What About This Intel Foundry Story? A Reality Check

Intel’s stock is up more than 160 percent this year. A company that spent a decade falling behind TSMC is suddenly the centerpiece of America’s chip-independence narrative: a new Arizona fab currently running the most advanced process node in production on U.S. soil, a $5 billion Nvidia investment, a preliminary manufacturing deal with Apple, a starring role in Elon Musk’s Terafab project. The federal government acquired a 9.9 percent stake in the company in 2025. The story tells itself — American manufacturing is coming home, and Intel is leading the comeback.

How much of this is real progress, and how much is narrative?

What’s Real

Start with what’s real. Fab 52 in Chandler, Arizona, is running. It’s the first facility in the country producing on Intel’s 18A node. Panther Lake and Clearwater Forest ship from it. Intel reports numbers that imply significant progress is being made.

What Intel actually disclosed last quarter is production volume, not the yield rate: 18A output exceeded internal targets by roughly 25% and rose more than 50% quarter over quarter. That’s more wafers moving through the fab, which is real activity — but it isn’t the same claim as “yields are improving,” and Intel still won’t give a yield percentage.

Management has said wafer costs and cycle times are improving too — CFO David Zinsner said the primary Panther Lake chip’s cost fell roughly 50 percent this year — but cost improving isn’t the same disclosure as a yield number benchmarked against TSMC’s. Tan’s version of an answer, in January: yields were “in line with our internal plans” but “still below what I want them to be.” No number. That’s not disclosure. Wafers are moving. Whether they’re moving at a competitive cost per good chip is a separate question Intel hasn’t answered.

Relative percent improvements and other vague reporting aren’t encouraging on their own. Comparing Intel to Intel isn’t very useful when the question is whether Intel can compete with TSMC.

Trump’s Thumb on the Scale

The customer list looks different up close. Nvidia’s $5 billion stock purchase landed five weeks after the U.S. Treasury converted $8.9 billion of CHIPS Act money into a 10 percent equity position — a stake worth several times what the government paid for it after Intel’s stock more than doubled this year.

Washington isn’t just holding leverage over Intel anymore. It’s sitting on a large paper gain that shrinks or evaporates if the foundry story falls apart, which gives the government its own financial reason to keep the narrative alive, separate from whatever national-security case got the deal done in the first place.

Nvidia has its own reasons to stay close to this White House: the administration controls whether Nvidia can sell chips to China at all. Commerce revised H200 export licensing in January 2026. Nvidia’s own SEC filings show its access to the Chinese market has stayed directly dependent on discretionary U.S. export policy throughout the same period it became one of Intel’s largest strategic investors. Huang personally joined Trump’s May visit to China while Nvidia was trying to reopen H200 sales there. A company that dependent on the administration’s export licensing has every incentive to also show up for its industrial-policy priorities.

Apple’s path was more direct. In August 2025, Tim Cook was in Washington lobbying against a proposed 100 percent tariff on semiconductor imports. He got his exemption amid a larger U.S. investment pledge, while, the Wall Street Journal reported, Trump and Lutnick were pressing him to route some of Apple’s chip production through Intel’s fabs. The government’s 10 percent Intel stake was announced the same month. Ten months later, Trump announced the Apple-Intel deal himself, on Truth Social, before either company confirmed it.

Elon Musk (aka Mr. 10x, as I’ve written)’s Terafab — a $55–120 billion chip project for Tesla, SpaceX, and xAI that promises, per Musk, 50 times the world’s current AI chip output — named Intel a “strategic partner” the same season, on a process node, 14A.

When the government is simultaneously one of Intel’s largest shareholders, the tariff authority, and the party WSJ reported personally pushed Apple toward Intel’s fabs, “commercial deal” and “political favor” stop being different categories.

Where’s the Beef

Here’s what’s actually happening, measured in dollars: Intel Foundry’s external revenue — chips made for anyone other than Intel itself — was $293 million out of $5.8 billion in foundry revenue last quarter. For all of 2025, external customers accounted for $307 million of $17.8 billion in foundry revenue: less than two percent. The division lost $2.1 billion in the second quarter alone, an improvement from $3.2 billion a year earlier, but still a loss on a business Intel has told investors will turn profitable in 2027.

Here’s what’s aspirational, measured in announcements: Nvidia’s stake buys collaboration on chip design, not a signed foundry commitment. Apple’s deal is preliminary — a reported agreement, not a contract with volume attached. Musk’s Terafab arrangement licenses a process node, 14A, that hasn’t produced a chip yet.

Three headlines, no disclosed production revenue yet in the number that actually counts.

The Company That Didn’t Need Any of This

Meanwhile the company that already runs the world’s leading foundry is building in the same city — and committed to funding the overwhelming majority of it itself. TSMC’s Arizona campus is now a $265 billion commitment — the largest foreign direct investment in American history — spanning ten fabs, two packaging plants, and an R&D center. Its first fab is in production. Its second fab, targeting 3-nanometer, finished construction ahead of schedule.

Company-wide, TSMC raised its 2026 capital budget to $60–64 billion — a single year’s spending, funded from its own cash generation and debt rather than issuing new equity, with 70 to 80 percent going toward the next nodes: 2-nanometer now ramping, the A16 process debuting later this year, early work on A14 for 2028. TSMC didn’t need a government stake to get Apple’s business — Apple has been its customer for over a decade — and it didn’t need to touch equity markets to fund the future.

TSMC did take $6.6 billion in CHIPS Act grants for Arizona, the same program that seeded Intel’s deal. The difference is scale: TSMC’s grant is a rounding error against a $265 billion commitment funded overwhelmingly from its own operating cash flow.

