Intel's Foundry Lost $2.1 Billion Last Quarter. Most Of The Chip Industry Isn't Buying From It.

Intel's data-center and AI business generated about $2.5 billion of operating profit in Q2 2026. In the same quarter, Intel Foundry, the company's in-house chipmaking arm, lost roughly $2.1 billion on $5.8 billion in segment revenue. External foundry customers, the outside companies whose orders would prove the business is real, contributed about $293 million, or roughly 5% of the segment, according to a Trefis analysis in Forbes. The rest of the segment, roughly $5.5 billion, was other Intel divisions buying chips from itself.

That gap is the architecture of the Intel bet. The company is now a hybrid: a profitable data-center and AI franchise cross-subsidizing a deliberately money-losing chip factory. The question is whether the factory side ever stops being a drag.

The hybrid has cash behind it, for now. Intel ended the quarter with roughly $30 billion in cash and short-term investments, a $10 billion revolver, and about $10 billion in non-core assets the CFO has flagged as monetizable. Capital spending is above $20 billion in 2026 and guided "significantly higher" in 2027. The trailing-twelve-month operating margin is 7.6% on $57 billion of revenue, down from roughly 34% at Intel's prior peak, and the company is still posting a profit, with revenue up 7.5% year over year on a trailing basis. AI-driven businesses, including the data-center group, now generate about 70% of quarterly revenue (Trefis in Forbes).

The market is pricing the foundry's pivot. Intel's stock is up 278% over the past year and trades at 8.2 times sales, near the top decile of its own decade. That multiple is not a bet that the foundry stays where it is. It is a bet that the gap between the $2.5 billion data-center profit and the $2.1 billion foundry loss closes.

Three things have to happen for that to be true.

First, external customers have to multiply. The July 2026 deal with Fortinet made the network-security company Intel Foundry's first named external customer for security chips, a small but real proof of life outside the company's own divisions (CNBC, TechTimes, Motley Fool). Going from $293 million to a number that meaningfully offsets a $2.1 billion quarterly loss means the customer list has to grow fast.

Second, the cost curve on Intel's own silicon has to keep falling. Panther Lake, the company's current-generation consumer and PC processor family, has already cut unit cost by about 50% year to date, with another 20% expected in 2026 according to management. Each percentage point of that curve turns foundry losses into smaller foundry losses, and at some point into breakeven, before any external customer revenue arrives.

Third, the cash bridge has to hold. Roughly $30 billion in liquidity, plus the revolver and the non-core asset sales, is the cushion against capital spending above $20 billion in 2026 and a higher 2027 bill. If rates stay elevated, if a customer win slips, or if a non-core asset sale falls through, the bridge gets tested.

The 278% move is, in effect, the market saying it believes all three will land. None of them are certain. What is certain is the size of the prize. A foundry segment that loses $2.1 billion in a quarter stops being a drag the moment external customers plus a steeper Panther Lake cost curve flip the segment's gross margin from deeply negative to roughly zero. From there, every additional customer adds profit instead of subtracting it.

The next test is Q3. If the external-customer number moves off $293 million, if Panther Lake hits its cost target, and if the capex guide does not get revised up again, the hybrid starts to look self-funding. If any of those slip, the data-center profit is paying for a factory that is still mostly selling to itself.