Snapshot: Executive Summary
TSMC is the closest thing the technology economy has to a true bottleneck monopoly. It fabricates ~90% of the world's leading-edge logic and effectively all frontier-AI silicon: every Nvidia GPU, every hyperscaler accelerator, Apple's flagship SoCs, and that position buys real pricing power. In Q2 2026 it printed records across the board: revenue +33.7% YoY (US$40.2B; NT$1,270.38B, +36% YoY), gross margin 67.7%, operating margin 60.3%, and a record 55.6% net margin1. Management raised FY2026 USD revenue growth guidance to "slightly above 40%" (from "above 30%") and lifted 2026 capex to US$60–64B.
So why isn't this a screaming buy? Because the price already reflects much of it. Our DCF, a capex-heavy model that charges the ~$56B capex in full, implies ~$425/ADR, roughly the current ~$423 price (detailed below). But be clear-eyed about what drives that: it is not a conservative result. It rests on a 5.0% Gordon terminal growth, and terminal value is ~86% of the figure, so the perpetuity rate is the valuation. At a GDP-anchored ~3% terminal the DCF falls to ~$300 (≈30% overvalued). On cash flows, then, TSM is fairly-valued-to-rich depending entirely on how durable you believe the AI cycle is. The case for owning it anyway rests on three things a single-point DCF doesn't capture:
- A long runway the model under-captures. TSM will plausibly grow well above the 5% terminal for another decade, which a 5-year-explicit DCF structurally understates, so the "aggressive" 5% terminal is partly compensating for a too-short explicit window. Our $490 target is a judgment-weighted 12-month figure that sits below the ~$540 consensus, not a mechanical roll-forward.
- Relative value. ~22× forward earnings for a >90%-leading-edge near-monopoly is the cheapest high-quality way to own AI compute: versus ASML at ~47×, AVGO richer, and a ~37× semis median.
- Pricing power. Multi-year 5–10% advanced-node price hikes are sustaining record margins, and two binding bottlenecks (CoWoS packaging, N2 capacity) are being broken, converting demand into revenue.
We rate TSM a Buy, 12-month PT $490 (bull $590 / bear $355), but with eyes open. The margin of safety is thin-to-negative on conservative terminal assumptions, so the rating is explicitly a bet on two things: that the AI super-cycle extends into the terminal (the 5% growth), and the cheap-vs-ASML relative-value comp (~22× vs ~47×). It is not cushioned by an independent margin of safety, and a reasonable analyst could rate this a Hold on the DCF alone. Downgrade trigger: the AI-accelerator CAGR visibly rolling over, or the multiple having to expand to justify the price. The bear is real: an un-hedgeable Taiwan single-point-of-failure, a still-pending US ITC patent ruling (a live binary, final Commission decision ~Oct), the H1 monthly-sales wobble (a soft April, then a record May) now steadied by the Q2 beat, and capex at ~44% of revenue.
Tactical: TSM at ~$423 sits ~16% below our $490 12-month target, the cheapest high-quality way to own AI compute (~22× fwd vs ASML ~47×) with a >90% leading-edge near-monopoly and record margins. But our DCF says it is roughly fairly valued today, so this is a quality-and-cycle Buy, not a deep-value one. Swing factors: AI-demand durability and the Taiwan / ITC tail.
Investment Thesis
Bull Case
- AI demand sustains a mid-to-high-50s% accelerator CAGR; the FY26 guide is raised again; revenue compounds faster and longer than the ~21% the price implies
- Pricing power + N2/A16 mix lift gross margin toward ~70%; CoWoS bottleneck clears cleanly
- The multiple holds ~22–25× as the cycle's durability de-risks; WACC compresses as the Taiwan tail discount narrows
- DCF at a ~9.0% WACC / 5.0% terminal on the bull cash flows supports a ~$577 intrinsic; the $590 target adds a year of growth
Base Case
- Revenue ~$160B FY26E (+31%) fading to ~$314B by FY30 (~21% CAGR); EBITDA margin holds ~67–69%
- DCF intrinsic ~$425 today at ~9.3% WACC / 5% terminal, a year of ~15–20% growth rolls it to ~$490
- ~22× forward holds; consensus ~$540; the AI cycle stays intact but doesn't run hotter than priced
- Capex ~$56B normalizes as a share of revenue; margins absorb overseas-fab dilution
Bear Case
- AI-accelerator demand digests; the raised "slightly above 40%" guide slips and monthly sales keep undershooting
- The multiple compresses toward ~16× as the cycle de-rates; ~44%-of-revenue capex meets under-utilization
- The Taiwan / ITC tail re-rates the concentration risk (an adverse ITC ruling, or a geopolitical scare)
- DCF at ~9.7% WACC / 4.8% terminal on lower revenue re-tests ~$315 intrinsic; the $355 target assumes a less-severe path
Rating: Buy. The probability-weighted blend (45% base / 30% bull / 25% bear) is ~$486, above the ~$423 spot. We rate TSM a Buy on the AI cycle, the pricing power, and the cheapest-quality-AI-compute comp, but, unlike a deep-value name, the DCF says you are paying ~fair value, so the upside is the cycle's durability rather than a re-rating from cheap. The Q2 2026 beat and the raised FY26 guide ("slightly above 40%" revenue, capex lifted to $60–64B) reinforce the bull case; we hold the house target at $490, deliberately below the ~$540 Street. Downgrade trigger: the AI-accelerator CAGR visibly rolling over, or a structural Taiwan-risk re-rating.
