The Foundry Race Is Already Over. The Only Question Left Is Who Pays to Pretend It Isn't.

Open Intel's 2024 annual filing to the foundry segment and you find a number that has no precedent in the history of chipmaking: an operating loss of $13.4 billion in a single year, on a business whose actual outside customers paid it just $385 million.² That is the entire contest in two figures. One number is the cost of trying to build a world-class foundry. The other is how much of the world actually bought one. The gap between them is roughly thirty-five to one, and it is not closing—it widened by nearly $6.5 billion in a single year.

The official story is that the foundry race is a three-horse contest—TSMC, Samsung, and Intel, each chasing the lead in the most advanced chips on earth. That framing is comforting, geopolitically convenient, and roughly a decade out of date. The race for the lead is over. What remains is a different question entirely: whether two governments can afford to keep their losing horses on the track long enough for the running to matter.

TSMC isn't winning the foundry race. It has already won it—and is now competing only against the gravity of its own scale. By the close of 2024 it held 67% of pure-play foundry revenue, a share that rose to a record 70.2% by mid-2025.⁵⁸ In an industry where one point of share is worth billions, two-thirds belonging to a single company is not a lead. It is a category.

The accidental architecture that made it unbeatable

Here is the part that gets mythologized and shouldn't. TSMC is usually described as the product of Morris Chang's grand vision to reinvent the semiconductor industry. It wasn't. Chang was recruited to Taiwan's research institute in 1985 by a government that wanted a chip industry and had handed him the assignment. Taiwan was weak at chip design and weaker at marketing—so the only model that could work was the one nobody had built: a factory that made chips for other people's designs and never competed with them.⁷ TSMC was founded in 1987 with almost half its capital coming straight from Taiwan's National Development Fund and a quarter of its equity traded to Philips for the technology it didn't have.⁷ The 'never compete with your customer' promise that became TSMC's deepest moat was born from a weakness, not a strategy.

That accidental architecture turned out to be the whole game. A pure-play foundry is a trust machine. A chip designer hands over its most valuable secret—the design—to a factory, and the factory's single non-negotiable promise is that it will never use that secret to build a competing product. TSMC could make that promise absolutely, because it sells nothing but manufacturing. In 2024 it ran 288 distinct process technologies to build 11,878 different products for 522 customers.¹ Every one of those customers knows the factory has no chip of its own to favor. That is the asset. And it is precisely the asset Intel and Samsung structurally cannot offer.

Why the hybrid model bleeds and the pure one prints

Both challengers are integrated device manufacturers—they design and sell their own chips while also offering to manufacture yours. This is the IDM-hybrid model, and it carries a contradiction the pure-play model simply doesn't have. When you ask Intel to build your chip, you are asking the company that makes your competitor's chip to also make yours, on shared capacity, with shared roadmaps. The conflict isn't theoretical; it's the reason Intel Foundry's external revenue in 2024 was a rounding error of $385 million against $17.5 billion in total segment revenue—the overwhelming majority of which is Intel building chips for itself and booking the internal transfer as 'revenue.'² Strip out the company buying from its own factory and the real foundry business is barely visible.

Samsung tells the same story in slow motion. It is genuinely the world's number-two foundry, and that ranking is real—but the trajectory underneath it is brutal. Through Q3 2024 it held around 12% of foundry revenue; by Q4 2024 TrendForce measured it at 8.1%, a slide that continued to 7.3% by mid-2025 even as the overall foundry market hit record highs.⁶⁵⁸ Number two is shrinking inside a growing market. That is the signature of a company losing share faster than it can be masked by the rising tide. Samsung now reportedly targets profitability by 2027 and a 20% share, anchored on Tesla contracts and its Taylor, Texas fab.⁸ Both challengers, notably, have set their break-even on the same date.⁴⁸

Intel said this in April 2024, alongside guidance that 2024 would be the year its foundry losses peaked.³ The direction was right—2024 was indeed meant to be the trough. The magnitude was a catastrophe: the actual loss came in at $13.4 billion, nearly double the prior year's $6.955 billion.² By December the CEO who set that framework, Pat Gelsinger, was gone; his interim replacement reaffirmed the 2027 break-even hope on the next earnings call, after the company had already cut 15% of its workforce in August.⁴ When the trough is double what you forecast and the architect is fired before the year ends, 'on track for 2027' is a sentence carrying a great deal of weight.

The honest counter: maybe profit was never the point

The fair objection to all of this is that I'm scoring a geopolitical contest on a financial scoreboard. Intel Foundry and Samsung Foundry may never need to beat TSMC on margin or share. Their job is to exist—to give the United States and South Korea a sovereign capability to make advanced chips that does not run through a single complex of fabs ninety miles from China. On that scoreboard, a $13.4 billion loss isn't a failure; it's a national insurance premium, and a cheap one against the cost of having no domestic foundry at all. This is the real reason the race looks like a three-way contest: two of the players are competing for a prize that isn't market share.

That objection is correct, and it is exactly why the race is over in the only way that was ever in doubt. Concede the geopolitics entirely and the strategic picture sharpens rather than blurs. The question is no longer 'who beats TSMC'—nobody does, on the current evidence. The question is whether two unprofitable challengers can be kept alive by national will long enough to ever matter commercially. Subsidy can fund a fab. It cannot manufacture the one asset TSMC built by accident: the unconditional trust of a customer who knows the factory will never become its rival. Money buys capacity. It does not buy the absence of a conflict of interest, and that conflict is structural, not financial.

TSMC backed into the most defensible position in technology by being too weak, in 1987, to do anything but manufacture. Forty years later, its two largest rivals are spending tens of billions to acquire by force the one thing it got for free: a customer's certainty that the factory has no horse in the race. Intel and Samsung can build the fabs. They have set the same target year, 2027, to stop losing money. But the deepest asset in this industry was never the silicon or the capital—it was the promise never to compete with the people who trust you. You can subsidize everything in a foundry except that. And the company that can keep it is the one that never had a chip of its own to protect.