Intel's foundry problems and what they mean for

Intel reported a $7 billion operating loss in its foundry division last year. Seven billion. That’s not a rounding error. That’s a company bleeding from its core.

For most of the computing era, Intel was the company. The chip inside your computer said “Intel Inside” and that wasn’t just marketing, it was fact. Intel designed chips and Intel made chips, in Intel fabs, on Intel equipment (well, ASML equipment, but still). Vertical integration. Design to silicon. One company, one pipeline.

That model is breaking.

What happened

The short version: Intel fell behind on process technology. TSMC moved to 7nm, then 5nm, then 3nm, while Intel was stuck trying to get 10nm to work. By the time Intel caught up (sort of), the world had moved on. Apple, NVIDIA, AMD, Qualcomm, all of them design chips. None of them make chips. They send their designs to TSMC. TSMC makes them.

Intel’s response was to split itself into two businesses: Intel Products (which designs chips) and Intel Foundry (which makes chips for anyone). The foundry would compete with TSMC and Samsung for outside customers.

The problem is that nobody trusts Intel Foundry yet. The 18A process node (Intel’s name for what other companies call “roughly 2nm”) is delayed. Yields are reportedly low. The first major external customer announcements have been quietly walked back. Microsoft was rumored to be a launch customer. The timeline keeps slipping.

And building fabs costs tens of billions. Intel is spending $100 billion over the next five years on new manufacturing capacity. That’s CHIPS Act subsidies plus Intel’s own money plus debt. All of it bet on the idea that Intel Foundry can compete with TSMC.

The scenarios

I see four paths.

Path one: Intel Foundry succeeds. 18A works. Yields improve. External customers arrive. Intel becomes a credible second source to TSMC. The US gets domestic leading-edge chip manufacturing. Good outcome.

Path two: Intel Foundry stumbles but survives. 18A is late and yields stay mediocre. Intel Foundry operates at a loss for years, propped up by government subsidies and the geopolitical argument that America needs a domestic foundry. Not profitable, but strategically important. A zombie that serves a purpose.

Path three: Intel Foundry is spun off or sold. The losses become unsustainable. Intel separates its product design from its manufacturing. The foundry becomes an independent company (or gets acquired by someone like Samsung or a private equity consortium). Intel Products becomes a fabless chip designer, like AMD.

Path four: Intel Foundry collapses. Customers don’t come. The technology doesn’t compete. The fabs become expensive buildings that make inferior chips. Intel retreats to being a design company. The US loses its bid for domestic leading-edge manufacturing.

None of these is implausible. The range of outcomes is enormous. And the implications extend far beyond Intel’s stock price.

Why this matters to everyone

If you care about AI, you should care about where AI chips get made. Right now, approximately 90% of the world’s most advanced logic chips are manufactured by TSMC in Taiwan. Ninety percent. Of the most critical technology of the 21st century. On an island 100 miles from a geopolitical adversary.

Intel Foundry is the main American attempt to change that number. The CHIPS Act money, the Ohio fabs, the Arizona expansion, all of it funnels into Intel’s ability to make leading-edge chips on US soil.

If Intel Foundry doesn’t work, the backup plan is… unclear. TSMC’s Arizona fab helps, but TSMC’s primary operations (and most of their R&D talent) remain in Taiwan. Samsung’s Texas fab is behind schedule. There’s no path C.

The concentration of advanced chip manufacturing is a fragility that most people don’t think about. But every AI model you use, every GPU training run, every autonomous vehicle sensor, depends on a supply chain that runs through a single island in the Pacific. Intel’s foundry problems aren’t just an Intel problem. They’re a structural risk to the entire technology industry.

The human side

I visited Intel’s Hillsboro campus a few years ago. The people there are incredibly talented. The engineers working on 18A know what they’re doing. Intel’s problems aren’t talent problems. They’re execution problems, bureaucracy problems, culture problems.

TSMC has a culture of relentless execution that borders on obsessive. Twelve-hour shifts. Engineers sleeping at the fab during yield ramps. A singular focus on making the process work, whatever it takes. Intel has historically been an engineering-first company, but decades of market dominance bred complacency. The urgency wasn’t there because the competition wasn’t there.

Now the urgency is there. The question is whether it’s too late.

I genuinely don’t know. I find myself rooting for Intel Foundry not because I hold Intel stock (I don’t) but because the concentration of chip manufacturing worries me at a fundamental level. The world needs more than one place that can make 2nm chips. If Intel can’t be that place, who can?

The answer might be nobody, for a while. And that’s the scenario that should keep us all up at night.

SemiAnalysis has good deep dives on the technical details. Reuters has been tracking the business side closely. I keep both feeds open. This story is going to define the next decade of computing, one way or another.

Related thinking: