Financials

Financials — What the Numbers Say

Reporting currency is New Taiwan Dollar (NT$). Multiples (P/E, EV/EBITDA, P/B) are computed against the TSM ADR market capitalisation in USD and are unitless — they read the same in either currency.

TSMC just printed the cleanest set of foundry numbers ever recorded: NT$3.81 trillion of revenue (+31.6% YoY), an operating margin of 50.8% (a record for a company at this scale), NT$1.00 trillion of free cash flow, and net cash of NT$2.10 trillion — and management is still raising capex into the AI build-out. The financial story is not the growth rate (foundry has cycled through worse years); it is that the operating margin keeps climbing while capex absorbs a third of revenue, which only works if pricing power and node leadership are real. The single number that determines the next 12 months of the underwriting case is the gross margin trajectory, because at this level of capex intensity each point of margin is worth roughly NT$38 billion of operating profit, and the multiple is pricing 60%+ GM as the new floor.

1. Financials in One Page

How to read this strip. Operating margin is operating profit divided by revenue — what's left of every sales dollar after cost of wafers, R&D, and selling, general and administrative expense. Free cash flow (FCF) is cash from operations minus capital expenditure — the cash management can actually distribute or reinvest after keeping the factory running. Net debt is total debt minus cash; a negative number means cash exceeds debt (net cash, a fortress balance sheet). Return on equity (ROE) is net income divided by shareholders' equity — how productively the company employs the capital book owners have left in the business. P/E and EV/EBITDA are price multiples on USD-denominated metrics (the stock trades as a Taipei-listed common stock and a US ADR), so they read the same in either currency.

2. Revenue, Margins, and Earnings Power

The income statement does three jobs: it tells you how big the business is (revenue), how much of each sales dollar drops to operating profit (margins), and what the per-share owner earnings look like (EPS). For TSMC, the most important fact in this section is margin durability through the cycle: in the 2023 downcycle the gross margin compressed by 5 points and the operating margin by 7 points, but the structure never broke — and on the way back up margins have already exceeded the pre-downcycle peak.

Twenty-year revenue and operating income

Revenue has more than doubled in three years (NT$1.59T → NT$3.81T from FY2021 to FY2025), and operating income has nearly tripled over the same window. The single-year revenue decline in FY2023 — the only one in twenty years besides FY2009 — was the post-pandemic memory and consumer-electronics inventory correction, not a structural break. The bar-on-bar relationship between the two series tells the operating-margin story: in the 2010s the operating bar was roughly 35-40% of the revenue bar; from FY2020 onward it sits closer to 45-51%.

The margin structure

Three things to notice. First, the FY2020 step-change: when 7nm and 5nm hit volume in iPhone and HPC, gross margin jumped from 46% to 53% in a single year — the first time foundry economics decoupled from the 45-48% norm of the prior decade. Second, the FY2022 spike to a 59.6% gross margin reflected pandemic super-cycle pricing, and FY2023 took back 4-5 margin points as inventory destocked — that downcycle compression is the natural test of the business model. Third, FY2025 already prints 59.9% gross and 50.8% operating, slightly above the FY2022 peak — and the latest quarter (Q1_FY2026) is running at a 66% gross margin. Margins are not normalizing; they are accelerating as the 3nm and 5nm node mix climbs.

The quarterly trajectory — the most important chart on this page

Revenue rose every single quarter from Q1_FY2024 through Q1_FY2026 — nine straight quarters of growth, with the gross margin moving from 53% to 66% over the same span. The Q1_FY2026 print is higher than the peak FY2022 quarterly margin even though the cycle is supposed to be more capacity-constrained, more capex-intensive, and more geographically dispersed (the Arizona, Japan, and Germany fabs are dilutive to margin per management commentary). The honest read: AI logic at the leading edge is now a separate margin regime from the historical foundry cycle, and TSMC is the only company in the world realizing it.

3. Cash Flow and Earnings Quality

A high reported margin does not matter if it does not convert to cash. The test is whether operating cash flow tracks net income, and whether free cash flow (cash from operations minus capital expenditure) is positive and growing. For TSMC the answer is yes — but with one structural feature every investor must understand: TSMC is the most capital-intensive megacap in the world, so a large share of operating cash is recycled into new fabs rather than returned to owners.

Net income vs. operating cash flow vs. free cash flow

The middle bar (operating cash flow) is consistently 1.3-1.9× the left bar (net income) — that's the depreciation add-back of a company that spent NT$680B on D&A in FY2025. The right bar (FCF) is much smaller because capex absorbs the rest. FCF in FY2025 of NT$1.0T is the highest in TSMC's history, finally clearing the FY2022 peak after two years of suppressed conversion driven by the 3nm capacity ramp.

