TSMC Q2 2026: $40.2B revenue, 67.7% GM, Q3 $44.6–45.8B — Aether-backed scorecard, outlook and risks

Updated after TSMC ($TSM) became fully searchable on Aether. Quarter figures trace to TSMC's July 16, 2026 investor materials and Form 6-K accession 0001046179-26-000451. FY2025 baselines and risk language are from the Aether-indexed Form 20-F filed April 16, 2026 (accession 0001628280-26-025362). Research, not investment advice.

If you follow NVIDIA ($NVDA), AMD ($AMD), Broadcom ($AVGO), Micron ($MU) or Apple ($AAPL), Taiwan Semiconductor Manufacturing Company ($TSM) just printed the manufacturing quarter underneath all of them. The headline is clean: revenue at the top of guide, margins above guide, HPC now two-thirds of the mix, and management pointing Q3 at a steep 2-nanometer ramp.

The threat is subtler. Investors can treat a top-line hit as proof the AI cycle is intact while skipping filed details that change how you read both the quarter and the guide. Net profit got a NT$63.20 billion boost from Vanguard International Semiconductor (VIS) disposal and mark-to-market gains. Gross margin already carries overseas-fab dilution even in a beat quarter. Inventory days jumped on the N2 ramp. And the next test is not Q2 — it is whether Q3 revenue of $44.6–45.8 billion holds while margins stay in the mid-60s.

We laid out that framing before the call in our July 12 preview: revenue would be mostly settled by June sales; margin, N3 tightness, capex and CoWoS language were the scorecard. Here is how Q2 scored — plus a denser outlook and risk stack grounded in Aether-retrievable company text.

How we sourced this (Aether + primary filings)

TSMC is now in Aether's equity corpus end-to-end for this post:

Every number below is company-disclosed. Where we do simple arithmetic (H1 totals, remaining H2 capex, Q3 midpoint), we label it as math, not guidance.

Primary sources

Filed / official (July 16, 2026):

Form 6-K cover — accession 0001046179-26-000451 (earnings package exhibits on EDGAR).

TSMC Q2 2026 quarterly results and Q3 guidance — earnings release, presentation, and management report (platform mix, cash flow, capex).

TSMC 2026 monthly revenue — April–June consolidated sales (June posted July 13 after typhoon delay).

Aether-indexed baseline:

TSMC 2025 Form 20-F — accession 0001628280-26-025362 (platform, geography, node revenue, customer concentration, 2026 capex plan, export-control and utility risks).

TSMC Q1 2026 results and Q2 guidance — prior-quarter actuals and the guide Q2 just beat.

The preview scorecard

Revenue behaved exactly as the preview predicted once June landed. Margin was the surprise — not a miss. Capex spending accelerated hard in the quarter; whether that still maps to the high end of the full-year range is a trajectory question, not an empty cell.

FY2025 baseline: the AI foundry year the quarter sits on

Q2 only makes sense on top of the FY2025 20-F. Aether returns the platform table cleanly:

HPC alone added NT$716.0 billion (+48% YoY) from 2024 to 2025 — more than the entire company grew in some prior years. Smartphone still grew (+11%, NT$105.7B), but lost six percentage points of mix. That is the structural shift Q2 just extended: HPC went from 58% of FY2025 to 66% of Q2 2026.

Customer-geography revenue (by HQ country, same 20-F):

Wafer revenue by node (FY2025, NT$ millions):

Capex path filed in the same 20-F:

Annual capacity exceeded 17 million 12-inch-equivalent wafers in 2025 (primary expansion in 3nm). The press release adds scale color: in 2025 TSMC deployed 305 process technologies and manufactured 12,682 products for 534 customers.

That is the runway Q2 is accelerating on — not inventing.

June sales made Q2 revenue arithmetic, not drama

Monthly disclosures for Q2:

At TSMC's reported Q2 exchange rate of NT$31.60 per US dollar, that sum converts to roughly $40.2 billion — matching reported net revenue. June came in above the midpoint implied in the preview (~NT$427.6B) and near the top of the band (~NT$446.6B).

Sequential revenue rose 12.0% in NT dollars and 33.7% year on year in US dollars. Shipments rose only 3.9% QoQ — so ASP / mix did real work alongside volume. The beat was margin and mix, not a revenue surprise hidden in the guide.