The Foundry Monopoly
TSMC's moat is not a product; it is a position. Building a leading-edge fab costs >$20B and takes years, the process know-how compounds with volume, and the customer ecosystem (EDA tools, IP, packaging) is built around TSMC's nodes. The result: ~72% of all foundry revenue and ~90%+ of leading-edge (≤3nm), with effectively no second source for frontier AI chips. That position is what lets TSMC raise advanced-node prices 5–10% on a multi-year basis2.
Revenue by platform (Q2 2026)
HPC/AI is now 66% of revenue (a record, ~3× smartphone), the company has transformed from a smartphone-led foundry into the AI-compute supply chain's bottleneck.
Revenue by technology node (wafer, Q2 2026)
Advanced nodes (≤7nm) are 77% of wafer revenue (N5 33% / N3 30% / N2 3%). The leading edge is where the margin and the monopoly live; mature nodes are a stable base.
Why the monopoly holds
- Capital + know-how barrier. Only TSMC, Samsung and Intel even attempt leading-edge, and TSMC is years ahead on yield and ramp, the metrics that actually determine cost-per-good-die. The gap is widening: TSMC's foundry share rose to 72.3% in Q1 2026 (from ~70% the prior quarter), versus Samsung #2 at ~6.5% and SMIC ~5.1%6, roughly 11× Samsung's foundry revenue. On the node that matters, Samsung's SF2 (2nm) yields are stuck in the mid-50s% (below the ~60% mass-production bar) while TSMC's N2 runs 60–70%, so Qualcomm kept its 2026 flagship Snapdragon on TSMC N2P; Intel's 18A is in production but its external foundry traction is still negligible (~$174M in Q1 vs TSMC's ~$36B/quarter). There is no credible second source at the frontier.
- Ecosystem lock-in. Customers co-design to TSMC's process design kits; switching foundries means re-spinning silicon. The frontier-AI customers (Nvidia, AMD, hyperscalers, Apple) are effectively single-sourced here.
- Pricing power, finally exercised. After years of customer-friendly pricing, TSMC is pushing 5–10% advanced-node and CoWoS price increases, a structural margin lever that the bull case compounds. Supply-chain reports (not official TSMC guidance) point to a further up-to-15% 3nm-specific hike in 2H 2026 and another 5–10% in 2027 on tight Fab 18 capacity, upside to the margin story if confirmed.
AI / HPC: The Engine
The single most important fact about TSMC in 2026 is that it has become the toll booth on the entire AI build-out. HPC/AI hit a record 66% of revenue in Q2 2026 (~3× smartphone), and management has guided to a mid-to-high-50s% AI-accelerator revenue CAGR through 2029. Crucially, TSMC captures this demand regardless of which chip wins, Nvidia, AMD, Google TPU, Amazon Trainium, a startup's ASIC, because they all fab here. It is a bet on AI compute in aggregate, not on any one design.
Two June 2026 data points make the breadth concrete. Nvidia's Vera Rubin (336B transistors) entered full production on TSMC's N3 with CoWoS-L (GTC Taipei, June 1), keeping 3nm demand elevated; and the custom-ASIC pipeline that broadens the base beyond Nvidia keeps filling, every major hyperscaler accelerator (Google's TPU "Ironwood", Meta MTIA, and the OpenAI/Broadcom "Jalapeño" inference ASIC unveiled June 24) is fabricated and packaged here. Nvidia is also the lead (reportedly only) customer for A16 via its Feynman generation. The toll is widening, not narrowing, even as it concentrates: on 2026 estimates Nvidia is ~22% of revenue and Apple ~18%, so the top two approach ~40%.
HPC/AI revenue ($B, est.)
