TSMC The Company Every Chip Depends On

1. The name that is not on the box

The last three videos in this series were about chips: how they are made, and what the numbers in their names really mean. Every one of them kept running into the same company, and it is a company most people have never heard of.

Its name is Taiwan Semiconductor Manufacturing Company, usually shortened to TSMC.

Look at how two of the world's best-known chip designers describe their products. NVIDIA's own page says its Blackwell chip is "manufactured using a custom-built TSMC" process. Apple's announcement of its A20 Pro chip says only that it is built on two-nanometre technology, and does not name who built it.

That difference is not an accident, and it is not a secret. It is the business model.

TSMC designs nothing that it sells under its own name. It makes chips that other companies have designed, and puts their names on them, not its own. So it never competes with a single one of its customers.

That one decision, made almost forty years ago, explains nearly everything else in this video.

So this video covers five things. Where the idea came from, and how much of the market it now holds. Who pays for it, where the factories are, and why nobody has simply copied it. The machines inside the factories were covered in the first video of this series, so we will not go back over them here.

We start with a man who spent twenty-five years at an American chip company.

2. A factory with nothing of its own

Morris Chang was born in nineteen thirty-one, in Ningbo, in China. In nineteen fifty-eight he joined Texas Instruments, and he stayed until nineteen eighty-three. By the end he was running the company's worldwide semiconductor business.

In the mid nineteen eighties, the Taiwanese government recruited him to lead its Industrial Technology Research Institute, a government research body. So a career that began in China, and was built in the United States, moved to Taiwan.

In nineteen eighty-seven he founded TSMC, and the Computer History Museum describes what was new about it. It calls TSMC a "dedicated silicon foundry", "an independent factory available to anyone for producing integrated circuits". Integrated circuits is the formal name for chips, and a foundry is simply a factory that makes things to someone else's design.

That was an unusual idea at the time, and it is worth seeing why.

Until then, a company that wanted to sell chips usually had to design them and build them too. That meant owning a factory, which cost a fortune and sat idle whenever its own products did not sell. Chang's model split the two jobs apart: one company designs the chip, and a separate company, with no products of its own, builds it.

Splitting the job that way sounds like a small change. It turned out to reshape the whole industry.

3. Why a factory for hire works

The reason it works comes back to trust.

If you design chips, you do not want your competitor to see your designs. A factory owned by a rival is a risk, however good it is. A factory that sells nothing of its own has no reason to copy you, so you can hand it everything, and so can your competitors.

Because so many companies can share one set of factories, each of them avoids building its own. And the factory stays busy, because when one customer's orders fall, another's rise.

The scale of that sharing is in TSMC's own figures. In twenty twenty-five, it says, it served five hundred and thirty-four customers. Between them they had it make twelve thousand, six hundred and eighty-two different products. It offered three hundred and five different manufacturing processes to make them with.

A process here means a particular recipe for building a chip, tuned for speed, for low power, or for cost.

A company that designs chips but owns no factory is called fabless, from the word fab, which is the industry's name for a chip factory. The fabless company spends its money on designers and software. The foundry spends its money on buildings and machines, and every fabless customer shares the bill.

That arrangement is now the normal way to make an advanced chip. And one foundry makes far more of them than anyone else.

4. Seventy-two and a half per cent

TrendForce, a market research firm, ranks the world's largest foundries every quarter by the money they earn.

Its ranking for the second quarter of twenty twenty-six was published on the ninth of September. It put TSMC's revenue at about forty point two billion dollars. The next largest foundry, Samsung, earned about three point two six billion. China's SMIC was third, at about three billion.

Put those numbers side by side and the gap is hard to take in.

Measured as a share, TrendForce says TSMC took seventy-two and a half per cent of the revenue of the ten largest foundries. Samsung took five point nine per cent. SMIC took five point four per cent.

That figure is often quoted, so it is worth being exact about what it measures.

It is a share of foundry revenue: money earned making chips to order for other companies. Chips that a company makes in its own factories are not part of it. TrendForce put the ten largest foundries' combined revenue at about fifty-three point five billion dollars for the quarter, a record.

Seventy-two and a half per cent of a record is a great deal of money. What does that look like for TSMC itself?

5. The money

TSMC publishes its results every quarter, and its sales every month.

In the second quarter of twenty twenty-six, TSMC reported revenue of forty point two billion US dollars. Of every dollar it took in, fifty-five point six cents was left as profit after all its costs and taxes. That share is called the net profit margin.

The monthly figures are in New Taiwan dollars, and they show the shape of the year so far.

Month by month, twenty twenty-six has mostly climbed. February was the low point, at about three hundred and eighteen billion New Taiwan dollars. August was the highest month of the year so far, at about five hundred and fifteen billion, up fifty-three per cent on a year earlier.

Those figures are the company's own, and it notes that the monthly numbers have not yet been audited.

TSMC also says what it expects next. For the third quarter of twenty twenty-six, it guided revenue of between forty-four point six and forty-five point eight billion dollars. That is a forecast, and it is shown here as the range the company gave, not as a single number.

So where does all of that money come from?

6. Who pays

TSMC's annual report to American regulators, filed in April twenty twenty-six, breaks its revenue down in several ways.

By the kind of product, fifty-eight per cent of its twenty twenty-five revenue came from high-performance computing, which mostly means the chips in data centres. Smartphones were twenty-nine per cent. By region, customers based in North America brought in seventy-five per cent.

