Intel 7 launches this month in Alder Lake, and the first thing to know about Intel 7 is that it is not a 7nm process. It is the 10nm line, the one that arrived four years late and cost Intel the lead, refined into a strong node and then renamed to stop the bleeding. For years Intel’s 10nm sat next to TSMC’s 7nm in every comparison and looked behind on the number while being roughly equivalent in reality, so Intel did the only sensible thing left: it threw out its own naming and relabeled the node to match what the rest of the industry calls the same density. Intel 7. The label finally matches the density, which is all the rename was ever for.
The good news under the rename is that Alder Lake is the best thing Intel has shipped in years. The hybrid design, big performance cores next to small efficient ones, finally gives Intel an answer to the core-count beating AMD has been delivering on TSMC’s process, and Intel 7 is mature enough to clock it high and ship it in volume. After half a decade of 14nm refreshes, a node that actually moves is almost shocking. Intel is not back in front. But for the first time since 10nm broke, it is moving forward again, on schedule, with a product people want.
The deeper story this year is not the chip, it is the new boss and the strategy he brought. Pat Gelsinger came back to run Intel in February, an engineer who grew up inside the company during the years it could do no wrong, and he has done something the old Intel would have found unthinkable. He has decided Intel will become a foundry. IDM 2.0 is the name for it: Intel keeps designing and building its own chips, it uses outside foundries like TSMC when its own are behind, and, the radical part, it opens its factories to outside customers and builds their chips for them. The company that spent forty years considering foundry work beneath it is now asking for the business.
Through every one of the dominant years, Intel’s answer to becoming a foundry was the same flat no: our fabs are the best, our margins are too good to share, why would we build other people’s chips. Now the fabs are not the best, the margins are not what they were, and the only way to fill enormously expensive factories that have fallen off the lead is to find outside customers to help pay for them. Necessity did what argument never could. Intel Foundry Services launches this year as a real business with a real pitch: a Western alternative to TSMC, government-friendly, ready to build your chips on American soil. Against a TSMC that already builds more than half the world’s foundry silicon and a Samsung that is the only other name at the leading edge, IFS starts from zero customers.
The plan to make any of this credible is the most aggressive roadmap in Intel’s history, and that is saying something. Five nodes in four years: Intel 7 now, then Intel 4, Intel 3, Intel 20A, and Intel 18A by 2025, regaining the leadership Intel lost and doing it faster than anyone has ever advanced a process roadmap. It is a sprint designed to convince two audiences at once, the customers who need to believe Intel’s fabs will be worth using and the investors who need to believe the turnaround is real. The roadmap is either the comeback of the century or a set of promises Intel cannot keep, and right now, in late 2021, there is no way to know which.
I want to believe it. A second leading-edge foundry that is not in Taiwan is good for everyone, the strategy is finally honest about where Intel actually stands, and Gelsinger is the rare leader who understands the factories down to the physics. Perso, though, five nodes in four years is a schedule Intel could not hit even when it was the best manufacturer alive, and it is promising to hit it now, from behind, while also standing up a foundry business from nothing. The rename was the easy part, a number on a slide. What comes after is the part that cannot be relabeled: proving a company can rebuild the one thing it cannot fake, a factory that actually leads. The metronome is ticking again. Nobody knows yet if it keeps time.