The AI-compute platform compounding from a low-double-digit-$B base toward the dominant share of a >$160B revenue company, the engine of the "slightly above 40%" growth.
Revenue by geography (customer HQ)
North America is 76% of revenue, the US AI/HPC + Apple demand base. China has fallen to ~7% (from ~22% historically) on export controls. The demand is American; the fab is Taiwanese, the crux of the bear case.
The demand signal: a scare, then a record
The near-term tape told two stories in six weeks. April 2026 sales were soft (+17.5% YoY), feeding an "AI-demand-digestion" narrative; then May printed a single-month record: NT$416.98B, +30.1% YoY5, re-accelerating off the soft April. Jan–May cumulative revenue is +30.0% YoY (NT$), squarely on pace with the "above 30%" USD full-year guide once the NT$→USD translation (which has been running several points favorable) is added back. So the "first crack" read more like monthly lumpiness ahead of the N2/Rubin ramp than a maturing capex wave. The Q2 call (July 16) resolved it: a record quarter (revenue +33.7% YoY in USD, +36% in NT$) with FY guidance raised to "slightly above 40%" and capex lifted to $60–64B, rebutting the digestion narrative. The next monthly prints and the Q3 call (~Oct 15) are the arbiters from here.
Capex, CoWoS & the N2 Ramp
TSMC is spending at an extraordinary rate to convert AI demand into revenue. At the Q2 call it raised 2026 capex to US$60–64B (from ~$56B), still roughly 44% of revenue, and reiterated the next three years will be "significantly higher." That spend is the bull's bottleneck-breaker and the bear's free-cash-flow drag at the same time (it suppresses FCF to a ~57× P/FCF; see Financials). Our DCF conservatively holds the prior ~$56B base, so the raised guide is incremental spend the intrinsic value does not yet reflect.
Annual capex ($B)3
Capex stepped up sharply into the AI cycle and stays elevated, the price of the monopoly. The question the DCF answers: does it earn its return?
CoWoS capacity (k wafers/month, est.)
CoWoS advanced packaging, the binding constraint on AI-accelerator shipments, is scaling ~4× to ~120–140k wafers/month. The supply-demand gap is targeted to narrow from ~20% to ~10% by end-2026. (Nvidia reportedly secured >60% of 2026 capacity for Rubin.)
The N2 (2nm) ramp: the next leg
N2 entered volume production in Q4 2025, and 2026 brings the most aggressive ramp in TSMC's history, five fabs, scaling toward ~200k wafers/month by 2027, with slots largely sold out into 2027. N2P reaches volume production in 2H 2026; A16 (the 1.6nm-class node with Super Power Rail backside power) is production-ready in 2026, but its customer volume ramp was re-slotted to 2027 at the April 2026 Tech Symposium, with Nvidia the lead customer via its Feynman generation. The roadmap then extends to N2U and A14 (2028) and A13/A12 (2029): all without High-NA EUV through 2029, a multi-year leading-edge runway no competitor can match. The overseas fabs (Arizona / Japan / Germany) are the offset: they dilute gross margin ~2–3% in FY26 (widening to 3–4% later), the cost of de-risking the Taiwan concentration, though Arizona Fab 1 is already profitable.
Financial Health & Trends
Annual revenue ($B)
Gross & operating margin (%)
FY20–FY26E: the compounding machine
TSMC has nearly tripled revenue in five years, $45B (FY20) to $122B (FY25), through one cyclical dip (FY23) and out the other side into the AI super-cycle; our model carries FY2026E at a conservative ~$160B (+31%)4, below management's raised "slightly above 40%" guide (~$172B). Margins have stepped to records: Q2 2026 delivered 67.7% gross / 60.3% operating and a record 55.6% net margin. This is a hardware company with software-like profitability.
FY20–FY30E: revenue & operating margin
Revenue compounds on the AI build-out while operating margin holds in the mid-to-high 50s. FY26E is our modeled base (a conservative ~$160B, below the raised guide); FY27E–FY30E are author estimates used in the DCF.
The free-cash-flow nuance
Here is the catch behind the record margins: at ~44% of revenue, capex consumes most of operating cash flow, so free cash flow is far lower than the income statement implies, a P/FCF of ~57× on today's capex-suppressed FCF versus the ~22× forward P/E. TSMC earns like a monopoly but converts to cash like a fab during a build-out (the DCF's later-year FCF rises as capex eases as a share of a larger revenue base). The DCF below treats this honestly, capex is subtracted in full, which is precisely why the intrinsic value lands near the price rather than far above it.
