TSMC posts sixth straight record quarterly profit on AI chip demand

FACT (CONFIRMED). On 15 October 2026, TSMC — the world's largest contract chipmaker and the dominant manufacturer of advanced AI silicon — reported third-quarter 2026 results that beat analyst estimates and marked its sixth consecutive record quarterly profit:

• Consolidated revenue NT$1,494.24 billion (US$46.72 billion), up 50.9% YoY (+41.2% in USD) and +17.6% QoQ; above the company's own guidance range of $44.6–45.8B.
• Net income (attributable to parent) NT$763.06 billion (≈US$23.9B at the quarter's average rate), up 68.7% YoY and +8.0% QoQ.
• Diluted EPS NT$29.43 (US$4.60 per ADR unit), +68.7% YoY.
• Gross margin 66.4% (guidance 65–67%); operating margin 58.7% — above the guidance ceiling of 58.0%; net profit margin 51.1%.
• Results beat the LSEG SmartEstimate of NT$746.3 billion net income (compiled from 16 analysts, per Reuters) and NT$1.47 trillion revenue; Straits Times cited a separate analyst estimate of NT$739.5 billion.
• The prior quarterly record was NT$706.6B set in 2Q26 (TSMC's own 2Q26 figure: NT$706.56B).

COMPANY CLAIM. Alongside the results, TSMC raised its full-year 2026 revenue growth outlook to ~44% YoY in US-dollar terms (previously "greater than 40%"), reiterated its 2026 capital budget of US$60–64 billion, said 2026–2028 capex will be "significantly higher" than the prior three years, pointed to agentic-AI and open-source-model adoption as demand drivers, announced its second Arizona fab has completed construction (mass production expected 3Q27, slightly ahead of schedule), and struck a cautious tone for 2027 planning.

Before

• TSMC was already on a record streak driven by AI accelerator demand; 2Q26 net income was NT$706.56B, itself a record (the fifth in the streak).
• Full-year 2026 revenue growth guidance stood at "greater than 40%" in USD (from the 2Q26 call in July 2026).
• The 2026 capex budget had been raised in July 2026 (from $52–56B in January to $60–64B).
• The 2nm (N2) node had entered high-volume manufacturing in 4Q25 and was ramping; the second Arizona fab was under construction.
• Consensus (LSEG SmartEstimate) expected ~NT$746.3B net income and ~NT$1.47T revenue for 3Q26.

Change (15 October 2026)

• TSMC delivered NT$763.06B net income (+68.7% YoY) — a sixth straight record quarter and an 11th consecutive quarter of profit growth (Reuters) — beating estimates on both revenue and profit.
• Operating margin (58.7%) exceeded even the top of TSMC's own guidance range, evidence of realized pricing/mix strength.
• FY2026 growth guidance was raised to ~44% USD; 4Q26 guidance of $52.3–54.3B implies ~+14% QoQ and ~+58% YoY at the midpoint (up to +61% YoY at the top end, per Reuters).
• 2nm stepped from ramp to "steep ramp-up" (8% of wafer revenue in its second full quarter of HVM); N2 annual capacity is planned to more than double in 2027.
• The AI-demand narrative broadened: management explicitly added agentic AI CPUs and open-source model adoption as silicon demand drivers alongside accelerators and networking chips.
• Simultaneously, management introduced a 2027 caution marker — a new tone element versus prior quarters' unqualified optimism.

After

• TSMC enters 4Q26 with record backlog and guidance implying another record quarter (~$53B revenue at midpoint — for context, that single quarter would approach full-year-2023 company revenue levels).
• Capex intensity stays elevated: NT$443.21B (US$13.8B) spent in 3Q26 alone; ~US$40.6B spent through September (secondary analysis), implying ~$19–23B in 4Q26 to land within the $60–64B budget.
• Geographical diversification advances: Arizona Fab 2 construction complete (mass production 3Q27); Texas actively evaluated ("Texas is one… we sent teams to Texas" — C.C. Wei); Terafab discussions acknowledged, with the condition that any TSMC-involved fab be TSMC-operated.
• The market's bar is now very high: TSMC shares are up ~60% in 2026 with a market cap around US$2 trillion (Reuters), and analysts publicly warn results are "judged against perfection."

