TSMC Grew 45%. The World’s AI Bottleneck Is Also Its Biggest Geopolitical Risk.

July revenue reached NT$467.6 billion as advanced-node and AI demand accelerated. The investment case now turns on whether pricing power and packaging scarcity can outrun capital intensity, overseas-fab dilution, and Taiwan risk.

TSMC Is Growing Like a Chip Designer and Earning Like Infrastructure

TSMC is becoming more important to artificial intelligence faster than the semiconductor supply chain can become less dependent on Taiwan. July revenue reached a monthly record of NT$467.6 billion, up 44.7% year over year, while revenue for the first seven months of 2026 increased 37.0%.

The second quarter showed that this growth is arriving with stronger economics. Revenue increased 36.0% in New Taiwan dollars, net income increased 77.4%, gross margin reached 67.7%, and operating margin reached 60.3%. TSMC did not have to sacrifice current profitability to capture the AI cycle.

Unlike an individual chip designer, TSMC does not need to predict which accelerator, cloud platform, networking supplier, or premium device wins. Those competing products require access to the same advanced manufacturing and packaging system.

This diversity reduces product risk, but it does not remove cycle risk. If hyperscalers slow capital spending together, the shared foundry layer will feel the decline across customers.

The market already understands much of TSMC’s quality. At $415.50 per ADR, StockResearch calculates an implied equity value of approximately $2.15 trillion. The debate is no longer whether the moat exists, but whether earnings can compound fast enough to justify an extraordinary valuation while the company spends more than ever to expand it.

AI Has Rewritten TSMC’s Revenue Mix

High-performance computing has replaced smartphones as TSMC’s financial center. HPC represented 66% of Q2 revenue, up from 61% in Q1 and 60% one year earlier, while smartphones declined to 22%. TSMC is no longer mainly a smartphone foundry with an AI option; it is an AI and HPC manufacturing platform with a large smartphone franchise.

AI systems require more than one leading-edge die. Accelerators, central processors, networking silicon, high-bandwidth memory, advanced interconnect, and packaging can create manufacturing demand across several components at once. TSMC participates without taking the product risk carried by any individual chip designer.

The customer base is concentrated at the top. The largest customer represented 19% of 2025 revenue, the second largest represented 17%, and the top ten represented 78%. Large customers improve planning, utilization, and yield learning, but they also increase the damage from a delayed product, sourcing change, or synchronized slowdown.

That concentration validates TSMC’s moat while remaining one of its largest operating risks.

The Moat Is More Than a Two-Nanometer Label

A process-node announcement is not the same as a commercially competitive foundry platform. In Q2, technologies at 7 nanometers and below represented 77% of wafer revenue, including the first 3% contribution from N2. Yet density, yield, power, performance, design rules, packaging compatibility, customer support, and available volume matter more than the label alone.

TSMC combines those capabilities at unusual scale. In 2025, it manufactured 12,682 products for 534 customers across 305 technologies. That breadth accelerates manufacturing learning and supports a mature ecosystem of design tools, qualified intellectual property, packaging options, and customer workflows. Its pure-play model adds trust because TSMC does not sell products that compete with its customers.

Advanced packaging deepens the moat. AI systems combine logic dies, high-bandwidth memory, interposers, substrates, and demanding thermal designs. CoWoS and SoIC extend TSMC’s role from producing the logic die to enabling the architecture around it, although the company does not disclose packaging-only revenue or profitability.

Samsung and Intel remain credible challengers, but roadmaps are not proof of platform parity. The competitive test is shipped volume, yield, customer adoption, and economics. TSMC’s lead is cumulative, which makes it difficult to replicate but costly to maintain.

---

The $62 Billion Capital Test

TSMC spent $29.8 billion in 2024 and $40.9 billion in 2025. It now expects $60-$64 billion in 2026, more than doubling annual capital spending in two years at the midpoint.

Approximately 70% to 80% will support advanced process technologies, around 10% will support specialty technologies, and 10% to 20% will fund advanced packaging, testing, mask making, and related investments.

TSMC can afford the build. It generated NT$2.27 trillion of operating cash flow and NT$1.00 trillion of free cash flow in 2025. In the first half of 2026, it produced NT$1.48 trillion of operating cash flow and retained NT$635.6 billion after capital expenditures.

The risk is return on incremental capital. Leading-edge tools and each new process generation are becoming more expensive, so TSMC must earn more revenue and gross profit from a larger asset base merely to preserve returns. Pricing power, utilization, and yield improvement can offset that burden, but the spending arrives before the mature return is visible.

