TSMC Q2 2026 preview: revenue is almost known — watch margin, N3 and capex

If you follow NVIDIA $NVDA, AMD $AMD, Broadcom $AVGO, Apple $AAPL or the AI-memory trade, Taiwan Semiconductor Manufacturing Company $TSM is the manufacturing report underneath all of them. TSMC reports Q2 results on Thursday, July 16.

The threat is not simply that revenue misses. It is that investors celebrate a headline beat while overlooking whether the capacity and margin structure supporting the AI buildout is getting stronger or more expensive.

TSMC discloses monthly sales. Its June release, postponed because of a typhoon day-off in Taiwan, is now scheduled for Monday, July 13 at 1:30 p.m. Taipei time. Once that number lands, investors will know nearly all of Q2 revenue three days before the earnings call.

The useful research therefore starts after the headline. Can gross margin stay near the mid-60s while N2 and overseas fabs ramp? Is N3 still so tight that TSMC must break its normal capacity discipline? Does management keep its 2026 capital budget near the top of $52–56 billion?

If you want a one-line beat prediction or price target, this is the wrong article. Those questions are already answered in TSMC's filed 20-F and official Q1 materials. Here is the pre-earnings baseline.

Monday's sales release turns the revenue beat into arithmetic

TSMC's official Q1 results page guides Q2 revenue to $39.0–40.2 billion, using an assumed exchange rate of NT$31.7 per US dollar. That converts to roughly NT$1,236–1,274 billion.

Official monthly disclosures put April sales at NT$410.726 billion and May at NT$416.975 billion, or NT$827.701 billion combined. The implied June requirement is therefore:

This is EvidInvest arithmetic, not company guidance for June: Q2's guided range, converted at TSMC's own exchange-rate assumption, less reported April and May sales.

If June lands inside that band, Thursday's revenue number should contain little surprise. The call then becomes a test of margin and forward capacity, not a contest over one quarterly top-line estimate.

The 2025 filing shows what is now concentrated inside TSMC

TSMC's 2025 Form 20-F (accession 0001628280-26-025362, filed April 16, 2026) makes the scale shift visible.

Revenue grew 31.6% to NT$3.809 trillion. Gross margin expanded from 56.1% to 59.9%, while operating margin rose from 45.7% to 50.8%. The mix changed even faster:

This is no longer a broad foundry story with AI attached. HPC supplied more than half of revenue, and the three most advanced disclosed node families supplied nearly three-quarters of wafer revenue.

The customer side is equally concentrated. TSMC's ten largest customers generated 78% of 2025 revenue. The largest represented 19%, and the second-largest 17%. TSMC does not name them in the filing. That is why the July call matters beyond the ticker: a change in demand from a small group of large customers can move utilization, capex and packaging plans across the whole AI hardware chain.

The first number to watch is 65.5–67.5%

TSMC exited Q1 with $35.90 billion of revenue, 66.2% gross margin and 58.1% operating margin. For Q2, management guided:

Those are extraordinary manufacturing margins. They are also being pulled in opposite directions.

On the positive side, TSMC said Q1 gross margin beat the high end of guidance because utilization and cost improvement were better than expected. Management expects N3 gross margin to reach and cross the corporate average in the second half of 2026.

On the negative side, the Q1 call identified two deliberate sources of dilution. The initial N2 ramp is expected to reduce full-year 2026 gross margin by 2–3 percentage points. Overseas-fab expansion is expected to dilute gross margin by 2–3 points in its early stages, widening to 3–4 points later.

That is the tension inside Thursday's print: utilization and leading-edge pricing have to absorb the cost of building the next generation and geographically diversifying the current one.

N3 expansion is the demand signal hiding behind capex

TSMC normally does not add capacity to a node after it reaches its planned size. In April, management said it was making an exception for N3 because of AI demand.

The company is adding a new 3-nanometer fab in Tainan, scheduled for volume production in the first half of 2027. Its second Arizona fab is scheduled to begin 3-nanometer production in the second half of 2027. Japan's second fab is planned to use 3-nanometer technology with production in 2028. TSMC is also converting 5-nanometer tools to support N3 in Tainan.

The most useful sentence in the Q1 transcript is not promotional:

"Still our supply is very tight. Demand continues to increase."

That was CEO C.C. Wei's explanation for moving the 2026 capital budget toward the high end of $52–56 billion.

