Two-thirds of every revenue dollar is gross profit. Revenue is growing over 40%. The balance sheet holds net cash. And Warren Buffett bought it in 2022 and sold within months.
Not because he changed his mind about the business — he has been complimentary about it throughout. He sold because of where it is.
That distinction is worth sitting with, because it describes a category of risk that almost every framework you have been taught handles badly.
Two kinds of risk, and only one responds to diversification
| Risk you can diversify | Risk you cannot | |
|---|---|---|
| Examples | A product recall, a lost contract, a bad quarter, a lawsuit | A strait, a currency, an export rule, a regime |
| Are they independent? | Yes — one company's problem says nothing about another's | No — one event hits many holdings at once |
| Does analysis help? | Yes. Read the filings and you can estimate it | No. No amount of accounting reveals the probability |
| What reduces it | Owning more things | Owning less of the thing |
Here is why "just diversify" fails specifically here. If you respond to concentration in one chipmaker by buying more technology companies, you may have increased your exposure, because most of them depend on the same factories. Our Nvidia reports have returned a verdict of "too hard" twice for related reasons.
And it runs the other way too. Our report on the sole maker of the machines that make those chips found that China was about a third of last year's sales and is being cut back, with American lawmakers pushing further. Different company, different continent, same category of risk — decided by governments, priced by nobody.
Diversification answers 'what if I am wrong about this company?' It has nothing to say about 'what if something happens that has nothing to do with the company?'
Sizing, and what it actually buys you
The same holding, the same catastrophe, five times the damage
| Position size | If it fell 90% | You would then need |
|---|---|---|
| 2% | −1.8 points of the portfolio | +1.8% on the rest to be level |
| 5% | −4.5 points | +4.7% |
| 10% | −9.0 points | +9.9% on the rest |
| 20% | −18.0 points | +22.0% |
Read the first and last rows. Identical judgement about the company. Identical outcome. One is an annoyance, the other reshapes the next five years of your results.
This is why our own verdict was "accumulate — but size the position for the Strait". The first half is about the business; the second half is the only sentence that actually protects you, and it is a decision you make once, in advance, when nothing is happening.
And one more thing the screen does not show
The American depositary receipt trades at roughly 15% above the shares in Taipei. That is not a view about the company — it is a gap between two markets.
What it means for you is concrete: the shares have to rise about 15% before you have matched the entry price of someone who bought locally. A permanent haircut, taken at the start, on a holding you were already going to size conservatively.
Worth knowing before you buy, and it appears on no screener we have used.
Three questions, answered before you buy
For every holding, write the one sentence that describes what could go wrong that no amount of reading would have warned you about. Most businesses do not have one. The ones that do should be treated differently from the ones that do not.
If the answer is 'a third of the portfolio', you are not diversified in the way you think. Correlation in a crisis is what matters, and holdings that look unrelated in normal times frequently are not.
Sizing decided calmly is protection. Sizing decided after a 40% rise is just a rationalisation of what already happened — and that is exactly when these positions quietly become too large.
None of this is an argument against owning it. We scored the business 8.0 and called it one of the finest on earth — 67.7% gross margins, 77% of wafer sales on nodes with few rivals, a $265 billion American build-out under way, and a manufacturing lead that rivals spending tens of billions have not closed.
The argument is that "how much" is a separate decision from "whether", and that for a small number of holdings it is the more important of the two.
The finest factory on earth, the only company that equips it, and one we have twice declined to judge.
Including why we agree with Buffett's admiration and share his worry.
Frequently asked
Can diversification remove geopolitical risk?
Not this kind. Diversification works on risks that are independent of each other — one company's recall has nothing to do with another's lawsuit. A risk concentrated in one location, which also happens to make most of the world's advanced chips, is correlated with a great deal of what else you own. Adding more technology holdings can increase your exposure rather than reduce it.
Why did Warren Buffett sell TSMC so quickly?
He bought in 2022 and sold within months, and said the reason was where the company is rather than what it is — he has been complimentary about the business throughout. It is a clear example of a risk that cannot be underwritten by analysis: no amount of reading the accounts tells you anything about the probability, so the only response available is how much you own.
What is position sizing and why does it matter more than being right?
It is the decision about how much of your portfolio one holding gets, and it is the only defence against a risk you cannot analyse. If a holding lost 90%, a 2% position would cost 1.8 points of your portfolio and a 10% position would cost 9 — identical view, identical outcome, five times the damage. Sizing converts an unanswerable question into a survivable one.
Why does the TSMC ADR cost more than the Taipei shares?
At our September 2026 report the American depositary receipt traded at roughly a 15% premium to the shares in Taipei — a supply-and-demand gap between the two markets rather than anything about the company. It is a real cost: you need the shares to rise about 15% before you have matched the entry price of someone who bought locally.