H1 2026 math (not guidance): $35.90B + $40.20B = $76.10B. That is already ~62% of FY2025's US$122.42B print — before a guided Q3 step-up.

Profitability: manufacturing strength, plus a one-time lift

TSMC's management report attributes the gross-margin lift primarily to cost improvement and higher capacity utilization, partially offset by margin dilution from overseas fabs. That sentence matters: even a 67.7% quarter still carries geographic expansion cost in the gross line.

Separate from operations: NT$63.20 billion (~$2.0 billion at ~31.6 FX) of VIS disposal and mark-to-market gains drove most of the jump in non-operating income (total non-op NT$95.83B vs NT$28.83B in Q1). Operating profit still beat guide; net income beat by more than operations alone would explain. When you compare Q2 net profit or EPS to Q3 / FY models, strip that item.

Other P&L color from the same report:

• OpEx NT$98.98B (7.8% of revenue vs 8.3% in Q1) — operating leverage.
• Other operating income NT$5.27B, including insurance claims tied to the April 3, 2024 earthquake (the 20-F recorded ~NT$3B net quake loss in 2Q24 and ~NT$5.3B net in 1Q25 for the January 2025 quakes).
• Effective tax rate 18.1% (vs 16.8% in Q1), with higher tax on undistributed retained earnings.

Mix: HPC is 66% — node stack and geography moved with it

HPC rose five percentage points from Q1 and now supplies two-thirds of revenue, with sequential HPC revenue up 20%. Smartphone fell to 22% of mix (−4% QoQ), consistent with CEO C.C. Wei's Q1 warning that memory-price inflation could pressure price-sensitive end markets even as high-end handsets held up. North America–based customers are 78% of revenue — concentration by platform and by HQ geography.

The node table is the capacity story in numbers: 3nm is nearly one-third of wafer revenue; 2nm has already entered at 3% in early mass production. CFO Wendell Huang's written framing for Q3 is explicit:

"Moving into third quarter 2026, we expect our business to be supported by continued strong demand for our leading-edge process technologies, including the steep ramp-up of our 2-nanometer technology."

HPC products in the 20-F explicitly include AI accelerators, AI GPUs, AI ASICs, server processors and high-speed networking chips for AI, cloud and enterprise data centers — plus 3DFabric / TSMC-SoIC / CoWoS advanced packaging. That is the filing-backed AI link; it is still not a pure "AI revenue %" disclosure.

Balance sheet and cash: spending into the ramp

Inventory days rose seven days to 87, "primarily due to N2 ramp." Free cash flow fell sequentially even as operating cash rose, because capex ($15.70B in the quarter, $26.80B year-to-date) outpaced the OCF gain. That is what a steep leading-edge build looks like on the cash statement: still strongly FCF-positive, but the spend is real.

Capacity-reservation signal from the 20-F (year-end): temporary receipts from customers were NT$189.9B at Dec 31, 2025, down from NT$291.1B at Dec 31, 2024. Contract liabilities were NT$50.0B (from NT$89.4B). Lower prepayments can mean customers are drawing reserved capacity — or that new reservation intensity cooled. Either way it is a leading indicator to watch alongside monthly sales, not a one-line bull/bear.

Outlook: five numeric paths from here

1. Q3 guide — another ~$5B step-up, softer margins at the midpoint

Read the guide as three simultaneous statements:

1. Demand — another sequential revenue jump on the same order as Q1→Q2, driven by leading-edge demand and a steep N2 ramp.
2. Margin path — midpoint gross margin steps down from 67.7% into the mid-60s. That fits the Q1 roadmap: N2 ramp and overseas fabs dilute even when utilization stays high.
3. FX — Q3 assumes NT$32.0 per dollar vs 31.60 realized in Q2.

Dollar bridge (illustrative math, not company FY guidance):

H1 alone is already $76.1B. A midpoint Q3 puts the company past FY2025's full-year US dollar print ($122.42B) with a full Q4 still ahead — which is why the market cares less about "did they beat Q2" and more about whether the mid-60s margin + N2 ramp combination survives the next two prints.

