Intel's Foundry 'Choice' Wasn't a Choice. Washington Already Made It.

In the fall of 2024, Intel Foundry - the business meant to make Intel the manufacturing backbone of American chips - reported $67 million in external revenue for the quarter.³ Sixty-seven million dollars. For context, that same segment lost $11.148 billion over the first nine months of the year.³ Read those two numbers together and a strange shape appears: a business spending billions to serve customers who, by and large, were not yet there. That is not how a company enters a market. It is how a company fulfills an order.

The official story is that Intel made a daring strategic choice to open its fabs to rivals - a bold pivot to compete with Taiwan's TSMC for the world's chip-making business. The truer story is that the choice had already been made for it. Washington wanted American-soil fabs for national-security reasons, dangled CHIPS Act money to get them, and Intel - the last U.S. company with leading-edge manufacturing - was the only realistic taker. The fork wasn't whether to build. It was whether to admit there was never really a second road.

A foundry that launched three years before it existed

The legend starts on March 23, 2021, when CEO Pat Gelsinger unveiled 'IDM 2.0' and Intel Foundry Services, anchored by roughly $20 billion for two new Arizona fabs and 'strong enthusiasm' from across the industry.¹ But enthusiasm is not revenue. More than two years later, in the first quarter of 2023, the foundry's external sales were $118 million - and falling, down 24% year over year.⁷ The actual structural separation - renaming the unit Intel Foundry, giving it its own profit-and-loss, and forcing Intel's own chip designers to buy from it at arm's length - didn't formally take effect until early 2024.⁸ So the thing announced in 2021 spent three years as a press release wearing the costume of a business.

This matters because the sequence is backwards from how a healthy market entry works. Normally you win a few anchor customers, prove the economics, then build capacity to meet demand you can already see. Intel did the reverse: it committed the capital, the org chart, and the political relationships first - and then went looking for the customers. That inversion is the tell. When you build before demand exists, you are not chasing a market. You are answering to whoever is paying for the building.

Who actually placed the order

Follow the money and the obligation comes into focus. In March 2024, Intel and the government announced a preliminary, non-binding deal for 'up to $8.5 billion' in CHIPS Act support. The final binding award, signed that November, came in lower - $7.86 billion - trimmed partly because Congress steered some of the funding toward a separate $3 billion 'Secure Enclave' program to make chips for the Defense Department.⁵ Notice what that detail reveals: the money was never neutral capital. It carried strings shaped by what the state needed - secure, domestic, military-grade fabrication - not by what the chip market was asking Intel to build. The customer of first resort was the customer writing the subsidy.

Subsidy money looks like a gift until you see what it commits you to. Once Intel took it, the foundry stopped being an option it could quietly close if the numbers didn't work. It became a national project with Intel's name on the door - and projects like that are far easier to start than to abandon.

The losses got too big to walk away from

Here is the trap, in plain figures. Intel Foundry lost $5.17 billion in 2022, then $6.957 billion in 2023 on $18.9 billion of revenue - of which only $953 million came from outside customers.² On the April 2024 investor call, Gelsinger told the market that 2024 would be 'the trough' - the worst year, after which losses would shrink toward a 2027 break-even.²⁴ He was directionally right and catastrophically off on scale: the full-year 2024 foundry loss landed near $13 billion.⁴ A trough you predicted at single-digit billions arriving at thirteen is not a trough you control. It is a hole that keeps deepening while you describe the bottom.

Then the architect left. On December 1, 2024, Gelsinger 'retired,' in the company's word - though the move came as the losses he had promised to tame ran past every guidance he had given.⁶ The man who bet the company on the foundry was gone before its trough year had even closed its books. What remained was the bet itself, now too large to write off and too slow to vindicate.

But what if the foundry is supposed to lose money for now?

The fair objection is that this is exactly what a leading-edge foundry looks like in its build phase. TSMC didn't become indispensable overnight; advanced fabs cost tens of billions and take years before utilization and yield turn losses into profits. Intel might genuinely be early, not wrong - paying upfront for capacity that pays off after 2027, with a foundry independent enough (its own subsidiary, an arm's-length internal customer) to finally win the outside business it lacked.⁸ On that reading, the $13 billion isn't a bleak; it's a down payment.

That case is real, and it could prove right. But notice what it quietly concedes: the justification for the spend keeps shifting away from customers and toward time. The honest counter is that the demand was supposed to follow the capacity - and across three years it barely moved, from $118 million a quarter to $67 million.⁷³ A build-ahead strategy is defensible when customers are visibly arriving and you're racing to serve them. It is something else entirely when the capacity is built, the customers haven't come, and the loudest reason to keep going is that you've already spent so much. The break-even date can be true and still be the wrong question. The right one is whether anyone outside Intel and the government is actually buying.

So the foundry fork was never really a fork. A genuine strategic choice has two roads you could honestly take. Intel had one road it was paid to walk, a balance sheet that bled harder each year it walked it, and a CEO who left before the worst year closed. The decision to make chips for rivals will be vindicated only if the rivals show up to buy. Until then, Intel isn't competing with TSMC for the world's chip business. It's holding open a fab the country wanted built - and discovering, billion by billion, what it costs to keep a door open for customers who haven't yet walked through it.