Nvidia vs. TSMC Stock: Which AI Investment Fits You?

Buying Nvidia and TSMC gives an investor two businesses, but it does not create two independent bets on artificial intelligence. One sells computing systems; the other manufactures chips for customers across the industry. If AI investment slows broadly, changing the name on the share certificate does not remove the common source of demand.

That is the useful starting point for a Nvidia vs. TSMC stock comparison. TSMC offers a way to participate without requiring one chip designer to keep winning. Nvidia offers more direct exposure to the economics of its computing platform. The choice depends on which risk an investor wants to own, and which risk they mistakenly think they have escaped.

There is also a financial surprise. Calculated from their latest reported quarters, the companies produced free-cash-flow margins of roughly 22% each. Those similar outcomes conceal very different demands on cash. Treating them as interchangeable would miss the investment decision.

Nvidia vs. TSMC: different businesses, connected demand

Nvidia’s latest results reported $96.221 billion in revenue for the quarter ended July 26, 2026. Data Center contributed approximately $89 billion, or 92.5% of the total by calculation. That concentration makes the company particularly sensitive to the scale and economics of computing deployments.

TSMC’s second-quarter earnings release describes a different model: a dedicated foundry that manufactured products for 534 customers in 2025. Its June-quarter revenue was $40.20 billion, up 33.7% in U.S. dollars from a year earlier. The reported growth rate in New Taiwan dollars was 36.0%; the currencies should not be mixed when ranking growth.

A broad customer roster helps answer the question of who wins a chip-design contract. It does less to answer what happens if customers collectively revise their spending plans.

Consider a hypothetical shift from one accelerator supplier to another. A foundry that manufactures the replacement design may retain work even when the original designer loses the order. Now consider a canceled data center instead. No supplier wins that replacement order because there is no order to replace.

Those are different investment risks. TSMC can offer some protection against choosing the wrong designer without offering protection against an industry-wide reduction in spending. The first proposition is plausible; the second requires evidence that a customer count alone cannot supply.

TSMC’s 66% HPC share qualifies the diversification case

TSMC’s July earnings-call transcript puts high-performance computing at 66% of second-quarter revenue and smartphones at 22%. Management also said its capacity planning incorporates discussions with customers and their customers, including cloud-service providers.

HPC is not a standalone AI-revenue disclosure. It should not be placed beside Nvidia’s Data Center percentage and treated as an identical segment definition. The numbers instead show that TSMC’s broader product mix still contains a large computing exposure.

For portfolio construction, that distinction matters more than a simple “diversified versus concentrated” label. An investor already holding several AI infrastructure companies should look through the holdings to the spending decisions behind them. Separate tickers can depend on the same projects being built on time.

There is no basis here for a numerical claim about how much portfolio volatility an NVDA–TSM combination removes. That would require return data, a defined measurement window and a model of correlations. Business exposure is the subject of this comparison, not a backtested claim that either share is a hedge.

The practical question is narrower: would the reason for buying TSMC still hold if computing investment disappointed? If the answer relies entirely on AI capacity staying full, the investment remains exposed to that spending cycle.

A 22% cash-flow margin does not mean the same thing

TECHi’s Nvidia financial statements page highlights the gap between quarterly operating cash flow and reported profit. It shows approximately $24.08 billion of operating cash generation against $59.69 billion of net income for the latest quarter. That prompted a closer comparison of cash retained after investment, rather than another ranking by headline growth.

The company’s release reports $21.341 billion of free cash flow after subtracting purchases and principal payments related to property, equipment and intangible assets. Dividing by quarterly revenue gives a 22.2% free-cash-flow margin.

TSMC’s quarterly presentation reports NT$783.36 billion of operating cash flow and NT$496.00 billion of capital expenditure. Its resulting NT$287.36 billion of free cash flow, divided by NT$1,270.38 billion of revenue, gives 22.6%.

The capital-spending comparison is more revealing:

• Nvidia: Property, equipment and intangible purchases equaled about 2.8% of quarterly revenue, before the additional principal payments used in its free-cash-flow calculation.
• TSMC: Capital expenditure equaled about 39.0% of quarterly revenue.
• Period limitation: Nvidia’s quarter ended July 26; TSMC’s ended June 30. These are the latest reported periods, not matching calendar windows.
• Accounting limitation: Free cash flow is a company-defined measure. Its components should be checked before treating the percentages as perfectly comparable.