2. Capex path — H2 still has ~$25–29B to spend if the band holds

H1 at $26.8B is roughly half of a $52–56B year if H2 matches. Q2 alone at $15.7B is a hot spend quarter; sustaining that pace would overshoot the filed band, so either H2 slows or management revises. The 20-F says 2026 spend focuses on N2/N3 capacity (Fab 20, 21, 22), specialty + advanced packaging (Fab 24), and R&D.

3. N2 / advanced-mix outlook

Bull path: N2 ramps without breaking the GM band; HPC stays ≥65% of mix; monthly sales track the Q3 bracket. Base path: GM lands mid-band (66%), N2 share rises slowly, overseas dilution visible every quarter. Bear path: Q3 revenue sticks but GM tests the 65% floor while inventory days stay elevated — the market then reprices "AI foundry" as a margin-dilution story.

4. Overseas footprint and the US build

A March 2025 6-K (Aether-indexed) said TSMC intends to expand US investment to US$165 billion (prior $65B Arizona plan plus $100B for three new fabs, two advanced packaging facilities and an R&D center). That is the multi-year dilution and resilience trade in one number: more geographic redundancy, structurally higher cost vs Taiwan megafabs. Q2's own language — gross margin helped by utilization, partially offset by overseas fabs — is the near-term P&L expression of that strategy.

5. What to watch next (30–90 days)

• Monthly sales into Q3 vs the $44.6–45.8B bracket
• N2 wafer-revenue share after the early 3% print
• Gross margin vs the 65–67% band as overseas capacity and N2 ramp
• Capex pace vs remaining ~$25–29B of a $52–56B full-year frame
• CoWoS / advanced-packaging tightness language once the call transcript is out (Q1: large-reticle CoWoS still the main approach; packaging "very tight"; CoPoS pilot "a couple of years later")
• Smartphone sequential trend after the −4% QoQ soft patch
• Temporary receipts / contract liabilities on the next balance sheet (capacity reservation intensity)

Risks: what the beat does not retire

A record-looking foundry quarter does not erase the risk stack. These are the ones that still matter after July 16 — grounded in company language and the filed 20-F, not a vibe check.

1. Customer concentration is still rising into the AI cycle

From the Aether-indexed 2025 20-F (Item 2):

The top-ten share rose eight points in two years. The largest customer's share fell, but the second-largest rose to 17% — concentration broadened inside the top two rather than disappearing. Q2's own geography print is even more North America–heavy at 78%. HPC at 66% of revenue means the AI / high-end compute cycle and a small set of fabless (and system) buyers are the same trade.

The 20-F is explicit that a more concentrated base "may subject our revenue to seasonal demand fluctuations," that system companies designing their own silicon can create "significant variations in our sales if the growth of their products and services, particularly in the AI sector, is volatile," and that export controls aimed at major customers can cut demand.

2. One-time gains can flatter the "quality" of EPS

VIS disposal / MTM gains of NT$63.20B are disclosed and large. Operating margin beat guide without them; net income and EPS did not. Models that annualize Q2 net profit without stripping non-op will overstate run-rate earnings power into Q3–Q4. Strip ~NT$2.0B / ADR ~$0. something of that distortion before you chain EPS growth.

Rough quality check (illustrative): Q2 net income NT$706.56B less NT$63.20B VIS-related ≈ NT$643B — still a huge quarter, but closer to the operating story (+~12% QoQ on that adjusted lens vs +23.4% reported).

3. Margin dilution is structural, not a footnote

Management already says overseas fabs diluted Q2 gross margin inside a beat. Q3 guides GM 65–67% and OM 56–58% — below Q2 actuals at the midpoint. N2 ramp inventory (87 days) is the balance-sheet twin of that dilution story. The bull case is utilization and cost improvement keep overpowering dilution; the risk is the dilution stack (N2 + overseas + US$165B US build) wins a quarter before demand does.

4. Capex and free-cash-flow timing risk

Q2 FCF fell QoQ because capex jumped to $15.70B. That is not distress — ending cash & marketable securities are NT$3,518B — but it is a commitment. If demand visibility softens while spend stays high, the market reprices TSMC as a cash-conversion story, not just a revenue story. If spend slows while AI demand stays tight, the constraint shifts back to capacity / packaging.

