMC had reported a 44.7% year-on-year increase, and the latest figures show monthly growth accelerated further. Year-on-year growth went from 44.7% to 53.3% in a single month. That is the opposite of what a decelerating capex cycle produces.
For the January-through-August period, TSMC’s consolidated revenue stood at NT$3,386.87 billion, a 39.3% increase from the same period in 2025. The trend line is not bending.
Who Is Buying and Why It Matters
In the second quarter, 2-nanometer chips contributed 3% of wafer revenue, while 3-, 5-, and 7-nanometer processes brought the total from advanced nodes to 77%. High-performance computing represented 66% of quarterly revenue. That mix is the tell. HPC is almost entirely AI accelerators and custom silicon for hyperscalers, which means the customers driving these numbers are Nvidia, AMD, Broadcom, and Apple, all of them ordering more, not less.
TSMC is ramping its 2-nanometer process, and reports in Taiwan’s press have described a target of 100,000 monthly wafers by year-end. Analysts on average are expecting 46.8% sales growth for the current quarter. August alone came in at 53.3%. The quarter’s first month has already cleared that bar by a wide margin.
The Context: Oracle Reports Tonight
TSMC’s August figure does not exist in isolation. Oracle releases its first-quarter fiscal 2027 earnings today, September 10, after the market closes. Management has guided for Q1 total revenue growth of 27% to 29%, with total cloud revenue expected to increase 58% to 64%. Oracle spent $55.7 billion in capital expenditures in fiscal 2026 alone. That spending goes somewhere. Much of it flows to chip orders that ultimately route through TSMC’s fabs. The two reports together form the clearest picture available of where the AI factory build actually stands right now.
Bull Case, Bear Case
The bull case is straightforward: TSMC is the sole company capable of manufacturing at the process nodes AI customers require, and August shows those customers are not pulling back. TSMC holds roughly 73% of the pure-play foundry market, with second-placed Samsung at about 7%, according to Counterpoint Research. That structural monopoly on leading-edge capacity means pricing power compounds alongside volume.
The bear case is less about demand and more about cost. TSMC’s third-quarter targets call for gross margin of 65% to 67% and operating margin of 56% to 58%, but overseas fab costs, the rapid 2-nanometer ramp, currency movements, and depreciation from a much larger investment program can pull in the opposite direction. Management raised its 2026 capital-spending outlook to $60 billion to $64 billion in July. Revenue at record highs does not automatically mean returns are keeping pace.
What to Watch Next
September revenue is scheduled for October 8 on TSMC’s financial calendar. That number will determine whether August was a peak or a stepping stone. Also worth tracking: TSMC and ASML this week announced an initiative to advance the industry’s transition to next-generation chipmaking technology, with TSMC planning to use ASML’s High NA equipment in high-volume manufacturing for advanced nodes starting in 2030. The capex cycle is not a 2026 story. It is a multi-year commitment being locked in right now.
August’s $16.3 billion is the hardest real-time data point available on AI infrastructure demand. It says the build is still compounding. Every investor with exposure to Nvidia, AMD, ASML, Applied Materials, or Broadcom should understand what that means for the rest of the year.
