TSMC’s Q3 Report: Are AI Chips Finally Turning the Foundry Market? What TSM’s Earnings Mean for CapEx, Pricing and Taiwan’s Supply‑Chain Risk

TSMC posted a 39.1% year‑over‑year surge in third‑quarter profit, with revenue up 30.3% to NT$989.92 billion (about $33.1 billion), beating consensus. Net income reached NT$452.3 billion, ahead of expectations and 13.7% higher than the previous record quarter. Management credited broadening AI adoption and strong demand from key HPC customers for the upside.

The mix shift is striking. High‑performance computing, which includes AI and 5G applications, rose to 57% of sales. Advanced nodes at 7nm and below contributed 74% of wafer revenue. Industry analysts point to robust traction at 3nm and very high utilization at 4/5nm, driven by ongoing GPU/HPC processor orders and premium smartphones—an alignment that tightens capacity at the leading edge and supports throughput and pricing discipline.

The company raised its full‑year 2025 revenue growth forecast to the mid‑30% range (up from roughly 30%) and lifted its CapEx floor to about $40 billion from $38 billion, signaling conviction in the AI cycle’s durability. Management said it is monitoring U.S. tariff developments as Taiwan and Washington weigh reciprocal rates and sector‑specific duties; exemptions are likely in some cases, and TSMC’s U.S. investments provide partial insulation against policy shocks.

In market terms, TSMC’s shares continue to reflect the structural re‑rating tied to AI. Recent pricing places the stock near $298 and within reach of its 52‑week high above $311, well above its $134 low, as investors discount multi‑year demand for advanced compute.

AI has changed the foundry’s utilization math by concentrating demand at the most advanced nodes. The current tightness at 3nm and high utilization at 4/5nm signal that hyperscaler and accelerator demand continues to outstrip near‑term supply, while premium smartphones provide incremental support. That skew favors better blended ASPs and resilience in gross margins even as legacy nodes face more conventional cycles.

Customer concentration within AI and HPC brings both benefits and responsibilities. On the plus side, sustained GPU and accelerator roadmaps from leading designers enable TSMC to plan capacity with greater visibility through 2026, underpinning its CapEx upgrades and facilitating disciplined pricing. The trade‑off is that demand is more dependent on hyperscaler build‑outs and the cadence of large‑scale AI deployments, which can amplify order volatility if macro or policy conditions tighten.

Pricing power remains strongest where learning‑curve advantages are steepest. At 7nm and below, TSMC’s scale, yield learning, and ecosystem lock‑in help defend margins, particularly when capacity is full. That said, if policy friction (tariffs or export controls) raises downstream costs for consumer‑sensitive end markets, TSMC will need to balance pricing discipline with strategic concessions to preserve platform share and long‑term roadmaps. For now, the mix shift and utilization backdrop argue for margin support at the leading edge.

TSMC’s higher CapEx floor to roughly $40 billion—paired with a heavier tilt to N3/N4/N5—feeds directly into wafer‑fab equipment demand, with lithography as a bottleneck resource. ASML’s latest update helps triangulate the timing and magnitude of the next capacity wave: the company reported €5.4 billion of Q3 bookings, maintained expectations for about a 15% annual sales increase with gross margin around 52%, and guided Q4 sales to €9.2–€9.8 billion with a 51–53% margin band.

Crucially, ASML now expects 2026 sales to be no lower than 2025—effectively setting a floor—while acknowledging a significant decline in China sales next year. That geographic shift aligns with the AI‑led investment cycle in the U.S., Europe, and parts of Asia outside China, where hyperscalers and national industrial strategies are underwriting advanced‑node capacity. Market focus is already moving toward 2027 as the next inflection point when additional AI capacity is slated to come online.

For TSMC, this backdrop suggests improving tool availability for the advanced nodes central to its roadmap and validates a multi‑year capex runway. A China demand downdraft for ASML could be partially offset by AI capacity in other regions, but it also heightens the risk that future export‑control changes tighten supply to certain customers or complicate service and upgrades. Overall, the feedback loop between TSMC’s elevated CapEx and ASML’s order book supports a sturdier 2026–27 visibility, contingent on policy stability.

Geopolitics is the wild card. China’s newly expanded rare‑earth export curbs require licenses even for products containing trace amounts of these materials, asserting leverage over supply chains for electronics, EVs and defense systems. Exports of critical minerals have already fallen sharply on a year‑ago basis. With Beijing controlling the lion’s share of rare‑earth processing capacity, near‑term substitution is challenging, and the policy weapon is potent.

The U.S. response has included the threat of sweeping tariffs—up to 100%—and continued export‑control focus on advanced semiconductors and tooling. Policymakers have signaled they will not soften their negotiating stance due to equity market volatility, linking a high‑investment, AI‑centric industrial policy to longer‑term competitiveness. For semiconductor capex planners, that means budgeting for policy friction rather than treating it as a transient shock.

TSMC’s de‑risking includes significant U.S. fab investments that could ease tariff exposure and secure exemptions for specific products. Still, diversification is a partial buffer at best. Rare‑earth supply disruptions can reverberate through components and capital equipment, potentially elevating costs or extending lead times. While strategic stockpiles and alternative sources are expanding, experts estimate that building a competitive non‑China processing base could take years. The prudent stance for investors is to assume periodic policy shocks that intermittently tighten supply chains through 2026.

Base case: AI demand sustains high advanced‑node utilization through 2026, keeping pricing constructive and supporting an elevated CapEx run‑rate. ASML’s 2026 floor and TSMC’s raised CapEx floor suggest a multi‑year equipment cycle centered on EUV‑heavy nodes. Policy friction occasionally adds volatility but does not materially interrupt hyperscaler roadmaps. In this scenario, leading‑edge foundries and top lithography suppliers outperform, while legacy nodes and China‑exposed segments lag.

Bull case: Hyperscaler build‑outs accelerate, pulling forward 2026–27 tool demand as 3nm ramps faster and early 2nm risk production overlaps with sustained 4/5nm utilization. Policy exemptions and reciprocal tariff arrangements mitigate cost pass‑through to key semiconductor products, preserving end‑market elasticity. Under these conditions, WFE leaders see stronger orders and earnings leverage, and AI chip designers benefit from faster time‑to‑capacity and improved supply assurance.

Bear case: Trade escalation and stricter rare‑earth controls coincide with a steeper China demand decline, denting tool shipments and elongating product cycles. Tariff drag compresses price‑sensitive consumer electronics demand, curbing unit elasticity even as AI investments remain healthy. In this downside, mixed node economics tighten: leading‑edge nodes stay utilized but face higher cost‑of‑goods and potential service constraints; legacy nodes weaken further.

Positioning: Investors leveraged to N3/N5 ramps, EUV supply chains, and AI chip design stand to benefit from the base and bull paths, with a watchful eye on policy‑driven drawdowns as add opportunities. Risk management favors diversified exposure across regions (U.S./Europe/other Asia) and a preference for balance sheets capable of absorbing transient policy shocks and inventory swings. Macro context—rates, labor markets, and the shape of the yield curve—remains relevant for discount rates, but AI‑led cash‑flow visibility is increasingly the dominant driver for this subset of semis.