Capital Allocation & Returns
TSMC runs a net-cash balance sheet (~$64B; ~$98B cash less ~$34B debt) and returns capital primarily through a growing dividend (~0.6% yield, ~24% payout) that rises with earnings. The capital story, though, is not returns, it is reinvestment: ~$56B of modeled annual capex (raised at Q2 to a $60–64B guide) into the AI build-out, at high incremental ROIC for now. Buybacks are not a meaningful part of the model.
The dividend grows steadily and the balance sheet is fortress-grade, but an investor is not buying TSMC for capital return, they are underwriting a multi-year, ~44%-of-revenue capital program and trusting it earns its cost of capital across the AI cycle. The DCF is the test of that trust.
Valuation & Comps
Two things are simultaneously true. TSM at ~22× forward earnings is the cheapest high-quality way to own AI compute: cheaper than ASML (~47×), comparable to or below AVGO, roughly in line with Nvidia, and below the ~37× semis median, for a >90%-leading-edge near-monopoly growing 40%+ at record (~56% net) margins. And the stock has re-rated hard, from a ~13× trough P/E in the 2022 down-cycle to the top of its own five-year range, so the cheapness is relative, not absolute.
TSM forward P/E history
From the ~13× cyclical trough (2022) to ~22× today, the AI re-rating is fundamental (more growth, higher margins), but it has compressed the cheapness.
Forward P/E: AI-compute peers
At ~22× TSM is the cheapest of the leading-edge AI-supply-chain names, notably below ASML (~47×), the other monopoly, despite comparable growth and the broadest demand capture.
On other lenses TSM screens similarly full-but-not-extreme: ~20.5× EV/EBITDA and ~16× sales, against ~30× trailing / ~22× forward earnings. And the Street has kept moving up, not down: through late June 2026, BofA raised its target to $590 (from $490) and Susquehanna to $575 (from $500), both reiterating Buy/Positive on agentic-AI demand, CoWoS capacity, and the advanced-node price hikes, with zero downgrades7; the Q2 beat and raised guide pushed estimates and targets higher again. Consensus now sits ~$540 (Strong Buy). Our $490 house target sits deliberately below the Street: we hold it because the DCF says TSM is ~fairly valued, so we underwrite the cycle's durability rather than chase the multiple.
Why a multiple alone is the wrong frame
"22× for 30% growth" is a clean bull soundbite, but a P/E ignores the ~44%-of-revenue capex that suppresses cash conversion. The rigorous test is a cash-flow model that charges TSMC for its capital intensity, a DCF. That is next, and its answer is the most important (and least flattering) number in this note.
DCF + Reverse DCF
A semiconductor DCF must charge for capex. We project revenue × EBITDA margin, then subtract capex, cash taxes and working-capital change to get unlevered free cash flow, discount the explicit years, add a Gordon-growth terminal and net cash, and divide by ADR shares. Toggle Base / Bull / Bear / Reverse and drag the WACC and terminal-growth sliders, the per-ADR value recomputes live. The honest result is the point: at reasonable inputs the intrinsic value lands near the price.
Base: revenue ~$160B → ~$314B by FY30 (~21% CAGR), EBITDA margin ~67–69%, capex ~$56–65B. WACC 9.3%, terminal growth 5.0% → a DCF intrinsic ~$425 ≈ the price. Important: that 5.0% terminal is the lever, terminal value is ~86% of the figure, and at a ~3% GDP terminal the intrinsic is ~$300 (see the grid below). So the base is fairly-valued-to-rich, not undervalued. Our $490 12-month PT is a judgment-weighted target that sits deliberately below the ~$540 Street, not a mechanical roll-forward.
Bull: the AI cycle runs hotter (revenue ~$385B by FY30), margins lift toward ~70%, and the Taiwan-risk discount narrows; WACC 9.0%, terminal growth 5.0% → ~$577 intrinsic, supporting the $590 12-month PT.
Bear: AI demand digests (revenue ~$282B by FY30), margins slip toward ~65% on under-utilization, and the discount rate lifts; WACC 9.7%, terminal growth 4.8% → ~$315 intrinsic, below the $355 12-month PT, which assumes a less-severe path.
Reverse DCF: holds base-case cash flows and solves for the terminal growth the current $423.00 price implies at the slider WACC. The output tells you how much AI-cycle durability is already in the price.
DCF Inputs
▶ DCF intrinsic ≈ market, TSM is roughly fairly valued on conservative perpetuity cash flows.
