TSMC
Comparable Company Analysis

Summary

Comps are the wrong tool for TSMC. The four multiples I tested disagree wildly, from about 53% downside to 129% upside.

EV/EBIT is the better fit in theory, but the only multiple I’d put any weight on is EV/EBITDA, and it implies TSMC is worth 15% to 26% less than its current price. That gap has a simple explanation: TSMC earns far higher margins than any of its peers, and the peer group is too different to value it fairly. So the useful output of this exercise is less a number than a set of reasons why comparable companies are the wrong tool here.

How I Set It Up

I compared TSMC (NYSE: TSM) with five listed companies that make chips for others, to varying degrees: SMIC, UMC, GlobalFoundries, Samsung Electronics and Intel. Every company is on the same footing: financials are for the trailing twelve months to about 30 June 2026, and net debt is taken from the Q2 FY2026 balance sheets (Samsung converted from KRW to USD, TSMC at the most recent spot). All figures are in US$ millions unless stated.

For each company I calculated EV/Sales, EV/EBITDA, EV/EBIT and P/E, plus net debt/EBITDA as a check on leverage. I then applied the peer mean, the peer median and a weighted multiple to TSMC’s own numbers to see what price each one implied.

Table 1. Market Data (US$ Millions, Price in US$)

Table 2. Trailing Twelve Month Financials (US$ Millions)

The five peers

• UMC. A pure-play foundry with the same business model as TSMC, just at mature and mainstream nodes instead of the leading edge. The closest match in the set.
• GlobalFoundries. Also a pure-play foundry, but smaller and more focused on specialty nodes (radio frequency, automotive) than UMC’s broader mainstream base.
• Samsung Electronics. A genuine leading-edge foundry competitor, but foundry is a minority of a far larger consumer electronics, memory and mobile conglomerate, which dilutes the comparison.
• Intel. Owns and runs its own fabs, but Intel Foundry is a new and largely unproven external-customer business. Intel is still mainly a chip designer that manufactures for itself, not a foundry like TSMC.
• SMIC. A pure-play foundry and, structurally, the most direct comparison in the group: foundry only, with no consumer, memory or mobile business to distract from it. It is also China’s flagship foundry.

Why I Started With EV/EBIT

My team and I initially believed EV/EBIT would be the most suitable multiple for a PPE-intensive company like TSMC. Its capex announcements keep growing, and its ventures across Arizona, Taiwan, Germany and Japan keep expanding. EV/EBIT is struck after depreciation and amortisation, so it captures the cost of the machinery, fabrication sites and other fixed assets that this capex builds up. EV/EBITDA leaves that cost out, which flatters a business that has to keep reinvesting just to stay at the leading edge.

The numbers support the premise. TSMC’s net property, plant and equipment grew from NT$1,555.6bn in 2020 to NT$3,691.8bn in 2025, up 137% and higher in every single year. Growth slowed to 5.6% in 2024, then picked back up to 14.1% in 2025 as spending stepped up again.

The 2026 capex budget has been raised to US$60–64bn, against US$40–42bn in 2025, with roughly 70% going to advanced processes. TSMC’s D&A over the last twelve months was about US$20.8bn (EBITDA of US$99.6bn less EBIT of US$78.8bn), or around 21% of EBITDA. As the new fabs come online and start depreciating, that share is likely to rise, which is exactly the effect EV/EBIT picks up and EV/EBITDA ignores.

Where the capex is going

The build-out spans Taiwan, Arizona, Japan and Germany, and the timeline runs out to 2028 and beyond. Some projects have no dates in the sources I used (Kumamoto Fab 2 in Japan, Dresden in Germany, the extra Arizona fabs), so they’re covered in the appendix rather than on the timeline.

Table 3. Selected TSMC Fab and Packaging Milestones by Year

Either way, I decided to look at EV/EBIT, and to put a weighted average multiple on top of it, to get a better sense of where TSMC sits across its competition.

What the Multiples Say

Table 4. Valuation Multiples, Peer Statistics and TSMC

Table 5. Implied TSMC Valuation by Method (equity value = implied enterprise value plus TSMC’s US$77.0bn net cash; 4,927.1m shares)

EV/Sales: TSMC looks overvalued, but the multiple is the wrong lens

TSMC trades at 14.8x sales against a peer mean of 6.65x and a median of 6.62x, which implies about 53% downside. But a sales multiple ignores profitability, and that is TSMC’s whole edge. It turns about 71% of sales into EBITDA, while the peers convert between 28% and 51%. Applying peer sales multiples to TSMC is bound to understate it.

EV/EBITDA: the most reasonable read

At 21.0x EBITDA TSMC trades above the peer mean (17.7x) and median (15.2x), with peers ranging from 8.4x (Samsung) to 34.1x (Intel). That implies -14.9% downside on the mean, -26.3% on the median and -20.2% on the weighted multiple. It is the tightest and most believable of the four results, and it points to a premium that’s easy to explain by margins alone.

