Intel Bet the Company on Catching Up. It Forgot to Fix Why It Fell Behind.

In July 2020, on a routine earnings call, Intel buried the most expensive sentence in its modern history inside the engineering euphemism of a 'defect mode.' Its 7nm process — the technology that was supposed to keep it ahead of the world — was running roughly twelve months behind the company's own internal target.⁴ Eight months later, a new CEO walked on stage and called the fix a bold new strategy. The fix was overdue, and it would soon reproduce the exact failure it was meant to cure.

The official story is that IDM 2.0 was Intel reinventing itself — a proactive pivot into foundry services and a manufacturing comeback. The real story is that Intel had already been quietly shipping work to other people's fabs and had already confessed, on the record, that its flagship process was broken before a turnaround plan ever existed. IDM 2.0 wasn't a leap forward. It was a company catching up to a hole it had spent six years digging.

The plan was right. The root cause was never touched.

On March 23, 2021, Pat Gelsinger announced IDM 2.0: an estimated $20 billion to build two new Arizona fabs and a new business, Intel Foundry Services, to make chips for outside customers.¹ As a diagnosis it was sound — Intel had to keep its own fabs, lean on external foundries when needed, and turn manufacturing into a product. But notice what the strategy treated as solved. It assumed the manufacturing engine could be pointed in a new direction. It never addressed why the engine kept stalling in the first place.

Go back to 10nm. Intel didn't get unlucky; it got greedy. It set transistor-density goals so aggressive it gave them an internal name — 'Hyper Scaling' — and then discovered that packing that many features onto a wafer drove defect density up and yields down. Intel later acknowledged that this very ambition caused lower-than-expected yields and costs above its mature 14nm process.² A node planned for volume in 2015 slipped to 2016, then 2017, then 2018, and didn't reach mass volume until 2019 — four years late.³ That is not a delay. That is a chosen design philosophy crashing into physics, repeatedly, on schedule.

How a rescue plan inherited the disease it was meant to cure

Here is the mechanism, worked all the way down. A leading-edge process node is a bet placed years in advance: you commit to a density target, then spend years discovering whether you can actually manufacture it at yield. Intel's institutional reflex was to bet the most aggressive density on the table — because for decades, density leadership was its entire identity. That reflex worked when Intel was the only one running the race. It became a trap the moment the density target outran the company's ability to control defects, because every additional patterning step is another chance for something to go wrong, and the chances compound across hundreds of millions of transistors.

IDM 2.0 changed the org chart, the customers, and the capital plan. It did not change that reflex. So when 18A arrived — the node carrying RibbonFET transistors, PowerVia backside power, and the entire weight of the turnaround — it was reportedly stuck below 10% yields in late 2024, a level at which mass production simply isn't possible, with a named foundry trial customer said to have walked away.⁵ Intel's own CFO later described the failure in language that could have been lifted verbatim from the 10nm post-mortem: the company tried to do too much at once.⁹ Same overreach. Higher stakes. The cornerstone of the comeback cracked the same way the original problem did.

The fair objection: maybe the plan just needed more time

The honest counter is that turnarounds of this size take a decade, not a few years, and that judging IDM 2.0 by early 18A yields is like grading a marathon at mile six. Yields climb. RibbonFET and PowerVia are genuinely advanced bets that, if they mature, leapfrog rather than catch up. And the harshest 18A numbers trace to a single Korean media report, not a primary filing — so the sub-10% figure deserves a skeptic's asterisk.⁵ All of that is true, and it is why this is a real argument and not a verdict.

But the objection misses where the spine of the criticism actually sits. The problem was never that 18A was immature in 2024 — early nodes always are. The problem is that Intel's CFO named the cause as the identical overreach that had buried 10nm and 7nm.⁹ A company can survive a slow turnaround. What it cannot survive is repeating its founding mistake at the exact moment it has bet everything on not repeating it. The board appears to have reached the same conclusion: in December 2024 Intel said Gelsinger 'retired,' while multiple outlets citing people familiar with the matter reported he was given the choice to retire or be removed after the board lost confidence in his plan.⁶⁷ The strategy didn't run out of time. It ran out of patience for the same bug, shipped a third time.

Intel spent six years, a new strategy, and a fortune in credibility learning a lesson it already had in writing. The 'Hyper Scaling' overreach that broke 10nm in 2015 was diagnosed, named, and confessed — and then it broke 7nm, and then it broke the node that was supposed to redeem all of it. IDM 2.0 was the right map drawn over the wrong terrain: it pointed the company toward a comeback while leaving the engine that kept stalling exactly as it was. The catch-up was never going to fail for lack of ambition. It failed because ambition was the bug, and nobody patched it.