5-Year FCF Forecast ($B)
| ($B) | FY26E | FY27E | FY28E | FY29E | FY30E |
|---|---|---|---|---|---|
| Revenue | 160 | 200 | 242 | 280 | 314 |
| EBITDA margin | 69% | 69% | 68% | 68% | 67% |
| EBITDA | 110 | 138 | 165 | 189 | 210 |
| – Capex | (56) | (62) | (65) | (65) | (63) |
| Unlevered FCF | 40 | 58 | 78 | 98 | 118 |
FCF = (EBITDA − D&A)×(1−16% tax) + D&A − capex − ΔWC, with D&A ≈ 50% of capex. The ~44%-of-revenue capex is what holds the intrinsic near the price. All forecast values FY27E+ are author estimates.
Sensitivity: DCF value / ADR ($) vs WACC × terminal growth
Base-case cash flows across the grid (WACC × terminal growth).
Reverse DCF: what is the market pricing in?
The reverse DCF needs an honest caveat: because we calibrated the base cash flows to land near today's price, inverting for the price-implied terminal mechanically returns ~5%, it echoes the base input rather than being a wholly independent check. The informative read is the gap to a normal terminal: against a GDP-anchored ~2.5–3% perpetual rate, ~$423 embeds roughly two extra points of perpetual growth (and ~21% revenue CAGR, ~68% EBITDA margins), i.e. the price pays for the AI super-cycle being near-permanent. You are not buying a mispriced asset; you are buying a great one at a full price and underwriting that the cycle lasts.
PT calculator (forward-earnings cross-check)
FY27E ADR EPS ~$24.5 × ~20×, the current ~22× multiple easing slightly as earnings compound, lands at our $490 base. The PT does not assume a re-rating; it assumes the earnings keep growing while the multiple drifts modestly lower.
Risk / Reward calculator
A "weak" near-term R/R is consistent with the thesis, not a contradiction: at ~fair value the 12-month risk/reward to the base target is modest. The asymmetry the Buy is paid for sits in the multi-year cycle, the bull ($590, +39%) vs the bear ($355, −16%), not in a tight 12-month trade.
Note: The assistant reasons from the dashboard's data snapshot and thesis sections, it does not browse the web or access real-time fundamentals beyond what's in data.js. Treat its responses as scenario-modeling support, not primary research. Author judgments on rating, PT, and probabilities remain with the analyst.
Upcoming Catalysts
| Catalyst | Window | Why it matters |
|---|---|---|
| Q3 2026 earnings call | ~Oct 15, 2026 | The next print. Q2 (Jul 16) already delivered a record beat (revenue +33.7% YoY) and RAISED FY guidance to "slightly above 40%" with 2026 capex lifted to $60–64B; Q3 (guided US$44.6–45.8B revenue / 65–67% GM) sets the next guidance step and tests the margin path as the 2nm ramp and overseas fabs dilute ~2–3 pts. |
| Monthly sales | ~Sep 10, 2026 (monthly) | Momentum check on the raised "slightly above 40%" FY guide after H1's swing (soft April +17.5%, record May +30.1%); each print reads AI-accelerator demand between the quarterly calls. |
| ITC patent ruling (337-TA-1443) | Final Commission decision ~Oct 2026 | Could in theory bar 7nm-and-smaller imports into a market that is ~75% of revenue, a binary headline risk amplified by a May 22 GOP-lawmaker letter urging an import ban, though a negotiated settlement remains the base case.8 |
| N2P volume / A16 production-ready | 2H 2026 (A16 ramp 2027) | N2P reaches volume production; A16 (Super Power Rail) is production-ready, its customer volume ramp re-slotted to 2027 (Nvidia via Feynman), extending the 2nm-class family as a long-lived, high-ASP node. |
| CoWoS supply-gap milestone | End-2026 | The supply-demand gap narrowing from ~20% to ~10% as capacity expands ~4×, the gate on Nvidia Rubin and broader AI-accelerator shipment volume. |
Risk Factors
- Taiwan single-point-of-failure (the structural tail). The pre-eminent risk and the reason TSM trades at a discount to its fabless customers: >90% of the world's advanced logic is fabricated ~100 miles from mainland China. The risk is un-timeable and, if it materializes, catastrophic (some estimates put a conflict's global cost at ~$10T). The ~$265B US commitment (an additional $100B announced in 2026 on top of the prior $165B Arizona build: 6 fabs + 2 packaging + R&D) and the US-Taiwan Agreement on Reciprocal Trade (signed Feb 2026, a 15% reciprocal-tariff cap, non-stacking, plus formal Section 232 semiconductor "preferential treatment" tied to ≥$250B of Taiwanese US investment) reduce, but cannot eliminate, the concentration, and some argue diversification erodes Taiwan's "silicon shield" deterrence.