EV/EBIT: the distorted one

On paper EV/EBIT looks dramatically undervalued: the peer mean of 61.8x implies about 129% upside. It isn’t a real signal. The peer multiples range from 9.2x (Samsung) to 128.7x (Intel), and the extremes come from SMIC and Intel, whose D&A is 82% and 74% of EBITDA respectively. Both are mid-capex-cycle with very thin EBIT (margins of 9.0% and 7.8%), which inflates their EV/EBIT multiples. That distortion then feeds straight into the mean and the weighted multiple when applied to TSMC. The median (33.6x, +25.9%) is far less affected.

P/E: close to meaningless

Intel made a net loss of US$11.3bn over the period, so its P/E isn’t meaningful and I excluded it. SMIC is on 92.5x, which pulls the peer mean to 39.9x against a median of 27.8x. The implied move flips sign depending on the method: −9.7% on the median, +29.4% on the mean and +33.8% on the weighted multiple. A multiple that can’t agree on the direction isn’t telling me much.

Leverage is not the story

Net debt/EBITDA is negative (net cash) for TSMC and for UMC, GlobalFoundries and Samsung, and positive only for Intel (1.2x) and SMIC (1.1x). Most of this peer set is financially conservative, so the valuation gaps come from margins and multiples, not from leverage.

The Weighted Multiple

The weighted multiple was highly subjective, which in the end was part of the reason it felt unusable next to EV/EBIT and EV/EBITDA. To get the weights, I built a Google Form for my team, accessible here, and also asked a range of AI chatbots to rate each peer on how similar it is to TSMC, out of 10. I gave each respondent the market cap and a short description of every company to work from.

For each respondent I turned the ratings into percentages of that respondent’s total, then averaged those percentages across all nine respondents (me, seven chatbots and one team member) to get one weight per company.

Table 6. Similarity Ratings Out of 10, by Respondent

Applied to the multiples, the weights give 6.6x EV/Sales, 16.6x EV/EBITDA, 58.0x EV/EBIT and 42.4x P/E (with Intel left out of the P/E and the remaining weights rescaled).

I couldn’t rely on it, for four reasons:

• Little to go on. Respondents only had market cap and a short description, so a rating out of 10 is close to a gut feel.
• Not much is known about the peers. Beyond market cap there was little to separate them, and AI chatbots aren’t industry judges.
• A small, uneven sample. Nine ratings, most of them from chatbots, is not a robust way to set weights.
• It amplifies the problem. SMIC gets 24% of the weight, so its 109x EV/EBIT flows straight into the weighted result.

Why Comps Fall Short for TSMC

Between the volatile results above and the nature of TSMC’s business compared with its competition, I concluded that a comparables valuation isn’t suitable for a company like TSMC, for two main reasons.

• It holds the largest market share in its industry. The five peers together hold just 18.7% market share by the figures in my rating brief, and none of them is a like-for-like rival at TSMC’s scale.
• Its competitors don’t replicate what it does. Companies like Intel and Samsung source only a percentage of their revenue from foundry operations. Compare that with the mobile device market, where Apple, Samsung, Huawei and others all sell essentially the same product in their own way, so their multiples are genuinely comparable. There is no equivalent group for TSMC.

The margins tell the same story. TSMC turns 56% of sales into EBIT, against 8% to 37% for its peers, and it holds US$77bn of net cash while Intel carries net debt.

Conclusion: the multiple that fits versus the one I trust

Appendix: TSMC Fab Footprint

Table 7. Operating, Under Construction and Planned Capacity by Region

Capex figures: 2026 budget of US$60–64bn against US$40–42bn in 2025, roughly 70% to advanced processes. Fab counts are per TSMC’s FY25 Form 20-F. Timelines are as reported by the sources below and may change.

References

1. 1. Yahoo Finance, TSMC 2026 capex
2. 2. DIGITIMES, TSMC Arizona, market and Germany in 2026
3. 3. TSMC FY25 Form 20-F (SEC)
4. 4. TrendForce, TSMC speeds up expansion in Taiwan (23 Feb 2026)
5. 5. TrendForce, TSMC boosts Arizona investment by US$100bn (16 Jul 2026)
6. 6. Taipei Times, TSMC’s Kumamoto fab swings to profit in Q1 (18 May 2026)
7. 7. Taipei Times, new fab and 2nm expansion in Tainan (5 Mar 2026)
8. 8. New Electronics, TSMC accelerates fab construction
9. 9. TechPowerUp, TSMC to build three new 2nm fabs in Taiwan

Financial data for TSMC and the five peers, the similarity ratings and the PPE table are from the author’s own spreadsheet. Fab and capex information and links above were provided by the author.

Views expressed are for learning and discussion. They are not investment advice.

Add a Thought

Something I got wrong, could have done better, or a view I haven’t considered? I’d like to hear it, and every note gets read.