TSMC 39.6B Record Revenue|AI Chip Capacity Sold Out 2026

A Record That Exposes a Ceiling

TSMC reported June 2026 revenue of NT$442.68 billion, a 67.9% jump from the same month a year earlier and the best single month in the company's history. That figure pushed second-quarter revenue to roughly $39.6 billion, edging past the top end of the company's own guidance range and setting a new quarterly record. Every AI chip that Nvidia ships, every Apple silicon wafer that goes into an iPhone — they flow through TSMC's foundries, which is why this number lands differently than a typical earnings beat.

Yet TSMC's own CEO, C.C. Wei, stated at the April earnings call that the company would be unable to fulfill demand from American customers for years, even as new capacity comes online. CoWoS advanced packaging — the technology that physically binds chiplets into a complete AI accelerator — carries lead times of 52 to 78 weeks and is fully booked through 2026. The record revenue is not a signal that production is accelerating to meet demand. It is the ceiling itself, running at full speed.

That gap — between a result that looks like acceleration and a constraint that is structurally fixed — is exactly what forces a decision for anyone watching TSMC right now. The full Q2 earnings report, including Q3 guidance and an update on CoWoS capacity, lands Thursday, July 16. Holders must decide whether to stay through that print; watchers must decide whether to enter a record-revenue name two days before the number that either validates or reframes the whole story.

The Constraint Is Physical, Not Financial

The bottleneck the record revenue cannot resolve is CoWoS packaging, and understanding it is what separates the surface reading of this result from the structural one. Sravan Kundojjala of SemiAnalysis described TSMC's advanced N3 manufacturing node — the process targeted by every leading AI GPU and CPU shipping this year — as effectively sold out. That is not a marketing formulation. It means that the machines required to expand this capacity take years to acquire and install, and the queue is already full.

This reframes the 67.9% revenue jump from a growth story into a throughput story. TSMC generated its best-ever monthly revenue because it ran its existing lines at maximum utilization — not because output expanded meaningfully above prior capacity. The company supplies roughly 73% of the global pure-foundry market as of Q1 2026, and that share has held even as demand grew far faster than any competitor could close the gap. The record result is what full utilization looks like on a constrained but premium-priced base.

Two named analyst camps draw opposite conclusions from the same fact. SemiAnalysis argues the tightness is the bullish case — a sold-out foundry with pricing power and AI demand running ahead of supply for another year at minimum. Morgan Stanley, by contrast, published a note this week arguing that recent semiconductor weakness signals a leadership rotation away from chipmakers and toward the hyperscalers spending the capex — Microsoft, Amazon, and Meta. Both reads are grounded in the same constraint; one says scarcity supports TSMC, the other says the next marginal dollar goes to the buyer of the chips, not the maker.

Who Controls the Bottleneck

The capacity allocation data is where the constraint becomes a decision variable. Nvidia has reserved approximately 60% of TSMC's total CoWoS production — roughly 595,000 wafers — and has already locked in more than half of the expansion capacity scheduled for 2026 and 2027. AMD holds approximately 11%, or about 105,000 wafers. The top three customers together account for more than 85% of all CoWoS output. That allocation was set years in advance; no one can buy their way into the queue at this stage.

AMD faces a constraint Nvidia does not. AMD's Venice EPYC CPU, which is transitioning to the 2nm node, competes for the same restricted CoWoS allocation as the MI400 GPU. Every wafer slot used for a CPU cannot be used for an AI accelerator. Nvidia's entire CoWoS allocation is dedicated to one product family. AMD is the only major player simultaneously scaling both its server CPU and AI GPU franchises through the same physical bottleneck — a structural disadvantage that the revenue record at TSMC does nothing to relieve.

Meanwhile, a JPMorgan macro research note reveals that international emerging-market funds have been trimming TSMC precisely because the stock's rally has pushed positions past mandate concentration limits — not because their conviction changed. The selling is mechanical, not fundamental. Gavekal estimates a net $70 billion of exposure to Korean chip names has been trimmed this year for the same reason. The apparent foreign outflow from the AI chip complex is largely position-limit driven, which means the pressure is temporary and the underlying demand signal is intact.

July 16 Is the Discriminator

Thursday's full earnings report is the event that converts the revenue headline into a tradeable read. Analysts expect Q2 net profit to rise 58.8% from a year earlier. TSMC guided Q2 gross margin of 65.5% to 67.5% and operating margin of 56.5% to 58.5% — those numbers either hold or compress, and the answer will be in the report. But the metric that most sharply discriminates the thesis is Q3 guidance denominated in US dollars, where currency moves can materially shift the reported figure versus the underlying NT-dollar demand.

For holders, the counter-evidence to monitor is real: Morgan Stanley's rotation call is source-grounded and names specific stocks that stand to receive the redirected capital. If Thursday's report shows gross margin compressing below the guided range or Q3 guidance implies revenue growth decelerating from 36%, the rotation thesis gets validation and the ceiling risk becomes the dominant frame. The read survives that test only if TSMC's CEO signals CoWoS capacity expanding ahead of the stated timeline — a concrete disclosure, not a general comment about AI demand.

The move becomes a confirmed entry setup if Thursday's call delivers a Q3 revenue guide that implies acceleration or maintenance of the 36% growth rate, and if management signals CoWoS expansion ahead of the 2027 schedule. It becomes a trap if Q3 guidance decelerates and C.C. Wei reconfirms that American demand cannot be met for years with no revised timeline. For the watch-list candidate, the CoWoS capacity language on Thursday is the single metric that decides whether the ceiling is a fixed constraint or a shrinking one — and that distinction, not the record revenue itself, is what this number was asking.