- The still-pending ITC patent case. The case in 337-TA-1443 has moved toward the final Commission decision (~October 2026). In theory it could exclude 7nm-and-smaller imports into the US (~75% of revenue), and the headline risk was amplified by a May 22 letter from four GOP lawmakers urging the ITC to impose an import ban. A negotiated settlement remains the base case, but it is a live, near-term binary that has weighed on the ADR through the summer.
- AI-demand digestion. April 2026 sales were soft (+17.5% YoY), reviving a digestion narrative, but May rebounded to a single-month record (+30.1%) and the Q2 print (Jul 16) confirmed the re-acceleration, with FY guidance raised to "slightly above 40%," so the scare has, for now, been rebutted. The residual risk is real: if the hyperscaler capex wave matures, even the raised guide is at risk and the premium multiple compresses.
- Capex intensity & under-utilization. 2026 capex, raised at Q2 to ~$60–64B (the model holds ~$56B, ~44% of revenue), with the next three years "significantly higher," is brutal operating leverage in reverse: if AI demand normalizes, under-utilized ultra-expensive fabs crush margins and ROIC. Overseas fabs already dilute gross margin ~2–3% (widening to 3–4%).
- Customer concentration. Roughly ~19% of revenue is Nvidia and ~17% Apple (2025 estimates), two customers near ~36% of the business, widening to ~22% / ~18% (~40% combined) on 2026 estimates as Nvidia, now the #1 customer, pulls ahead on the Vera Rubin ramp. That displacement reinforces the "toll on AI" thesis, but a roadmap stumble or in-sourcing shift at either would be felt directly.
- The thin margin of safety. Our own DCF says TSM is ~fairly valued today, so unlike a deep-value name there is little embedded pessimism to cushion a disappointment. The Buy depends on the AI cycle lasting.
Scenario Stress Tests
| Scenario | Mechanism | DCF intrinsic | vs base ~$425 |
|---|---|---|---|
| Base | ~21% rev CAGR; ~67–69% EBITDA margin; WACC 9.3%, tg 5% | ~$425 | — |
| Taiwan-risk discount narrows | WACC to 8.5% as the tail de-rates; base cash flows | ~$524 | +23% |
| AI growth halves | Rev CAGR to ~12%; margins held; WACC 9.3% | ~$304 | (28%) |
| Margin under-utilization | EBITDA margin −5pts on a demand air-pocket; base WACC | ~$378 | (11%) |
| Multiple/tail re-rating | WACC to 11% (geopolitical scare); base cash flows | ~$302 | (29%) |
| Full bear | AI digestion + 9.7% WACC + 4.8% terminal on lower rev | ~$315 | (26%) |
| Bull: cycle runs hot | ~$385B FY30 rev, ~70% margin, WACC 9.0%, tg 5.0% | ~$577 | +36% |
All values are DCF intrinsic values today (not the 12-month PTs, which roll forward ~a year of growth), each shock changes only the inputs noted, holding the 5.186B ADR-share count and $64B net cash constant. Deltas are vs the ~$425 base intrinsic.
Bull vs Bear Debate
| Issue | Bull view | Bear view |
|---|---|---|
| Is the DCF a problem? | A DCF on a 20%+ compounder rolls forward ~20%/yr; ~fair value today becomes a clear gain in 12 months, and the comp (~22× vs ASML 47×) is cheap. | The DCF charges for the real ~44%-of-revenue capex and still only gets to the price, there is no margin of safety, only a bet the cycle lasts. |
| AI demand durability | HPC/AI 66% (a record) and a mid-to-high-50s% accelerator CAGR; the Q2 record beat and the raised "slightly above 40%" guide rebutted the April scare. TSM captures demand regardless of which chip wins, a toll, not a bet. | April's soft +17.5% showed the tape can wobble, and the ITC ruling is a live binary; if hyperscaler capex digests, even the raised "slightly above 40%" guide and the premium multiple both crack. |
| The Taiwan tail | Arizona (~$265B) + Japan/Germany diversify; the US has every incentive to defend the supply chain; the discount already reflects the risk. | >90% of advanced logic sits ~100mi from the mainland, an un-hedgeable, un-timeable tail that no DCF can price, and diversification may erode deterrence. |
| Pricing power & margins | 5–10% node price hikes + N2/A16 mix drove record 68% GM / 56% net margin, structural, not cyclical, given the monopoly. | Overseas fabs dilute GM 2–3% (widening), and a demand air-pocket turns the ~44% capex into under-utilization that crushes the same margins. |
| Is ~22× cheap? | Cheapest high-quality AI-compute exposure, below ASML, AVGO, and the ~37× semis median for the best asset in the chain. | ~22× P/E ignores ~57× P/FCF; on cash, not earnings, TSM is not cheap, and it has re-rated from a ~13× trough. |
Technical Analysis
TSM ADR trailing-12-month closes
RSI (multi-timeframe)
Neutral after the pullback from the $479 high, cooled from overbought, neither stretched nor washed out.