Where Intel Fabs Are At

On August 11, Intel priced a $20 billion public stock offering — investors reportedly put in roughly $100 billion of orders for it, five times what was on offer. That money is meant to cover Intel’s 2026 capital budget, which the company has guided at roughly $20 billion for the year. TSMC is spending roughly three times Intel’s entire annual capex this year, with 70 to 80 percent of that going toward advanced process technologies alone. Intel’s CFO has already said 2027 spending will be “significantly above” this year’s, on a foundry business still running around 5 percent external revenue last quarter — up from under 2 percent for all of 2025, but still a rounding error next to the capital being bet on it.

Nothing in Tan’s public career record shows prior experience running a leading-edge fab or designing processors — a résumé gap I’ve written about, among several other pieces on Tan’s record at the site. He spent over a decade running Cadence, a mature EDA software company that makes tools for chip designers. His clearest accomplishments at Intel so far have been financial and organizational: stabilizing the immediate problems, getting $20 billion out of Wall Street, cutting costs, and keeping investor confidence intact. Those are real skills.

They aren’t the same skills as what you need to build a leading-edge process or to produce competitive processors.

Fab 52 and the 18A process trace to Pat Gelsinger’s IDM 2.0 strategy, announced in 2021, with more than $100 billion in committed capital and 18A already in development years before Tan arrived. The board fired Gelsinger in December 2024 while outside reporting suggested 18A yields were still extremely low. Tan took over three months later. Yields don’t recover overnight because a new CEO gives a good earnings call or runs a reorganization — Tan cut headcount roughly 15 percent in his first year — they recover after years of tooling, process refinement, and capital already committed start paying off.

If the fab turnaround really is happening this fast under Tan, then the credit should go to Gelsinger. In my view, seventeen months isn’t enough time for any operational or execution improvement of Tan’s own to show up in a multi-year yield curve. Whatever’s driving the reported progress was already in motion before he arrived. This brings into question the board’s decision to fire him — and raises the question of whether he could have been kept on in some capacity to finish what he started. Either things were already going well before Tan arrived, which undercuts the case for firing Gelsinger, or they aren’t actually going as well as the story suggests. The board doesn’t get to have it both ways.

What The Board Actually Pays For

Tan’s hiring package, disclosed in March 2025, includes a new-hire performance stock award with a target value of $17 million: Tan earns 100 percent of the target shares if Intel’s stock roughly doubles over three years, and up to 300 percent of the target shares — which does require beating the S&P 500 — if it triples — on top of a separate $25 million option grant vesting over five years. None of that is tied to yield, external-foundry revenue, or foundry profitability.

Look at the actual scale of what Tan’s pay is built from. His 2025 annual cash incentive — scored against revenue, gross margin, operating expense, and a set of board-defined “Top Jobs” goals — paid out $2.37 million. Intel’s own proxy shows Tan’s 2025 reported compensation at roughly $93 million, while the SEC’s accounting measure called “compensation actually paid” came to about $162 million, almost entirely because the stock went up.

The magnitude gap tells the story: Intel does measure Tan on real operational metrics, and the amount of money attached to that measurement is a rounding error next to what’s attached to the stock price.

The board didn’t merely ask Tan to improve Intel’s operations. It placed an enormous additional bet on something else: the stock price. That bet gives him a powerful incentive to deliver it. You can’t blame management for doing exactly what they were paid to do. The compensation structure makes the stock price the overwhelmingly larger financial incentive. If Intel’s most urgent need is competitive yield, real external customers, and foundry economics that work, it’s a fair question why the giant turnaround-sized award isn’t tied to any of those things instead.

That cuts both ways. If Tan fixes the fundamentals and the market doesn’t notice for years, the part of his pay that’s actually large pays him nothing — stock price and underlying business health can decouple for a long time. Microsoft’s did: the business grew and stayed profitable through most of the 2000s while the stock sat essentially flat for thirteen years. A pure absolute-price incentive like Tan’s new-hire PSU could have paid little or nothing during that stretch no matter how much the underlying business improved.

A Deadline Only If Washington Lets It Be One

Political power can move capital, accelerate headlines, and manufacture the appearance of momentum. It cannot manufacture yield, cost parity, or a customer’s decision to actually ship volume through a fab rather than just stand next to it at a press conference. Fab 52 is running and 18A output is beating Intel’s own targets — that much is earned. Whether the underlying yields are earned enough to make Intel a credible external foundry, rather than just a capable one for itself, is the question the external revenue number — roughly 5 percent last quarter, still nowhere near what the capital spending assumes — hasn’t answered yet.

The political scaffolding around Intel’s story isn’t temporary regardless. Trump has roughly two and a half years left in office. He still holds the tariff authority that got Apple to the table, the equity stake that makes Washington a stakeholder in Intel’s success, and the direct line to Cook, Huang, and Musk that helped produce this burst of deals and announcements. Nothing requires him to stop leaning on that scale before 2027, or after.

The Test

Tan’s compensation is now structurally tied to the stock holding up — the three-year performance window that determines how many PSU shares he earns runs from March 2025 to roughly March 2028, sitting entirely inside Trump’s remaining term. None of this requires anyone to be conspiring. Trump wants an American chip success story. Washington owns Intel stock worth several times what it paid. Apple wants tariff relief, Nvidia wants export licenses, Musk wants fabrication capacity and political goodwill. Every one of those incentives points the same direction — toward keeping the story intact — without any of the people involved needing to coordinate it.

Washington can create demand for Intel while Washington is leaning on the scale. The test of Intel Foundry begins when it no longer has to.

The Intel foundry story isn’t fiction. It’s a five-percent business wearing a hundred-percent narrative.