Free cash flow conversion

FCF margin (FCF divided by revenue) tells you how cash-generative the business is per dollar of sales — software companies sit at 30-40%; TSMC's 26-30% is exceptional for a capital-intensive manufacturer. FCF/net income (cash conversion) is the test of earnings quality: a number persistently below 70% would signal accruals are flattering reported earnings or capex is eating the cash. TSMC's conversion ratio swings hard with the capex cycle — it bottomed at 34% in FY2023 (peak ramp of N3) and recovered to 75% in FY2024 / 59% in FY2025. There is no earnings-quality red flag here; the cash gap is explainable by capex, depreciation lag, and customer prepayments visible in the working-capital lines, not by aggressive revenue recognition or capitalized intangibles.

The capex story

Capex went from NT$258B in FY2015 to NT$1,272B in FY2025 — a five-fold increase to fund N3, N2, advanced packaging (CoWoS), and the Arizona / Kumamoto / Dresden geographic build-out. Capex intensity peaked at 53% of revenue in FY2021 (peak N5/N3 spend) and has since fallen to 33% as revenue scales faster than fab additions. The investor question is not "is capex too high" — the customer commitments (Apple, NVIDIA, AMD pre-bookings) underwrite it — but "what does FCF look like when capex normalizes?" If FY2025 capex held flat into FY2026 and revenue grew another 25%, free cash flow would expand to roughly NT$1.6 trillion (~NT$308/share) with no margin help.

Stock-based compensation and dilution

Stock-based compensation (SBC) is dilutive non-cash pay management adds back to operating cash flow. For TSMC, SBC of NT$1.25 billion in FY2025 is 0.07% of net income — essentially a rounding error. Compare this with US large-cap tech where SBC routinely runs 8-15% of net income, eroding free cash flow per share through buyback-of-issuance. TSMC compensates almost entirely in cash, and the share count has been remarkably stable at ~5.19 billion shares for nearly a decade. This is one of the cleanest accounting setups among megacap technology.

4. Balance Sheet and Financial Resilience

The balance-sheet question for a capital-intensive manufacturer is: does the company have the flexibility to keep building when the cycle turns down, without bondholders forcing decisions on shareholders? For TSMC the answer is overwhelmingly yes.

The cash and debt picture

Cash has tripled in five years to NT$3.13 trillion. Total debt of NT$1.03 trillion is essentially flat to FY2021 — TSMC issued long-dated bonds during the 2021-2022 ramp, but has not added debt since. Net cash is now NT$2.10 trillion (~$66 billion). For context, net cash alone exceeds the market capitalization of every foundry competitor except Samsung's entire diversified electronics business.

Leverage and coverage

Net debt / EBITDA at -0.84× means cash exceeds debt by roughly the same as one year of EBITDA. The 2021 trough (-0.43×) was the most levered TSMC has been since 2013; even then it was net-cash. Interest coverage (EBIT / interest expense) has consistently run above 50× — the bond market essentially treats TSMC as a sovereign analogue.

Working capital and liquidity

Current ratio (current assets / current liabilities) above 2.0× across the cycle. Inventory is essentially flat YoY despite +32% revenue growth in FY2025 — inventory turns have improved sharply, meaning the FY2023 destock was finally cleared without leaving overhang. Receivables grew slower than revenue (+3.7% vs +31.6%), implying customers are paying faster, which is consistent with the AI-customer mix (Apple, NVIDIA, AMD prepay).

5. Returns, Reinvestment, and Capital Allocation

The return-on-capital question separates real businesses from large ones. The test: does each new dollar of capital deployed earn more than its cost? For TSMC, returns have been structurally above 25% on equity for almost the entire 21-year history, and FY2025 prints the highest ROIC of the decade at 43.7%.

ROE, ROIC, and ROA

Return on invested capital (ROIC) — net operating profit after tax divided by debt plus equity capital — measures economic productivity independent of capital structure. TSMC's FY2025 ROIC of 43.7% is higher than the cost of equity for almost any reasonable assumption, meaning each marginal dollar reinvested compounds book value at a rate that would justify the high P/B if sustained. The downcycle floor (FY2023: 27%) is still better than peers' peak.

Capital allocation

Three observations. First, capex dominates — TSMC reinvests roughly 70-80% of operating cash flow in growth. Second, dividends have grown from NT$259B to NT$467B in five years (a 12% CAGR), and the payout ratio in FY2025 was 27% of net income — disciplined and rising. Third, buybacks are negligible (NT$3.1B in FY2024, zero in FY2025); management explicitly chooses reinvestment + dividends over buybacks. This is not a buyback story; it is a reinvestment story that pays a growing dividend on the side.

Per-share economics

EPS has grown at a 27% CAGR over five years without any meaningful dilution — the share count is virtually unchanged. FCF per share has nearly tripled in two years (NT$56 → NT$193), and on the FY2026 capex run-rate it could clear NT$300. The combination of low dilution, rising dividend, and rising FCF/share is a textbook high-quality compounder signature.

6. Segment and Unit Economics

TSMC reports revenue mix two ways that matter: by process technology (the node — 3nm, 5nm, 7nm and so on) and by platform (the end-market — HPC, smartphone, automotive, IoT, DCE). Geography is also disclosed but less load-bearing than node and platform.

