History

The Story

Figures converted from TWD at historical FX rates — see data/company.json.fx_rates. Ratios, margins, and multiples are unitless and unchanged.

TSMC's narrative has compressed in five years from "5G/HPC growth driver" into "the company AI cannot exist without," and management has not had to walk anything back to get there. The 2022 chip-shortage surge gave way to a real revenue contraction in 2023, but rather than blame demand, management explained it as inventory destocking and re-anchored the story on a single multi-year wave — AI-related HPC — that has subsequently delivered. The risk discussion has migrated from COVID and Taiwan-power constraints to tariffs, export licensing and trade-secret theft, which is what happens when a company becomes geopolitically essential. Guidance has been beaten every single quarter in the reviewed window, and the only quantitative promise that narrowed the story (2025 capex $38–42B) landed inside the range.

1. The Narrative Arc

The current chapter of the business began in 2018 — the year founder Morris Chang retired, C.C. Wei became CEO and Mark Liu became Chairman — and accelerated when Wei consolidated both roles in June 2024 after Liu's December 2023 retirement. The strategic content of the chapter, however, was set by three external forces TSMC then organized itself around: U.S./China decoupling (Huawei September 2020), the COVID-era digitalization surge, and from late 2022 onward, generative AI.

2. What Management Emphasized — and Then Stopped Emphasizing

The earnings-call vocabulary has rotated decisively. "5G" and "smartphone" were the 2021 anchors; "AI," "HPC" and "leading-edge process technologies" are now the entire script. The shift was not announced — it was simply done. By Q1 2026, the prepared remarks describe the business as "supported by continued strong demand for our leading-edge process technologies" and nothing else.

The shift in revenue mix is the receipt:

Quietly de-emphasized. "5G" as a standalone driver has effectively disappeared from prepared remarks; smartphone seasonality is now framed as a headwind to be offset by AI rather than a primary business. China geographic mix dropped from 20% (2019) to roughly 8% (2025) without management ever announcing a China strategy change — the U.S. export-control regime did the work. Mention of COVID and pandemic acceleration, ubiquitous in 2021, was gone by 2024.

3. Risk Evolution

The risk-factors discussion has roughly doubled in surface area, almost all of it in two clusters: (a) operating across multiple jurisdictions with conflicting rules, and (b) the geopolitical fragility of being the chokepoint of the AI supply chain.

New since FY2024. Tariffs (IEEPA reciprocal tariffs in April 2025; Section 232 advanced-computing tariff in December 2025; February 2026 Supreme Court ruling and the January 2026 U.S./Taiwan trade pact capping reciprocal rates at 15%). Trade-secret theft escalated from a generic line item to a full case file: three indicted in August 2025 over 2nm leakage to a Tokyo Electron supplier; former SVP Lo Wei-jen indicted in November 2025 over alleged transfer of 2nm/A16/A14 process secrets to Intel; Taiwan's National Security Act invoked for the first time in a chip-secrets case in December 2025.

Materialized. Earthquakes were a perennial line item that became real: ~$92M loss in Q2 2024 from the April 2024 quake; ~$162M loss in Q1 2025 from the January 2025 quake. Both losses were absorbed without margin shock — gross margin held above 53% in the affected quarters.

Concentrated. Top-10 customer share rose from 70% (2023) to 78% (2025). The top customer's share fell from 26% (2021) to 19% (2025), but the second-largest jumped from 11% (2023) to 17% (2025) — the customer pool is narrower, not broader.

Faded. COVID disclosures (entire dedicated paragraphs in FY2021) are gone. The Huawei-specific carve-out has dissolved into a broader export-control regime. China-Taiwan strait risk is still listed but has not been re-prioritized despite the broader news flow — management has consistently treated it as a steady-state risk, not a rising one.

4. How They Handled Bad News

TSMC's record in the reviewed window contains only one episode that qualifies as bad news at the headline level — the 2023 revenue decline of -4.5% — and a series of smaller operational hits (two earthquakes, the Huawei chip-diversion notification, the 2025 trade-secrets case).

