If Intel has learned anything from management consultants, it’s that it’s never too early to sell the next project. It’s better to sell the sizzle than to sell the steak. At the first Intel Foundry Direct Connect confab, the company added the Intel 14A process (sizzle) to its public roadmap. It also affirmed that its ambitious five-nodes-in-four-years initiative is on track, but it has yet to deliver most of the 5N4Y steak.
We trust that CEO Pat Gelsinger will deliver but need verification, which can only come after the next few Xeon and Core Ultra generations ship. It will take until the end of 2025 before 5N4Y has clearly succeeded, failed, or somehow fallen in between. So far, the Gelsinger regime has delivered only one new process: Intel 4, the second of the five nodes. It’s the technology used for Meteor Lake’s CPU die but few (any?) other designs, indicating it’s an intermediate step toward Intel 3. Although its predecessor, Intel 7, is the first node, it’s a renaming of the Intel 10 nm Enhanced SuperFin process that predates Gelsinger and is—as the name suggests—a revision to the company’s 10 nm technology.
Meteor Lake also includes TSMC-fabricated dice, such as for GPU functions. Various reports indicate that TSMC will make future GPU and CPU dice employed in some Lunar Lake, Arrow Lake, and Nova Lake PC processors. We believe sourcing from TSMC is a result of decisions made before Intel’s fabs were organized to meet GPU designers’ requirements for quick changes and in an era of low confidence in Intel’s process-technology development. For Gelsinger, who learned the business at Andy Grove’s knee, it would be an easy decision to approve a backup plan.
That said, patting yourself on the back for 5N4Y while using TSMC is a bad look. Instead of turning to consultants for inspiration, Intel would be better off learning from software companies. Instead of steak, they know that to succeed they must eat their own dog food.