Before: Five straight record quarters; >40% FY26 USD growth guidance; N2 ramping; consensus expecting another record (~NT$746B net income).

Change: 3Q26 net income NT$763.06B (+68.7% YoY) beats estimates; OM beats TSMC's own guidance ceiling; FY26 growth raised to ~44%; 2nm in steep ramp; agentic-AI CPUs and open-source models named as new demand vectors; 2027 prudence flagged.

After: A manufacturing chokepoint reporting demand "far above" supply capacity for the coming years, committing $60–64B capex in 2026 and "significantly higher" spending through 2028, while hedging that consumer end-markets and macro conditions could pressure 2027 planning.

FACT. TSMC is a pure-play semiconductor foundry: it manufactures chips designed by others (Nvidia, Apple, AMD, Broadcom, and every major AI accelerator designer) and does not sell its own chips. Its quarterly financials are therefore the most direct public readout of global advanced-chip demand.

• Node mix drives revenue quality. In 3Q26, wafer revenue by technology was: 2nm 8%, 3nm 33%, 5nm 28%, 7nm 10%; advanced technologies (7nm and below) were 79% of total wafer revenue. AI accelerators are built on the leading nodes (3nm/5nm/2nm) and use TSMC's advanced packaging (CoWoS), so each incremental AI order flows disproportionately to high-margin leading-edge capacity.
• Platform mix shows the AI weighting. Per TSMC's 3Q26 earnings deck (reported by CNBC): high-performance computing (HPC) = 68% of revenue, smartphones = 21%; automotive was the fastest-growing platform QoQ (+23%). HPC includes AI accelerators, data-centre processors and networking silicon.
• The demand mechanism (COMPANY CLAIM, from the 15 Oct earnings call). Chairman C.C. Wei attributed rising silicon demand to: (1) the transition "from chatbots to AI assistants" — the emerging agentic AI market, which needs both accelerators and agentic AI CPUs; (2) growing adoption of open-source AI models, which "lower the barriers to future AI usage" and expand total compute demand; (3) continued networking chip demand for AI clusters. TSMC's stated AI-accelerator revenue CAGR (narrow definition) is "mid- to high-50s percentage" for 2024–2029, with Wei saying on the call it "probably will be higher" (no formal revision).
• Capacity check discipline (COMPANY CLAIM). Unlike a pure demand-follower, TSMC said it validates demand before building: it assesses customer financing, data-centre build-out progress and power supply — it "sent our own teams of people to check on the data center build-out progress and power supply situation in the U.S." and reported feedback "so far so good." Even after these checks, Wei said demand remains above supply capability.
• Why supply stays tight. Nearly 20 fabs worldwide are in progress (13 leading-edge/advanced-packaging fabs in Taiwan plus 5–6 overseas). N2 is ramping at Hsinchu and Kaohsiung with "good yield" (HVM since 4Q25); N2 annual capacity is planned to more than double in 2027; A14 risk production is on track for 2027 with volume production in 2028; a new Tainan 3nm fab starts volume production 1H27 with an additional 3nm fab 1H28. Tool supply is tight and tool price inflation is "real," which is why capex per unit of capacity is rising.

• Technical significance: AI compute demand is now legible in a foundry's income statement. 79% of wafer revenue on ≤7nm and 68% of revenue from HPC quantify how completely the leading edge has become an AI supply chain. The 2nm ramp is the next supply lever: its steepness in 4Q26/2027 determines how much AI silicon can ship at the best economics.
• Developer significance: More supply of 2nm/3nm/CoWoS capacity feeds down the stack — better availability and eventually better price-performance of AI accelerators and dev-facing inference capacity. Agentic-AI CPU and networking-chip demand means cloud inference fleets and cluster networking are scaling alongside training.
• Enterprise significance: TSMC's guidance (~44% FY26 USD growth; 4Q26 up to +61% YoY) anchors the assumption that AI infrastructure supply remains the pacing factor into 2027. Enterprises planning 2027 AI budgets are implicitly planning against this capacity outlook and its pricing power.
• Strategic significance: TSMC is the physical chokepoint of the AI build-out. Its own words — demand "far above" supply; "all the AI players are working with TSMC"; capex 2026–2028 "significantly higher" than the prior three years — are the strongest single data point in the window that hyperscaler AI capex plans are (for now) backed by shipped-capacity reality, not just order books. The simultaneous 2027 caution is the equally important counter-signal.