A manufacturer can spend heavily to create future productive capacity. A designer can report relatively little direct capital expenditure while committing substantial resources elsewhere. Looking only at the final cash margin misses both mechanisms.

The figures are useful precisely because they resist an easy winner. They do not establish that Nvidia’s current cash conversion is permanent, or that TSMC’s investment will earn an attractive return. They establish the next research questions: how collections develop at the designer and how productive the manufacturer’s expanding asset base becomes.

Nvidia’s low direct capex is not a complete risk measure

Nvidia’s July Form 10-Q disclosed $279 billion in supply commitments, primarily associated with memory and manufacturing facilities for its data-center infrastructure systems. The company also identifies export restrictions and changes in customer demand as business risks.

Supply commitments are not the same thing as capital expenditure already paid. They are not a number to subtract mechanically from one quarter’s free cash flow. Their relevance is that low spending on owned equipment does not mean the business can adjust every future cost immediately.

This changes how the comparison should be framed. TSMC visibly commits cash to manufacturing capacity; Nvidia can carry obligations through its supply arrangements. One route is easier to see in a capital-expenditure row, but both require demand assumptions.

An investor should therefore resist awarding Nvidia an automatic valuation premium simply because its direct capex ratio is lower. The question is whether the total obligations behind growth remain proportionate to economically durable demand.

The same discipline applies to TSMC. A factory investment can strengthen the business if customers use it at attractive prices. It can burden returns if capacity arrives ahead of demand. Spending itself proves neither outcome.

Valuation needs consistent periods and earnings definitions

At the September 18, 2026 U.S. regular-session close, TECHi’s integrated quote service returned Yahoo Finance fallback prices of $222.27 for NVDA and $434.67 for TSM. The respective timestamps were 20:00:00 and 20:00:02 UTC. U.S. stock markets were closed for the weekend when this analysis was prepared on September 19.

These are closing snapshots, not live weekend prices. The higher price per share does not make TSMC the more expensive company. Share counts, earnings and the security being purchased determine what a valuation ratio actually measures.

The NVDA quote page and TSM quote page provide the starting market context. For a purchase decision, match the price with earnings or cash flow from a consistent period and check the source timestamp. A forward multiple for one company and a trailing multiple for the other are answers to different questions.

Annualizing the latest quarter also creates a trap here. Multiplying a single cash-flow figure by four assumes that its collection patterns and investment schedule represent the year. The cash-margin comparison above deliberately does not make that assumption.

A hypothetical example illustrates the separate valuation risk. If earnings per share rise 20% while the earnings multiple falls 20%, the implied share price falls 4%: 1.20 multiplied by 0.80 equals 0.96. This is arithmetic, not a forecast for either stock.

A strong operating thesis still needs a purchase price that leaves room for disappointment. That is especially relevant when an investor’s case depends on several favorable assumptions arriving together: rising deployment, sustained pricing and timely capacity expansion.

Which stock fits the investment thesis?

TSMC is the more direct fit for an investor whose conviction is that advanced chip manufacturing will remain valuable while the winning designers change. That thesis requires confidence in manufacturing execution, utilization and the returns earned on new capacity. It does not require pretending the company is insulated from AI spending.

Nvidia fits an investor who specifically wants exposure to its computing platform and is willing to assess the associated concentration, supply and policy risks. Its latest growth is impressive, but extrapolating revenue without examining cash collection and obligations would leave the analysis unfinished.

Holding both can be reasonable if the objective is to spread exposure across different business models. It is a weaker argument if the objective is simply to reduce dependence on AI infrastructure investment. The distinction should be written into the investment thesis before deciding position size.

The next earnings releases should be used to revisit the comparison, rather than preserve a permanent winner. For Nvidia, track operating cash flow alongside revenue and commitments. For TSMC, follow capacity spending, utilization commentary and the cash retained after investment. Changes in those relationships are more informative than a one-day price ranking.

TSMC offers broader exposure to who designs the chips. Nvidia offers the more concentrated platform bet. The shared risk is the scale of spending that ultimately supports both. An investor choosing between them should start with that distinction, then decide what price adequately compensates for the risks left over.