Two patterns stand out. First, every operational hit has been quantified in the filings ($92M, $162M, top-10 customer percentages, 16nm license scope) — there is no soft-pedalling of dollar impact. Second, the 2023 demand miss was reframed but not denied: the prepared-remarks language pivoted from "slowdown" to "inventory adjustment," and the company simultaneously raised the visibility of AI as the cycle that would replace it. By Q1 2025, management was confident enough to flag tariff uncertainty before it had hit financials:

"While we have not seen any changes in our customers' behavior so far, uncertainties and risks from the potential impact from tariff policies exist. We will continue to closely monitor… and manage our business prudently." — Wendell Huang, CFO, Q1 2025

That single sentence is the only piece of forward-looking caution in six quarters of prepared remarks. It is also the only one that has not later been retracted.

5. Guidance Track Record

Across six consecutive quarters of detailed revenue/margin guidance, TSMC has cleared the high end of every revenue range and met or beat the high end of every margin range. Capex has tracked within the announced budget. The pattern is not a beat-by-pennies pattern — Q2 2025 revenue came in 4.4% above the top of the guided range, and Q3 2025 gross margin landed 200 bps above the guided upper bound.

Long-term promises (multi-year).

Credibility score: 9 / 10. TSMC's six-for-six on quarterly revenue guidance is unusual for a $120B+ business, and every multi-year capex, node-ramp and margin target in the window has landed. The single point deducted reflects two genuine residuals: (i) the 2022 capex range was met only after being reframed mid-year due to equipment delivery delays — a small but real walk-back; and (ii) management has never publicly quantified the financial exposure from the October 2024 chip-diversion disclosure, the 16nm export licensing or the 2nm trade-secrets case, so the soft tail of geopolitical risk is unverifiable from filings alone. The score would be 10 in a world where TSMC also disclosed quantified outcome ranges for those items.

6. What the Story Is Now

The story today is simpler than it has ever been: TSMC manufactures the chips that the world's largest companies cannot make anywhere else, and management is delivering the financials that statement implies. Q1 2026 gross margin of 66.2% and operating margin of 58.1% are not "foundry-business" numbers — they are software-business margins on a $140B/yr capital-intensive base, and they have been earned in a quarter that absorbed January-2025-quake reconstruction, tariff uncertainty and a high-profile IP-leak case.

De-risked. The "AI demand is real" debate is over for this management team. Capex of $52–56B in 2026 is being committed against a customer base that has been pre-buying capacity, and the long-term USD revenue CAGR target of ~25% through 2029 is the most aggressive long-term number TSMC has ever published — and is being delivered in real time. Customer concentration risk has been called out without softening, and the 16nm/Nanjing export regime has been quantified. The 2nm node is in production and contributed to margin expansion in Q1 2026.

Still stretched. Three items. (1) The customer pool: 78% of revenue from the top 10, top-2 customers at 19% and 17% — a single Apple/NVIDIA cycle pause could compress Q1 2026's 50.5% net margin meaningfully. (2) Geopolitical tail: the Section 232 25% tariff on advanced computing chips, the January 2026 U.S./Taiwan trade pact, and the still-unresolved A14/A16 IP-theft case sit outside management's control and are not quantified in guidance. (3) Long-term gross-margin claim of "56%+ through the cycle" relies on continued mix shift to 3nm/2nm at high utilization; the 2023 destocking quarter (GM low-50s) is the only data point on what a real cyclical trough now looks like, and the long-term claim has not been stress-tested against a deeper one.

What to believe vs. discount.

• Believe: the 25% USD CAGR target through 2029 (raised, not announced; consistent with announced capex and customer roadmaps), the 56%+ GM floor (already cleared by 10 ppts in Q1 2026), and the 2026 above-30% USD revenue growth.
• Discount: any implicit suggestion that geopolitical risks have been managed rather than deferred — the tariff regime, the trade-secrets exposure and the 16nm license renewals are running, not resolved. The base case is that they remain manageable; the tail is that any one of them re-prices the multiple in a single quarter.
• Watch: the second-largest customer's share (11% → 17% in two years). Either it continues to rise — meaning TSMC is becoming a two-customer business at the leading edge — or it normalizes, which itself implies an AI demand pause.

The current narrative is not stretched relative to what management has actually delivered. It is stretched relative to what management cannot control. Both have been true for several years; what has changed is that the financials no longer require any narrative help.