The Foundry the State Claimed

August 19, 2025: SoftBank had just put $2 billion into Intel, and Washington was floating something considerably stranger — the federal government taking a roughly 10% equity position in an American chipmaker. Not a loan. Not a grant with compliance strings and a ribbon-cutting. Ownership.

The commentary wrote itself, and it wrote itself badly. Half the country called it socialism, the other half called it a bailout with better branding, and both were relitigating 1985 while the serious objection went unmade: a government holding a large equity position in Intel becomes, simultaneously, Intel's regulator, the procurement authority for its largest customer, and its shareholder. Every export rule, subsidy allocation, and defense award from that day forward carries a thumb on the scale against AMD, Micron, and GlobalFoundries. That's an actual problem with the deal. Nobody raised it, because raising it would have required reading the terms.

Three days later the rumor stopped being one. On August 22 the deal was announced; on August 27 it closed. The government bought 433.3 million shares at $20.47 — $8.9 billion, a 9.9% stake — funded by converting the $5.7 billion of CHIPS Act money Intel had been awarded but not yet received, plus $3.2 billion already promised through the Pentagon's Secure Enclave program. Non-voting shares. No board seat. An agreement to vote with management on nearly everything.

Which is to say: on paper, the most passive $8.9 billion in the history of industrial policy. A stake carefully engineered to look like it wasn't one.

Then there was the clause nobody put in a headline.

Attached to the equity was a five-year warrant letting the government buy another 5% of Intel at $20 a share — exercisable only if Intel stops owning at least 51% of its foundry business. That is not an investment term. Nobody structures an option that pays off exclusively in the event their asset gets dismembered. That's a leash with a price tag on it. Translated out of securities language: spin off the fabs and we dilute you. The government wasn't buying upside. It was pricing a specific future — the one in which the last American company that still manufactures leading-edge logic decides manufacturing is somebody else's problem — and making it expensive.

How expensive is the entire question, and the answer got filed under the wrong story.

A year on, the trade looks embarrassingly good. By late April 2026 the stake was worth about $36 billion — a paper gain north of $27 billion, roughly 300% in eight months. Intel's 18A node hit high-volume manufacturing in January with yields above 60% and climbing. Panther Lake is shipping. Microsoft is running custom AI accelerators through Intel Foundry; Amazon is having Xeons and an AI fabric chip built there. The foundry business everyone had written off as a capital incinerator started acquiring the one thing it never had, which was customers who weren't Intel.

Everyone read that as the return on the investment. It is also, and more usefully, the return on the leash. A warrant to buy 5% at $20 is worth roughly nothing when the stock is at $20. It is worth billions when the stock has tripled. The appreciation didn't just make the trade look smart — it armed the clause. Whatever a foundry spinoff would have cost Intel in August 2025, it costs several times that today, and the government paid nothing extra for the upgrade. The deterrent got sharper for precisely the same reason the position got profitable. Those aren't two stories. They're one mechanism, and the coverage only ran the half with a percentage sign in it.

None of which means Washington got smart about semiconductors. Washington got lucky about timing. The stake appreciated because AI demand turned advanced silicon into a scarce commodity and lifted every company with a fab, not because anyone in Commerce understood EUV lithography. Buy any picks-and-shovels asset in mid-2025 and you'd be posting a similar number. The return is a coincidence — it just happens to be a coincidence that reinforced the one part of the deal that wasn't.

Because something in there was deliberate, and it was older and duller than the return.

Design is the brain. Fabrication is the building. Fabless was the entire religion of American semiconductors — let TSMC eat the capex, the yield curves, the twenty-billion-dollar bets on nodes that might not work, and keep the margins and the intellectual property here where the good weather is. A genuinely brilliant arbitrage, for exactly as long as geography stayed boring. Geography stopped being boring. You cannot rent a building that sits a hundred miles off a contested strait.

There's a second precedent buried in the funding line, and it may be larger than the warrant. The $5.7 billion wasn't new money. It was a grant Intel had already been awarded, converted after the fact into a claim on the cap table. That's a change in the form of American industrial policy, not just its target: every future recipient of federal money now has to price the possibility that a subsidy becomes equity later, on terms set at the moment the government's leverage peaks. It's also, not coincidentally, why the warrant was negotiable at all. You get to attach a leash when you're already holding the collar.

So here's the lesson, and it isn't "own the fabs."

Sovereignty is never about owning the high-margin layer. It's about owning the layer that can't be reconstituted under duress — and the ugly part is that the country had no organ for telling the difference. It didn't reason its way to fabrication. It needed a pandemic, a shortage, and an election cycle, and then paid $8.9 billion for the privilege of noticing something that had been true for a decade. The system finds out which of its dependencies is load-bearing by losing it. The stake wasn't the fix. It was the bill.

Which is why you watch the warrant, not the returns.

Every good quarter Intel prints from here, somebody credible will argue the state should take the win and exit — book the gain, declare industrial policy vindicated, get the government off the cap table. It'll sound like fiscal responsibility. It'll poll well. And the moment it happens, the 51% foundry clause goes with it, and the whole thing collapses back into a very profitable stock trade that accomplished nothing structural.

The equity was always the boring part. The paragraph about who owns the fabs is the entire document. I'll start the timer on the first serious op-ed calling for the sale.