It also shows how much depends on a few customers.

The largest single customer paid nineteen per cent of TSMC's revenue in twenty twenty-five. The second largest paid seventeen per cent. Together, the ten largest customers paid seventy-eight per cent of everything.

The report does not say who those customers are, and this video will not guess. What can be said is what the customers themselves say. NVIDIA names TSMC on its own product pages. Apple's announcement of the A20 Pro does not name its manufacturer at all.

A company with that many dollars flowing from so few names has an obvious question to answer: what exactly are they paying for?

7. The newest recipes earn the most

TSMC's results split its wafer revenue by process. A wafer is the thin disc of silicon that chips are made on, as the first video in this series showed.

Look at the second quarter of twenty twenty-six. The three-nanometre process brought in thirty per cent of TSMC's wafer revenue. Five-nanometre brought in thirty-three per cent. Seven-nanometre added another eleven. The newest process, two-nanometre, was already three per cent.

Those names are not measurements. As the first video in this series explained, a nanometre figure in a process name is a product name, not the size of anything on the chip. What matters here is how quickly the newest processes take over the revenue.

Taken together, TSMC's processes at seven nanometres and below earned seventy-seven per cent of its wafer revenue in that quarter.

The two-nanometre process is very new. TSMC's own page says it started volume production in the fourth quarter of twenty twenty-five, as planned. Within about six months, it already accounted for three per cent of wafer revenue.

Staying at the front like that is the most expensive habit in manufacturing.

8. The cost of staying ahead

TSMC's capital spending, meaning the money it puts into new buildings and machines, was about one point two seven trillion New Taiwan dollars in twenty twenty-five. For twenty twenty-six, it planned between fifty-two and fifty-six billion US dollars. Its factories could already produce more than seventeen million wafers a year, counted in the industry's standard twelve-inch size.

The reason the spending never stops is that every new process needs new equipment, and much of the old equipment cannot make the new chips.

A new process needs new factories, fitted with some of the most complex machines ever built. Each one has to be installed, tuned and learned before it produces anything useful. That learning is slow, and it cannot be bought.

It works as a loop. Customers pay most for the newest processes. That money pays for the next round of factories. The next round of factories wins the next round of customers, and the loop turns again.

The loop explains the money. It does not explain the map.

9. Almost all of it in one place

TSMC lists its factories on its own website, and the list is short.

In Taiwan it runs six of its largest factories, which it calls GIGAFABs, plus several older, smaller ones. Outside Taiwan it has one site in Arizona and one in Washington State in the United States, one in Japan, and one in China. A specialty factory in Dresden, in Germany, is being built.

A GIGAFAB is TSMC's word for its biggest factories, each one built to process the standard twelve-inch wafers.

So the most advanced of TSMC's manufacturing sits on one island. That is simply what the company's own list shows. It is also why the expansion in the next chapter matters so much.

The overseas sites are not copies of Taiwan. The Washington State factory works on older, eight-inch wafers, and the German factory is described as a specialty plant. Outside Taiwan, the site that TSMC is building up for its newest processes is Arizona.

So what is actually happening in Arizona?

10. Arizona

The first Arizona factory has been in volume production since late twenty twenty-four, making chips on TSMC's four-nanometre process. The second has finished construction, with volume production expected in twenty twenty-seven. Then, on the sixteenth of July twenty twenty-six, TSMC announced a further expansion.

The announcement was relayed the same day by the Arizona Commerce Authority, the state's economic development agency.

TSMC said it would invest another one hundred billion dollars, for four more factories at two nanometres or beyond. That brings its total planned investment in Arizona to two hundred and sixty-five billion dollars. The plan now covers ten factories, two advanced packaging plants and a research centre.

Advanced packaging is the step that joins finished chips together into one working part.

One claim in that announcement is worth singling out, because it is the one that matters most. TSMC says the first Arizona factory is running "with yields comparable to TSMC's facilities in Taiwan". Yield means the share of chips on each wafer that actually work, so a factory with poor yields throws away much of what it makes.

That claim comes from the company itself. It is attributed here as the company's statement, not as an independent measurement.

On a map, the size of the move is plain. Most of TSMC's largest factories are in Taiwan, and Arizona is on the far side of the Pacific Ocean. The first factory there makes four-nanometre chips, and the four new ones are planned for two nanometres or beyond.

11. Why nobody simply copies it

If the model is so good, why has nobody matched it? The figures in this video suggest four reasons.

The first is trust, because TSMC still competes with none of its customers. The second is scale: five hundred and thirty-four customers keep its factories full. The third is money, spent at a pace few companies can match. The fourth is learning, which can only be gained by running the newest processes for years.

None of those four can be bought quickly. Each one depends on the others, which is why the loop from chapter eight is so hard to break into.

The distance to second place makes the point. In the second quarter of twenty twenty-six, TrendForce put Samsung's share of foundry revenue at five point nine per cent. TSMC's was seventy-two and a half.

That is where the story ends up: a company whose name is on almost nothing, making the chips that almost everything else depends on.

Follow any advanced chip back to where it was made. The name on the box is the company that designed it. The factory behind it very often belongs to TSMC. That split, between the designer and the builder, is the idea Morris Chang started with in nineteen eighty-seven.

Whether that factory can be built again somewhere else is the question the company is now spending hundreds of billions of dollars to answer.