One Island Makes 90% Of The World's Chips

Here's the number that organizes everything else. Roughly 90% of the world's most advanced chips — the leading-edge processors that run modern phones, cars, data centers and AI systems — are manufactured by a single company: TSMC, headquartered in Taiwan. Across the broader foundry market it holds around 60% share, and about 72% of the pure-play foundry business.

Its nearest competitor, Samsung, sits at roughly 7%.

I want to be precise about the word "advanced," because the distinction matters. Plenty of countries make chips. Older, simpler chips are manufactured in dozens of places, including at scale here in China. What one island dominates is the leading edge — the smallest, fastest, most complex processors, the ones that go into the products that define a modern economy. For those, there is no meaningful second source. Not a more expensive one. Not a slower one. None at scale.

That's what makes this different from every other concentration story I cover. With most dependencies you can pay more and go elsewhere. Here, the elsewhere doesn't exist yet.

This is the part that turns a geopolitical curiosity into a line on your P&L.

Through 2026, TSMC has been raising prices on advanced nodes — reported increases in the range of 3% to 10%, with the most advanced 3nm processes seeing hikes around 15% in the second half of the year. At the leading edge the numbers are starker: a single 2nm wafer now runs above $30,000, compared with roughly $20,000 for 3nm — an increase of more than 50%.

You don't buy wafers. But Apple does. Nvidia does. Your automaker does. Every company whose product contains a modern processor does.

And when a supplier holds 90% of a market with no alternative, price increases aren't negotiated — they're announced. There is no leverage on the buyer's side. That cost moves downstream, through the device manufacturer, to the retail price, to the subscription fee, to the invoice.

The era of semiconductors getting cheaper every year — the assumption underneath four decades of consumer electronics pricing — has quietly reversed.

This didn't happen by accident, and it didn't happen quickly.

Beginning in the 1970s and 1980s, Taiwan made a deliberate national bet: rather than compete across the whole electronics industry, it would specialize in the single hardest part of it — manufacturing chips designed by other people. The pure-play foundry model, invented there, split chip design from chip production and let TSMC concentrate every resource on process technology.

Meanwhile, the West did the economically rational thing. Chip fabrication is brutally capital-intensive, low-margin relative to design, and unforgiving of error. It was cheaper to buy than to build. So Western firms moved up the value chain into design and software, and let manufacturing go.

Both sides optimized correctly for their own position. One optimized for margin. The other optimized for irreplaceability. Thirty years later, only one of those turned out to be strategic.

The United States understands the exposure and is spending accordingly. TSMC has committed roughly $165 billion to build six fabrication plants in Arizona — one of the largest foreign direct investments in American history. Capacity there is reportedly sold out through 2027 before it fully comes online.

But two facts complicate the story.

First, cost. Chips produced in Arizona currently run an estimated 5% to 20% more expensive than the same chips produced in Taiwan — a figure AMD's CEO has stated publicly. Reshoring capacity does not reshore the price.

Second, time. You can fund a fabrication plant in a year and build one in three. You cannot rebuild thirty years of accumulated process knowledge, supplier ecosystems, and trained engineers on that schedule. The people who know how to run these lines at yield are concentrated in the same place the fabs are.

So the concentration is being addressed. It is not being resolved, and certainly not within the planning horizon of anyone reading this.

Most operators I speak with have never mapped their own chip exposure. They know their landed cost, their freight rate, their labor line. They have never asked which of their products contain a leading-edge processor, or where it came from.

That question is no longer academic. If your product contains a modern chip — and increasingly, products contain chips that didn't a decade ago — you have a cost input controlled by a single supplier with no competitive check on pricing.

The exposure runs deeper than the obvious categories. It isn't only electronics companies. It's anything with a control board, a sensor, a connected feature, a display. It's the machinery in your plant. It's the AI service quietly added to software you already pay for, priced against compute that runs on those same chips.

I won't pretend to forecast the geopolitics, and I'd be suspicious of anyone who does. Taiwan sits roughly 130 kilometres from the Chinese mainland, and every analyst on earth has an opinion about what that means. I'm not going to add another one, and I don't need to.

The commercial reality is sufficient on its own. Even in a completely stable world, with no disruption of any kind, one company controls the pricing of the most important manufactured input in the modern economy — and it has begun using that position. That is happening now, on invoices being issued this quarter.

My read, after twenty-five years across factory floors in four countries: this does not become less concentrated within the decade. Plan against the pricing power, not against the headline risk.

First, find the chips in your own business. Go through your product line and your equipment. What contains a processor? Which supplier provides it, and do they know where it's fabricated? Most won't have been asked. The exercise itself surfaces exposure people didn't know they had.

Second, put a number on it. If your chip-related input costs rise 10%, what does that do to your unit cost, your margin, and your next quote? That's a ninety-second calculation and almost nobody has run it. If you want to model it against your real landed cost, the Tariff & Landed-Cost Calculator is free — and as a reader you get seven days of full Pro access, no card required.

Third, ask your supplier before your next order, not after. "Are chip costs rising on this component, and by how much?" Ask while you still have room to plan, requalify, or reprice. After the invoice arrives, you're absorbing it.

You cannot reshore a semiconductor industry from your desk. But you can make sure that when the cost moves — and it already is — you're the operator who priced it in advance rather than discovered it in arrears.