Revenue mix by process technology (Q1_FY2026)

Advanced nodes (7nm and below) are 74% of wafer revenue; 3nm alone is 25%, up from 6% in FY2023 (16% in FY2024 → 25% Q1_FY2026). The mix is shifting toward leading edge faster than at any prior generation, and the leading edge is where TSMC has near-monopoly pricing.

Revenue mix by platform (Q1_FY2026)

HPC (high-performance computing — datacentre, AI accelerators, server CPUs) has climbed from 35% of revenue in FY2022 to 61% in Q1_FY2026, finally surpassing smartphone as the dominant end-market. This is the AI shift in financial form: TSMC is no longer a "smartphone-cyclical" foundry; it is now a "datacentre-leading-edge" foundry with smartphone as the second leg.

Year-over-year mix shift (FY2024 vs Q1_FY2026)

In five quarters, HPC moved from 51% to 61% of revenue while smartphone fell from 35% to 26%. This is the single most important structural shift in TSMC's economics in a decade — datacentre AI demand is structurally less cyclical and higher-ASP per wafer than handset SoCs, which explains why margins held even as smartphone seasonality bit.

7. Valuation and Market Expectations

The valuation judgment is not "is 29× P/E expensive" in isolation. It is: what does a 29× multiple with these margins, this return on capital, this cash position, and this growth rate actually imply about future cash flows?

Historical P/E and EV/EBITDA

The 11-year P/E range is 11.9× (FY2022 trough) to 31.1× (FY2020 peak). Today's 29.1× is at the upper end but not above the prior peak. The story is different at EV/EBITDA: FY2025's 18.0× is the highest fiscal-year-end EV/EBITDA in the dataset, which means the market is paying more for EBITDA than ever before — a quiet warning that margin expectations are stretched even if earnings multiples are not at extremes.

Where the multiple sits relative to fundamentals

Every valuation metric is above its 10-year average, but P/B is the most stretched at 1.65× its history (driven by the ROE expansion to 35%). On forward earnings (consensus is bunched around FY2026E EPS of ~NT$385-410), the implied forward P/E falls to roughly 23-25× — still above history but with the AI cycle providing the growth bridge.

A simple bear / base / bull frame

The latest TWSE close is around NT$1,500. Bear (NT$5,040) feels punitive — it requires a margin retracement that the latest quarter expressly contradicts. Base (NT$9,600) implies considerable upside but on a heroic FY2026 EPS assumption. Bull (NT$12,600) is the AI super-cycle taking three more years to roll. The market is pricing somewhere between bear and base — i.e., it is not pricing the bull case yet, which is why the upside-skew thesis remains intact even at 29× trailing earnings.

8. Peer Financial Comparison

The relevant comp set is the rest of the foundry industry plus Intel's foundry pivot. Samsung Foundry is technically the #2 player but only discloses inside the diversified Samsung Electronics balance sheet, so it cannot be benchmarked cleanly at the segment level. SMIC, the China-focused #3 foundry, only files semi-annually and does not have full ratio coverage in this dataset.

Side-by-side — FY2025 (reporting-currency comparison)

The gap is enormous and the gap is justified. TSMC's operating margin (50.8%) is 2.7× UMC's (18.5%) and 4.3× GlobalFoundries' (11.7%). ROIC (43.7%) is 4-5× the pure-play peers. Intel's foundry pivot has destroyed value at the corporate level (negative ROE, negative FCF margin). Samsung's foundry segment grew negative in FY2025 (-3.9% YoY versus TSMC +32%). There is no foundry peer that can credibly bid for TSMC's position at leading-edge logic; the multiple premium reflects an operational moat, not a sentiment premium.

The legitimate question is not "why does TSMC trade at 29× when UMC trades at 15×" — it is "why does TSMC not trade at a bigger premium." Two answers: (1) Taiwan geopolitical discount, and (2) cyclical-foundry discount that may be permanently shrinking as the AI-leading-edge mix grows.

9. What to Watch in the Financials

What the financials confirm: every quality test passes — record margins, record cash generation, fortress balance sheet, fastest-growing leading-edge node mix in foundry history, and the cleanest accounting in megacap technology. The FY2023 downcycle showed that even the bad scenarios leave TSMC with 27% ROIC and a net-cash balance sheet — most peers' best year does not reach those numbers.

What the financials contradict: the narrative that "AI is a TSMC tailwind among many" — it is not; it is the margin and growth story now. HPC at 61% of mix means smartphone seasonality, China weakness, and consumer-electronics destocking are second-order issues. The thesis bear case must come from leading-edge slip (delayed N2 / A16), customer insourcing (Apple), or geopolitics — not from financial mediocrity.

The first financial metric to watch is gross margin in the next quarterly print. If the 66.2% Q1_FY2026 number is held or exceeded in Q2_FY2026, the evidence supports the 23-25× forward P/E thesis and points toward a 27-30× anchor. If it slips below 62% on guidance, the AI-structural-margin thesis comes under question — and a 29× trailing multiple has no margin of safety against that path.