Intel 10nm finally ships in volume this year, in Ice Lake laptops, roughly four years after it was supposed to and long after the only thing that mattered had already happened: somewhere in the delay, Intel stopped being the most advanced chipmaker in the world. The node works now. It took so long to make it work that TSMC’s 7nm passed Intel while Intel was stuck, and the lead Intel held for fifteen years is simply gone. 10nm is the node that lost it.
The failure is a specific kind, the kind that comes from being too good for too long. Intel set 10nm targets that were wildly aggressive, a density jump far bigger than a normal shrink, packed with new tricks like cobalt wiring, and it refused to use EUV, betting it could push the old deep-ultraviolet light one more generation on engineering alone. The bet that looked merely risky at 14nm became a catastrophe at 10nm. Yields were so bad the first product, Cannon Lake, barely shipped at all, a paper launch nobody could actually buy. Intel had out-engineered the physics for two decades and finally, on a node it made too hard on purpose, the physics won.
While 10nm did not work, Intel still had to sell chips, so it stayed on 14nm and refined it, and refined it again, and again, the same node stretched across Skylake and Kaby Lake and Coffee Lake and Comet Lake until the plus signs became a joke. 14nm+++ was a punchline for years of standing still. Customers got small gains, more cores bolted on to cover for a process that would not advance, and watched a company that used to ship a new node every two years ship the same one five times. The metronome did not just miss a beat. It stopped.
The reordering is the part that will matter for a decade. TSMC reached 7nm in 2018 and put it under Apple’s chips and AMD’s, and those chips are now beating Intel’s on the metrics Intel used to own. Samsung shipped a 7nm EUV chip this summer, and whatever its yield troubles, it committed to the EUV future Intel refused. For the first time since the 1990s, the most advanced logic on earth is not built by Intel. AMD, riding TSMC’s process, is taking server share Intel treated as permanent. The integrated model’s whole premise, that Intel’s fabs are the best, is no longer true.
This is where the closed-fab strategy turns from strength to trap. Intel kept its factories to itself because they were the best and the margins were too good to share, but now the factories are behind, the premium is eroding, and the fabs that were a moat are becoming a liability, enormously expensive to run and a node or more off the lead. A fabless competitor just rents TSMC’s better process and designs circles around Intel. Intel cannot rent anything; it is married to fabs that are no longer winning. The thing that made Intel invincible is now the thing it cannot get out of.
There is a version of the next few years where Intel claws back. 10nm works now, a 7nm node is on the roadmap, and a company with Intel’s resources and talent does not stay down forever; counting it out has burned people before. Perso, I think the recovery is real but slow, and slow is dangerous when TSMC is pulling away and AMD is taking the server room. The deeper problem is not one bad node. It is that Intel built its margins and its strategy and its whole sense of itself on a manufacturing lead it assumed was permanent, and that lead is gone. Getting the transistors working again is the easy part. Figuring out what Intel even is when it is no longer the best fab in the world is the part nobody in Santa Clara has an answer for yet.