MACD vs Signal
Rolled over from the spring highs on the ITC headline + the late-June selloff; momentum cooling but the uptrend intact.
Relative strength (2026 YTD)
Up ~39% YTD, ahead of the Nasdaq-100 (~16%) but far behind the red-hot SOX semis index (~96%, nearly a double): TSM is the steady quality name that lagged 2026's higher-beta AI-hardware surge, consistent with the "fair value, not a momentum darling" read. (Benchmark indices approximate.)
EMA stack (current)
Trader's view
- Uptrend intact but consolidating below the $479 all-time high after a year-long AI re-rating from ~$224, the stock is +39.2% YTD (from a ~$304 year-end-2025 close).
- Key support: the $395 spring breakout shelf; below it, the rising 200-DMA.
- Key resistance: the $479 all-time high; a break needs a clean monthly-sales re-acceleration or a guidance raise.
- Momentum (RSI/MACD) cooling off the highs, the chart is waiting on the July print and the ITC headline to clear.
Glossary & Methodology Notes
- Foundry / leading edge
- A foundry manufactures chips designed by others (fabless customers). "Leading edge" is the most advanced nodes (≤3nm today); it commands the highest margins and is where TSMC's ~90%+ share and AI demand concentrate.
- Process node (N3 / N2 / A16)
- A generation of manufacturing technology. N3 = 3nm-class, N2 = 2nm-class (volume production Q4 2025), A16 = the 1.6nm-class node with Super Power Rail backside power (production-ready 2026, customer volume ramp 2027). Each generation adds transistor density/efficiency and a price premium.
- CoWoS (advanced packaging)
- "Chip-on-Wafer-on-Substrate", the packaging that stitches GPU dies to high-bandwidth memory. It is the binding capacity constraint on AI-accelerator shipments; TSMC is scaling it ~4× into 2026.
- EBITDA vs free cash flow (the capex drag)
- EBITDA is earnings before interest, tax, depreciation and amortization. Free cash flow subtracts capital expenditure. For TSMC, ~44%-of-revenue capex makes FCF far lower than EBITDA implies (P/FCF ~57× vs P/E ~22×), the reason the DCF, which charges capex in full, lands near the price.
- ADR ratio
- The TSM ADR represents 5 TSMC ordinary shares. Per-ADR figures (price, EPS) are 5× the ordinary-share figure, converted at the prevailing NT$/USD rate, which makes USD EPS FX-dependent.
- DCF & Reverse DCF
- A discounted-cash-flow model values a company as the present value of its future free cash flows plus a terminal value. A reverse DCF inverts it: hold a discount rate and solve for the growth the current price implies, telling you how much of the AI cycle is already priced.
- WACC & terminal growth
- WACC (weighted-average cost of capital) is the discount rate; for TSMC it should embed a Taiwan-risk premium. Terminal growth is the perpetual growth after the explicit forecast, capitalized via the Gordon model. For a capital-intensive compounder, both move the value materially.
Methodology
- Snapshot anchor: August 7, 2026. Live ADR price patches via the Cloudflare-Worker quote proxy on page load. 1 ADR = 5 ordinary shares.
- All figures in USD. FY2025 + Q2 2026 are reported; FY2026 is from management's "slightly above 40%" growth guide (raised at Q2), which the DCF models conservatively at ~$160B; FY2027E–FY2030E and all DCF inputs are author estimates, the largest swing factor. ADR EPS is FX-dependent.
- The DCF charges capex in full (the point of using a cash-flow model for a capital-intensive name); the 12-month PT rolls the intrinsic value forward ~one year of growth.
- Conclusions are the author's view. Illustrative, not investment advice.
Sources & Citations
Inline citations
Superscripted numbers in the body link here. Click any N in the report to jump back to the source.