INTERPRETATION (grounded in the confirmed figures):

• AI capacity visibility through 2028. With $60–64B committed for 2026 and 2026–2028 capex "significantly higher" than 2023–2025's (~$30B/yr average), buyers can plan AI-infrastructure roadmaps against a multi-year capacity ramp that includes ~20 new fabs and a doubling of N2 capacity in 2027.
• Agentic AI as a silicon category. TSMC's explicit naming of "agentic AI CPUs" as a distinct demand stream — separate from GPUs/accelerators — legitimizes the inference-side hardware build-out (CPU-based inference fleets, agent runtimes) as a durable silicon market, not a training-only cycle.
• Open-source models as a demand amplifier. Management's logic: open-source adoption lowers usage barriers → more AI workloads → more leading-edge silicon. If realized, this decouples part of chip demand from any single lab's capex decisions.
• Geographic optionality. Arizona Fab 2 completing construction ahead of schedule (mass production 3Q27) plus active Texas evaluation makes US-based leading-edge capacity real rather than announced — relevant for sovereignty requirements and supply-chain risk budgets.

1. When will TSMC formally revise its AI-accelerator CAGR (mid-to-high 50s%, 2024–2029)? Wei hinted "it probably will be higher" but declined to update it.
2. Does the 4Q26 capex implied by guidance ($19–23B in one quarter) carry into an even larger 2027 budget, and when is the 2027 number published (typically January)?
3. Will the Texas evaluation convert to an announced fab, and on what timeline? Will the Terafab arrangement (Musk-built/TSMC-operated) proceed?
4. How much does agentic-AI-CPU demand add on top of accelerators — will TSMC ever disclose it separately?
5. Does the 2027 caution deepen at the January 2027 call, or does AI demand absorb component-cost pressure as it did in 2026?
6. Does Arizona Fab 2 hit its 3Q27 mass-production date, and at what node/cost structure?

Near-term watch points (through 1Q27):

• TSMC monthly revenue reports (Sep 2026 revenue was NT$511.86B, +54.6% YoY, per AI Affairs citing the monthly filing) — the next two prints indicate whether the 4Q26 guidance midpoint (~$53.3B) is being tracked.
• 4Q26 results (expected mid-January 2027): the 2027 capex number, updated 2027 growth view, and depth of the 2027 caution are the key disclosures.
• Hyperscaler/capex follow-through: whether Nvidia, Microsoft, Google, Meta, Amazon reaffirm datacenter plans consistent with TSMC's "demand far above supply."
• Texas fab decision and Terafab arrangement status.

Longer-term (2027–2028):

• Arizona Fab 2 mass production in 3Q27 (ahead of schedule); additional Tainan 3nm fab 1H28; N2 annual capacity more than doubling in 2027; A14 volume production 2028.
• High NA EUV adoption around 2030 (company-stated), with the ASML large-format photomask transition underway.
• Whether agentic-AI CPUs become a disclosed segment and whether the AI-accelerator CAGR gets formally raised.

TSMC's 3Q26 is the AI build-out's audited receipt: a sixth straight record profit (NT$763.06B, +68.7% YoY), 79% of wafer revenue on advanced nodes, 68% from HPC — physical confirmation that compute demand is still outrunning supply at the industry's single chokepoint. But the same release carried the story's tension: TSMC raised 2026 growth to ~44% and promised "significantly higher" capex through 2028 while its CEO openly prepared the market for a more selective 2027. The honest editorial line is neither boom nor bubble — it is "demand verified, conditions tightening": the AI economy's foundation is pouring more concrete than ever, and the people running it just told everyone to watch their footing.

--

Sources

See for the full source list with URLs and roles. Lab artifact:.