Analysis
Business and Industry
TSMC sits between a chip design and a working product. Customers need factories that can reliably turn designs into usable chips, plus packaging that lets those chips work together. Once an advanced design is built around a manufacturing process, moving it elsewhere can take substantial work. That gives TSMC bargaining strength when capacity is scarce. Customers can still choose a different supplier for their next design. [S01–S05]
TSMC can benefit across different chip designs, but several designers may rely on the same cloud operators' budgets. Its manufacturing advantage does not make that funding independent or durable.
Competitors could change the balance, but their announcements do not yet settle the economics. Samsung's advanced-process orders do not establish equal yields. Most of Intel's reported foundry revenue comes from its own operations, so that revenue is not a measure of equivalent outside demand. ASE's planned packaging capacity may offer alternatives for some work while complementing wafer manufacturing. The competitor detail explains what the reported figures do and do not show. [S12–S14]
Packaging can relieve a production bottleneck. We do not add a separate packaging value on top of TSMC's earnings: standalone profit and invested capital are not sufficiently disclosed to support it.
Investment Thesis
The spending is reaching TSMC: January–August 2026 revenue rose 39.3% from a year earlier. That establishes sales, but not cash collection or the profit those sales will ultimately produce. NVIDIA and Broadcom's growth and outlooks reinforce demand without establishing the returns earned by users of the equipment. [S05–S08; Evidence: Operating context]
The harder question is who can keep paying. Meta's quarterly operating cash was $31.86 billion, against $31.08 billion of investment including finance-lease principal. Investment consumed nearly all of that measure of operating cash. Amazon reported negative trailing free cash flow despite strong operating cash and cloud growth. Microsoft had more room between operating cash and cash equipment spending, but that comparison excludes some leases and other claims. These periods and definitions differ; they cannot be added into one funding total. [S09–S11]
Well-funded expansion would support TSMC's growth; disappointing returns on customer spending could slow several buyers at once. These figures do not prove the buildout will fail, establish a bubble or measure demand pulled forward. Nor do they provide a complete comparison of cloud operators' funding.
Our model gives more weight to durable or accelerating profitable growth than to setbacks. Strong demand, switching costs and scarce capacity support that judgment; customer funding and investment costs limit it. The exact probabilities are subjective interpretations of the dated evidence, not measured frequencies or new operating facts. The decision probabilities show those beliefs; the forward valuation preserves the operating assumptions and separate unweighted cases.
Financial Quality and Funding
TSMC must keep investing to remain competitive. Refusing necessary expansion could damage future earnings, so we deduct all cash equipment spending when asking what remains for shareholders.
Central-case cash after investment is about 2.3% of the company's equity value—the total market value of its shares—at the reference price. Much of the purchase therefore depends on investment producing future value. This is not a promised distribution or a probability-weighted cash forecast. The spending assumptions cover October 2026–September 2027 and are ours, not management's calendar-2027 budget. [Evidence: Cash after investment]
Reported profit is stronger than that cash figure might suggest, but profit needs interpretation too. The latest quarterly operating margin—operating profit as a percentage of sales—reached 60.3%; the earlier contraction shows why we should not assume a particularly good quarter lasts. Disposal gains, tax adjustments and earthquake insurance also complicate recurring earnings. Our forward model does not project the disposal gain or earthquake recovery. Payment timing can flatter cash as well: delaying supplier or tax payments leaves more cash temporarily without improving the underlying business. The earnings adjustments, financial history and cash reconciliation retain those distinctions. [S01–S05, S25, S26]
Some overseas entities already earn profits, but those profits do not establish unsubsidized investment returns: product mix, transfer pricing, subsidies and factory ramps matter. Management expects overseas expansion to dilute margins. We retain those costs without deducting ramp effects twice. [S03–S05; Evidence: Operating context]
TSMC's June cash and liquid securities substantially exceeded financial debt. The concern is the return on spending, not an imminent inability to fund operations. That balance does not prove all cash is distributable or establish precise covenant headroom. Governance provides no shortcut either: the combined chair/CEO roles, ownership and incentives do not by themselves prove poor alignment or good capital allocation. The funding and governance detail keeps the evidence and its limits visible. [S01–S03]
Valuation and Alternatives
At $451.05, buyers pay about 22.77 times modeled forward earnings: $22.77 in price for each $1 of earnings per ADR modeled for October 2026–September 2027. Those earnings are assumed, not guaranteed. The twelve-month return depends on the earnings investors expect a year from now and what they will pay for them. Current orders cannot settle either number. [Evidence: Forward valuation]
Using the same economic outcomes for both holdings gives weighted return estimates of 10.57% for TSMC and 6.75% for SPY. The 3.82-percentage-point advantage supports considering a purchase. These averages include possible large losses; they are not promised returns. The complete outcome table on the Evidence page explains the subjective earnings ranges behind them.
| Scenario | Subjective probability | TSMC average return | SPY average return |
|---|---|---|---|
| Severe ordinary setback | 8% | −39.34% | −22.98% |
| Soft growth and weaker conversion of sales into earnings | 25% | −8.80% | −2.34% |
| Durable growth | 40% | +15.52% | +9.00% |
| Accelerating profitable growth | 25% | +45.24% | +25.12% |
| Catastrophic Taiwan or ADR impairment | 2% | −80.00% | −35.00% |
Returns are rounded to two decimal places. Ordinary-scenario figures are averages within each scenario, where results can be better or worse; the catastrophic scenario uses representative impairment returns. Scenarios with negative average returns carry 35% of our probability weight. That does not mean a 35% chance of losing money: the averages do not measure how often losses occur within each scenario.