On July 16, listen for whether that language changes. Reaffirming the capex range matters, but the explanation matters more. A higher budget because equipment arrived earlier is different from a higher budget because customers extended demand visibility. A lower budget caused by construction timing is different from one caused by weaker orders.

CoWoS: separate company statements from industry estimates

Google coverage is full of precise CoWoS wafer-per-month forecasts. Those figures are generally estimates from analysts and supply-chain publications, not numbers TSMC has guided.

TSMC's own Q1 statement was narrower and more useful. The main supply approach today remains large-reticle CoWoS. The company is building a CoPoS pilot line, with production expected "a couple of years later." Management also said advanced-packaging capacity remained very tight and that it was increasing its own capacity while working with outside assembly and test partners.

That gives Thursday's packaging discussion a clean evidence test:

Does management still call supply very tight? Does it pull CoPoS timing forward? Does it describe capacity as constraining customer shipments, or only as something TSMC is working to expand?

Until the company supplies a capacity number, a precise wafer forecast should remain labeled as an outside estimate.

TSMC's AI number is narrower than the AI economy

TSMC said high-performance computing represented 61% of Q1 revenue, up 20% sequentially. HPC is not the same thing as AI.

Management's AI-accelerator definition covers data-center GPUs, ASICs and HBM controllers used for training and inference. It excludes data-center CPUs because TSMC says it cannot reliably distinguish conventional server CPUs from AI-server CPUs.

Within that narrower definition, management maintained a mid- to high-50s compound annual growth outlook through 2029 and said the observed trajectory was moving toward the higher 50s.

That distinction matters across the supply chain. NVIDIA $NVDA, AMD $AMD, Broadcom $AVGO, Micron $MU and SK Hynix $SKHY participate in different parts of the accelerator and memory stack. Apple $AAPL also uses leading-edge capacity, but smartphone silicon does not become AI- accelerator revenue merely because it is made on an advanced node.

When management discusses AI growth, check the definition before applying the number to every advanced chip.

The filing also names what can break the clean AI story

The 20-F does not present tight capacity as pure upside. It identifies three concentrations investors should keep beside the growth numbers.

First, 75% of 2025 revenue came from customers headquartered in North America. That is commercial exposure, not manufacturing geography, but it amplifies sensitivity to a small group of US customers and US export policy.

Second, key production equipment is available from a limited number of suppliers, while some silicon wafers, gases, chemicals and photoresist are sole-sourced. TSMC says shortages can create price adjustments and delivery delays. The Q1 call added that specialty chemicals and gases such as helium and hydrogen were multi-sourced with safety inventory, and management expected no near-term interruption.

Third, geographic concentration creates measurable operating losses. Taiwan earthquakes caused approximately NT$3.0 billion of losses in Q2 2024 and NT$5.3 billion in Q1 2025, net of insurance claims. TSMC also identifies electricity and water availability as ongoing production risks.

These are not reasons to predict a failure. They are the filed costs and dependencies against which the growth story has to be measured.

The four-line scorecard for July 16

The revenue question should be mostly settled by Monday's monthly-sales release. The filing-first scorecard for Thursday is:

1. Gross margin: Does Q2 land inside 65.5–67.5%, and does H2 guidance absorb the stated N2 and overseas-fab dilution?
2. N3 capacity: Is supply still described as very tight, and are the 2027–2028 expansion dates unchanged?
3. Capital spending: Does 2026 remain near the high end of $52–56B, and is the reason stronger demand rather than timing?
4. Advanced packaging: Does TSMC change its language on CoWoS tightness or CoPoS production timing?

The popular framing is that TSMC's report will prove or disprove the AI boom. One quarter cannot do that. It can show whether the industry's central manufacturer is still earning mid-60s gross margins while spending more than $50 billion to relieve capacity constraints.

That is a better test than a headline beat. The first step is small: search accession 0001628280-26-025362 in Aether, then keep the four-line scorecard open during Thursday's call.

Sources

• TSMC Q1 2026 results and Q2 guidance
• TSMC Q1 2026 earnings-call transcript
• TSMC 2025 Form 20-F, accession 0001628280-26-025362
• TSMC 2026 monthly revenue
• TSMC financial calendar
• TSMC 2025 annual report

Research assembled with Aether's indexed TSMC 20-F and TSMC's official investor materials. Exact filing accession supplied above. Research, not investment advice.