TSMC Posts $39.6B Q2 Record as AI Chip Demand Rewrites Its Calendar

Taiwan Semiconductor Manufacturing Company delivered numbers on Monday that left little room for debate: the AI buildout is not slowing down. TSMC reported second-quarter revenue of $39.62 billion, a 36% increase year-on-year that beat the top end of the company’s own April guidance and set a new all-time quarterly record for the world’s dominant chip foundry.

The milestone is striking not just for its size but for the seasonal story it disrupts. June revenue came in at NT$442.68 billion — a 67.9% surge versus June 2025 — snapping a pattern of summer softness that had held for four consecutive years. Historically, TSMC’s second quarter has been its weakest, dragged down by the cyclicality of consumer smartphones and PCs. The AI cycle has rendered that calendar irrelevant.

A Foundry That Now Sets Its Own Seasons

For TSMC, the shift is structural. Spending from hyperscalers on AI training clusters and inference infrastructure does not follow the back-to-school or holiday device rhythms that once governed the semiconductor industry. Nvidia’s accelerator chips — manufactured almost exclusively by TSMC on its 4nm and soon 2nm processes — are ordered in response to datacenter buildout timelines dictated by Microsoft, Amazon, Google, and Meta. Apple’s M-series silicon, also a TSMC exclusive, adds a second non-cyclical revenue stream as the Mac and iPad lines move toward more frequent refresh schedules driven by on-device AI features.

First-half 2026 revenue reached NT$2.4 trillion, representing 35.6% growth over the same period a year ago. Analysts at Morgan Stanley estimated that AI-related revenue now accounts for more than 55% of TSMC’s wafer revenue, up from roughly 38% in early 2025.

The Concentration Risk Nobody Wants to Discuss

The record numbers mask a structural dependency that TSMC’s own investors have been quietly flagging. Three customers — Nvidia, Apple, and Broadcom — collectively account for an estimated 60 to 65% of the company’s revenue. Nvidia alone may represent close to 30% after a series of capacity agreements signed during Blackwell’s ramp.

That concentration is a feature for investors as long as AI spending holds, and a serious vulnerability if it doesn’t. The Bank for International Settlements published a report this week warning that hyperscaler AI capital expenditure is now outpacing earnings and free cash flow across the five largest technology companies, forcing some to issue debt to sustain spending. Any deceleration in that pipeline flows almost directly into TSMC’s order book within two to three quarters.

TSMC has moved aggressively to diversify geographically, partly in response to geopolitical pressure and partly to reduce customer concentration risk. The company’s Arizona Fab 21 is on track for volume production of 2nm chips in 2026, and a second facility in Japan recently reached full operational status manufacturing N12e chips for automotive and industrial customers. Neither diversification track, however, comes close to matching the revenue density of Taiwan’s advanced node fabs.

Advanced Nodes Are the Only Game That Matters

TSMC’s pricing power is increasingly concentrated at the bleeding edge of process technology. Its N3 and N2 node families command dramatically higher wafer prices than mature nodes — industry estimates peg 2nm wafer prices at $30,000 to $35,000 each, roughly three times the cost of a 7nm wafer — and virtually every leading-edge AI chip in production or development today is routed through TSMC.

Intel Foundry, Samsung Foundry, and Rapidus in Japan all aspire to reclaim leading-edge manufacturing capacity, but none has demonstrated the yield consistency required to win Nvidia or Apple’s most critical workloads. The competitive gap, counterintuitively, may be widening: TSMC’s process technology lead and its ecosystem of packaging innovations like CoWoS-S and SoIC give it capabilities no competitor has replicated at scale.

The company’s next architectural step is its A16 process, a 1.6nm-class node featuring nanosheet transistors and backside power delivery. TSMC has confirmed OpenAI as one of the first customers for A16 production capacity, scheduled to begin risk production in late 2026 and volume production in 2027. That pipeline, if it holds, means TSMC’s revenue trajectory may not have peaked yet.

What the Numbers Mean for the Broader Ecosystem

The scale of TSMC’s Q2 result reverberates well beyond Hsinchu. For Nvidia, it is a validation that Blackwell orders are converting to revenue on schedule. For Apple, it signals that TSMC can absorb both AI accelerator demand and the iPhone’s fall production surge without one starving the other of capacity. For SK Hynix, which supplies the HBM3E memory stacked on Nvidia’s H200 and B200 GPUs, TSMC’s order strength is a leading indicator for its own trajectory.

For the Taiwan semiconductor ecosystem more broadly — the ODMs, substrate makers, and materials suppliers that cluster around TSMC — the record quarter translates into elevated orders through at least year-end. And for policymakers in Washington who have been pushing CHIPS Act incentives to build domestic capacity, the numbers underscore both the urgency of the effort and the scale of the gap that needs to be closed.

TSMC is scheduled to report full Q2 earnings, including margins and outlook, on July 17. Analysts will be watching gross margin guidance particularly closely: the mix shift toward advanced packaging, which carries lower margins than pure wafer fabrication, has become an important variable in a revenue story that otherwise looks almost uniformly positive.

In the meantime, the quarterly record stands as the clearest possible signal that whatever uncertainty surrounds the AI investment cycle at the macroeconomic level, the orders placed with the world’s most important chip manufacturer have not yet blinked.