5. End-market bifurcation (HPC vs smartphone)

HPC +20% QoQ and smartphone −4% QoQ is a clean split. FY2025 already showed the same pattern at annual scale (HPC +48%, smartphone share down to 29%). The risk is not that HPC fails tomorrow; it is that smartphone / broader consumer softness lasts longer than the AI narrative assumes, or that memory-cost pressure (Wei's Q1 theme) bleeds into handset builds that still matter for leading-edge utilization between AI peaks.

6. FX

Guides are given in US dollars with an explicit NT$ assumption. A move in USD/NTD changes NT-dollar margins and translation optics even when US-dollar demand is unchanged. Q3's jump from 31.60 realized to 32.0 assumed is itself a modeling input, not noise. At scale: every NT$0.10 move on ~$45B quarterly revenue is roughly NT$4.5B of translation — before any margin effect from a local-currency cost base.

7. Export controls and dual-use AI compute rules

The 20-F details the October 2022/2023 U.S. "October Rules" on advanced computing ICs and notes that in January 2025 the U.S. issued further rules under which TSMC "may need to obtain an export license prior to shipping products using 16-nanometer or below process to specified destinations unless specific conditions are met." Shipments "may be delayed or prohibited." TSMC Nanjing's Validated End-User authorization expired in December 2025; operations now depend on an annual export license with "no assurance" of renewal. China was still ~9% of FY2025 revenue and 6% of Q2 — not the growth engine, but a live compliance and demand risk.

8. Geopolitics and Taiwan manufacturing concentration

TSMC's safe-harbor language points readers to the 20-F risk factors for what can make forward statements wrong. Geographic revenue concentration in North America customers, manufacturing concentration in Taiwan, export-control regimes, and overseas fab execution are not retired by a strong June quarter. They are the reason the company is building outside Taiwan and the reason that build dilutes margins.

9. Electricity, water, and natural disasters

Filed utility risk is concrete: Taiwan electricity tariffs applicable to TSMC rose 25% effective April 1, 2024 vs 2023, then another 14% in October 2024. The 20-F flags drought and grid-stress risks for water and power in the science parks. Earthquake history is not abstract either — ~NT$3B net loss after the April 2024 quake and ~NT$5.3B net after January 2025 quakes — with Q2 2026 still booking related insurance recoveries.

10. Advanced packaging still the hidden bottleneck

The written July 16 release is light on new CoWoS capacity numbers. Until the call transcript is filed, treat third-party wafer / CoWoS unit forecasts as outside estimates — not company guidance. For NVIDIA, AMD, Broadcom and the HBM stack (including Micron), packaging remaining "very tight" (Q1 language) can cap sell-in even when TSMC wafer revenue prints well. The 20-F's own HPC section treats CoWoS / SoIC / InFO as part of the product, not an optional extra.

11. Customer consolidation and AI demand volatility

The 20-F warns that industry M&A among customers "will further decrease the overall number of our customer pool," and that AI-sector growth that is "volatile or not sustainable" can create significant sales variation. A 66% HPC quarter increases that beta: the upside and the single-cycle risk are the same exposure.

Valuation and related research

Click any ticker above (or below) for the EvidInvest fair-value page:

• TSMC ($TSM) valuation · earnings metrics
• NVIDIA ($NVDA) valuation
• AMD ($AMD) valuation
• Broadcom ($AVGO) valuation
• Micron ($MU) valuation
• Apple ($AAPL) valuation
• July 12 Q2 preview

Sources

• Form 6-K, accession 0001046179-26-000451
• TSMC Q2 2026 quarterly results
• TSMC 2026 monthly revenue
• TSMC Q1 2026 quarterly results
• TSMC 2025 Form 20-F, accession 0001628280-26-025362
• TSMC US investment expansion 6-K (Mar 2025)
• EvidInvest Q2 preview (July 12)
• Aether transcript_search / financial_search over TSM Q2 FY2026 press release, 2025 Form 20-F, and 2026 monthly revenue 6-Ks (retrieval date 2026-07-16)

Research assembled from TSMC's July 16 earnings release, management report, presentation, monthly revenue tables, and the Aether-indexed 2025 20-F. Call Q&A / official transcript may add CoWoS and FY color not in the written package. Illustrative bridges (H1 totals, remaining H2 capex, adjusted net income) are arithmetic on disclosed figures, not company guidance. Research, not investment advice.