- TSMC, Second Quarter 2026 Results (reported July 16, 2026): revenue +33.7% YoY (US$40.20B; NT$1,270.38B, +36% YoY, a record), gross margin 67.7%, operating margin 60.3%, net margin 55.6% (all records); net income NT$706.56B (+77% YoY); diluted EPS US$4.31/ADR (NT$27.25). FY2026 USD revenue growth guidance raised to "slightly above 40%" (from "above 30%"); 2026 capex raised to US$60–64B; Q3 2026 guided US$44.6–45.8B revenue / 65–67% GM. TSMC Investor Relations. ↩
- TSMC Q2 2026 platform & technology mix: HPC/AI 66% of revenue (record, ~3× smartphone), Smartphone 22%, IoT 5%, Automotive 4%; by wafer revenue, advanced ≤7nm = 77% (N5 33%, N3 30%, N7 11%, N2 3%); geography North America 76%. Leading-edge share >90%; ~72% overall foundry share. Per the Q2 2026 results / investor presentation. ↩
- TSMC 2026 capex raised at Q2 to ~US$60–64B (from ~$56B; the next three years "significantly higher"); the DCF models the prior ~$56B base. CoWoS capacity scaling ~4× to ~120–140k wafers/month; N2 (2nm) volume production from Q4 2025 with five fabs ramping in 2026 and N2P/A16 in 2H 2026; overseas fabs dilute gross margin ~2–3% (widening to 3–4%). Per management guidance on the Q1 / Q2 2026 calls. ↩
- TSMC FY2020–FY2025 revenue and margins per annual results: revenue US$45.5B → US$122.4B; FY2026E ~$160B is our conservative modeling base, below management's raised "slightly above 40%" USD growth guide (~$172B implied; TSMC gives a % range, not a $ target), an estimate. ↩
- TSMC May 2026 monthly revenue: NT$416.98B, +30.1% YoY / +1.5% MoM (~US$13.25B), a single-month record re-accelerating from a soft April (+17.5% YoY); January–May 2026 cumulative revenue NT$1,961.80B, +30.0% YoY. Per TSMC's monthly sales release / SEC Form 6-K (June 10, 2026). TSMC monthly sales. ↩
- Foundry market share, Q1 2026 (TrendForce): TSMC 72.3% (up from ~70.4% in Q4 2025), Samsung 6.5%, SMIC 5.1%, TSMC roughly 11× Samsung's foundry revenue. Samsung SF2 (2nm) yields reported in the mid-50s% vs TSMC N2 at ~60–70%; Intel Foundry external revenue ~$174M in Q1 2026. ↩
- Analyst actions: BofA raised its TSM target to $590 from $490 (Buy reiterated, June 24, 2026); Susquehanna raised to $575 from $500 (Positive, June 22, 2026); estimates and targets moved higher again after the July 16 Q2 beat. Consensus average target ~$540 (Strong Buy) per stockanalysis.com; zero downgrades. ↩
- US ITC investigation 337-TA-1443: the case has moved toward the final Commission decision (expected ~October 2026). On May 22, 2026 four Republican lawmakers wrote to the ITC urging an import ban on infringing TSMC chips (7nm-and-smaller); a negotiated settlement remains the base case. The overhang weighed on the TSM ADR through the summer. Per Taipei Times / IPWatchdog / market reports and the USITC docket. ↩
Background reading
- TSMC 2025 Annual Report (Form 20-F), financials, segment/geographic disclosures, risk factors.
- TSMC quarterly results + monthly revenue releases (TSMC reports monthly sales), revenue, margins, platform/node mix, capex.
- TSMC Q2 2026 earnings call + investor presentation, the raised "slightly above 40%" guide, the $60–64B capex raise, AI-accelerator CAGR, CoWoS/N2 roadmap, Arizona commentary (an additional $100B US investment).
- US ITC investigation 337-TA-1443, the 7nm-and-smaller patent case and its preliminary-ruling timeline.
- Peer disclosures (NVDA, AVGO, ASML), the forward-P/E comparison and the AI-supply-chain context.
- US-Taiwan Agreement on Reciprocal Trade (MOU Jan 2026, signed Feb 2026), the 15% reciprocal-tariff cap, Section 232 semiconductor "preferential treatment," and the ≥$250B Taiwanese US-investment pledge.
- TSMC April 2026 Technology Symposium, the N2 / N2P / A16 ramp timing and the N2U / A14 (2028) and A13 / A12 (2029) roadmap (TrendForce, Tom's Hardware coverage).
- TrendForce foundry-share data (Q1 2026); Nvidia GTC Taipei (June 2026, Vera Rubin); the OpenAI/Broadcom "Jalapeño" ASIC announcement (June 24, 2026), competitive and design-win context.
- Market data: stockanalysis.com / finviz (ADR price, shares, market cap, 52-week range, +39.2% YTD, forward multiples) as of August 7, 2026.
Disclaimer. This report is the author's institutional equity-research view, prepared for portfolio and educational purposes. It is not a recommendation to buy, sell, or hold any security. Forward-looking statements are subject to risk and uncertainty; past performance is not indicative of future results. Consult a licensed financial advisor before making investment decisions. All third-party trademarks are the property of their respective owners.