The key signal is operating cash flow relative to capital spending. If cash generation continues to outrun investment, the expansion remains self-funding and value-accretive. If capex stays elevated while utilization or pricing weakens, free-cash-flow conversion can deteriorate quickly.

Diversification Is Necessary, Not a Cure

Arizona, Japan, and Germany improve TSMC’s resilience without recreating Taiwan’s semiconductor ecosystem. The first Arizona fab began N4 volume production in late 2024, the second targets N3 in the second half of 2027, and a third is planned for N2 and A16 by the end of the decade.

Japan adds capacity near automotive and image-sensor customers, while Germany focuses on automotive and industrial technologies rather than the most advanced AI logic. These projects diversify part of the production footprint, not the entire leading-edge cluster.

Advanced fabs take years to build, equip, qualify, and ramp, while Taiwan continues advancing to the next process generation. Overseas facilities can diversify selected production, but they cannot quickly replace Taiwan’s concentration of suppliers, engineering talent, and leading-edge capacity.

The international footprint is also more expensive. Management expects overseas-fab expansion to dilute gross margin by 2 to 3 percentage points initially and 3 to 4 points later. That cost may be justified if pricing, subsidies, and customer commitments make future cash flows more durable.

The severe tail risk remains. A military, shipping, power, cyber, or trade disruption could impair production regardless of demand or valuation. Assigning a precise probability would create false confidence; the risk belongs in scenario analysis and position sizing.

Valuation Already Knows TSMC Is Exceptional

Our full report estimates 2028 earnings per ADR, applies a scenario-specific exit multiple, and discounts the value to September 2026 at 9%.

• The Bear case produces a present value of approximately $265 per ADR. It assumes AI infrastructure spending normalizes, leading-edge utilization weakens, overseas expansion and node ramps reduce margins, and the market returns to a more conventional semiconductor multiple.
• The Base case produces approximately $491. It assumes AI and high-performance-computing demand remain strong, N2 follows a normal yield curve, packaging expands, and overseas dilution does not break TSMC’s long-term margin framework.
• The Bull case produces approximately $735. It assumes merchant accelerators and custom silicon compound faster than expected, pricing remains firm, packaging scarcity persists, and diversification improves resilience without a major loss of efficiency.

Weighting those outcomes at 25%, 50%, and 25% produces an estimated value near $496 per ADR, approximately 19% above the cutoff price.

The range matters more than the midpoint. TSMC can deliver excellent operating results and still produce weak shareholder returns if its earnings multiple compresses. That is why our view is constructive rather than low risk: the base case offers credible upside, but the bear case is severe enough to favor staged exposure over paying any price for an irreplaceable asset.

What Would Change the View

We would become more constructive if August revenue and the third-quarter result confirm the current leading-edge trajectory, N2 expands while its initial margin dilution moderates, and operating cash flow continues to fund the capital program.

We would become more cautious if HPC growth slows across several major customers, gross margin weakens for reasons beyond a normal node ramp, capital spending outgrows cash generation, or a major customer reduces leading-edge commitments.

The immediate checkpoint is August monthly revenue on September 10. It should be assessed with July’s record and third-quarter guidance rather than treated as a standalone surprise.

The Bottom Line

TSMC is not merely selling wafers into an AI boom. It controls a manufacturing platform built on process technology, yield, scale, advanced packaging, design enablement, and customer trust. The latest results show that this position can generate rapid growth and expanding margins at the same time.

Verdict: CONSTRUCTIVE with disciplined sizing. At $415.50 per ADR, investors are already paying for durable AI demand, successful node transitions, packaging expansion, and manageable overseas costs. Investors without exposure should favor a staged entry; existing shareholders can continue holding while HPC demand, advanced-node mix, margins, and cash generation remain strong.

The paradox is the thesis itself: TSMC is valuable because the world cannot replace it quickly. That same concentration creates a risk that conventional earnings models cannot fully capture.

This Lite Research Report presents the central thesis and the evidence behind it. The complete StockResearch report adds the reproducible Bear, Base, and Bull valuation model, detailed node and packaging analysis, a TSMC-Samsung-Intel competitive matrix, the full risk framework, and an investor monitoring dashboard. Get StockResearch Pro to read the full report and follow future updates (7-day money-back guarantee).

Read the full report on the StockResearch platform

Disclaimer: This article is for informational purposes only and does not constitute financial or investment advice. Equities are volatile and carry risk of loss. The market-price snapshot is as of September 2, 2026; operating and financial data are through the quarter ended June 30, 2026, with monthly revenue through July 2026. Valuation scenarios are illustrative analytical judgments and may differ materially from actual outcomes. Conduct your own due diligence before making investment decisions.

.