A 5% reduction in valuation multiples applied only to TSMC turns that edge into a 1.65-point disadvantage. Investors could simply pay less for TSMC's particular risks, without an industry collapse. Weaker earnings assumptions can also leave little advantage. With a market-wide valuation decline, TSMC can still outperform SPY while both lose money. The sensitivity tests separate these cases.
A longer-term cash valuation adds a reason for restraint. Its central result is about $280 per ADR, below the reference price; a stronger durable-growth case reaches about $414. These are sensitive to distant growth, margins and reinvestment assumptions, and are not twelve-month targets. We give them limited weight in this rating, but they show how much work future cash generation must do. Neither this model nor the first-year cash yield has been turned into a new probability-weighted cash forecast. [Evidence: Long-term cash value]
SPY reduces direct exposure to one company's execution, but still has substantial technology exposure and shares AI and Taiwan supply-chain risks. Dated Treasury and insured-deposit references offer another comparison for someone who does not want equity risk, subject to their own price, eligibility, tax and withdrawal conditions. The alternatives keep those terms separate from modeled stock returns.
Risks and Countercase
The case for a lasting buildout rests on strong demand and limited manufacturing alternatives. Well-funded spending and productive new capacity would let TSMC turn that position into further earnings growth—a possibility we give substantial weight.
The reason to wait is that today's orders do not establish tomorrow's customer profits or factory returns, while depreciation—the expense for equipment in use—continues to rise. Customer cancellation terms, eventual AI monetization, standalone packaging returns and unsubsidized overseas returns remain insufficiently observable. IMF discussion of leverage and circular investment commitments suggests disappointments could be amplified; it does not forecast a TSMC downturn. [S01–S05, S19]
If several customers cut chip orders at once as AI spending slows, TSMC could collect less revenue while still paying for its factories and committed expansion. Earnings could fall further than our main estimate assumes. An independent reviewer uses a more severe earnings estimate for that kind of setback. With other assumptions unchanged, it leaves a smaller modeled advantage over SPY. That reduces our confidence in the buying case. The comparison depends on judgments about future profits and what investors will pay for them; it does not establish the precise chance of losing money or beating SPY. Our cash estimates also use sales periods that do not line up exactly. The supporting evidence explains these limits and the reviewer's alternative. [Evidence: Decision probabilities; Evidence: Probability sensitivities; Evidence: Cash after investment]
Taiwan can overwhelm the ordinary business story. Sustained disruption to production or access could prevent an ADR holder from realizing a quoted value; overseas expansion does not yet reproduce Taiwan's production ecosystem. Our 2% subjective catastrophic weight uses representative losses of 80% for TSMC and 35% for SPY. It is not an observed annual frequency, and those losses are neither maximum losses nor guaranteed executable prices. Temporary ordinary outages remain in the other outcomes. [Evidence: Probability sensitivities]
Export restrictions, subsidy conditions and clawbacks, energy and labor costs, and a stronger Taiwan dollar can also pressure returns while investment remains committed. A reported termination of one Marlin ITC investigation does not resolve every legal matter. Strong demand does not remove these constraints. [S01, S03, S17, S18]
What Would Change Our View
Well-funded demand and new factories producing recurring profit and cash would strengthen the buying case. Persistent customer weakness, margin erosion or poor capacity returns would favor reducing exposure. A price change does not automatically change the rating or trigger a trade.
Four fixed tests for January–June 2027 give that question observable terms: sales check whether activity deteriorates; recurring margin checks what the sales earn; cash after investment checks what remains; and equipment spending relative to sales checks how much expansion consumes. The test table preserves the exact thresholds, definitions and treatment of missing data. None is observable at this cutoff.
Passing would not settle the thesis. The sales threshold mainly catches deterioration, and supplier-payment or customer-receipt timing can flatter cash. The original cash test and the corrected diagnostic remain side by side, including the separately disclosed lease prepayment; neither silently replaces the other. The market will also be looking beyond the test period to the following year's earnings. These are terms for evaluating the assessment, not a promise of ongoing monitoring. [Evidence: Prices and return rules]
Rating Rationale
6 — Consider buying reflects a modest preference at the reference price and horizon. Strong demand and difficult-to-replace manufacturing support our estimated advantage over SPY, but customer spending and TSMC's investment returns limit conviction. The rating does not determine position size or authorize a trade.
5 — Consider other opportunities is credible for someone seeking stronger evidence of durable cash returns. Weaker earnings or a lower TSMC-specific valuation can erase the edge, and substantial value depends on productive reinvestment.
7 — Buy would require more confidence than this evidence supports. Plausible ordinary changes can reverse the preference, and customer funding, overseas returns, model uncertainty and Taiwan risk remain unresolved. Outperforming SPY would not prevent an absolute loss.
Public disclosures
The Publisher reports no holdings or partnerships relevant to this report.
Prepared with AI Assistance.