Intel May Ditch TSMC For Its Own Foundry To Build Nova Lake Chips
Up to this point, the prevailing expectation among leakers and supply chain watchers was that Intel would split Nova Lake production between its own fabs and TSMC's. Most of the major rumormongers agreed that Intel would manufacture many of the non-CPU tiles itself, but that the CPU compute tile would remain at TSMC for a substantial portion of the lineup. KeyBanc now argues that the balance has swung dramatically in Intel's favor, claiming the company could manufacture as much as 80% to 90% of Nova Lake's tiles internally rather than outsourcing the majority of its compute tiles to TSMC.
The report attributes the supposed shift to two primary factors. First, KeyBanc says Intel's 18A process has continued to mature at an impressive pace, with estimated yields climbing from roughly 65% last quarter to approximately 85%. Now, those figures aren't publicly confirmed, but if accurate, they would represent a substantial improvement in manufacturing readiness, especially in light of the company's new Ireland expansion. Combine that with claims that development of Intel's follow-on 14A node is progressing even more smoothly than 18A did at a comparable stage and it suggests the idea that Chipzilla's process development efforts may finally be settling into a more predictable cadence.
As another element supporting this story, KeyBanc also points to changing competitive dynamics between Intel Foundry and TSMC. According to the firm's report, Intel's improving manufacturing position has made it a more credible rival to Taiwan's finest. As a result, TSMC has purportedly started to offer Intel less favorable wafer pricing and reduced optimization support. Those assertions paint a picture of a customer relationship becoming increasingly complicated as Intel attempts to transform itself into a serious contract chipmaker.
That said, there are some problems with this idea. If the processors are indeed on track for launch around the turn of the year, much of the product validation and manufacturing preparation will have already been well underway. A wholesale decision to move desktop CPU tiles from one leading-edge process to another this late in the development cycle would be a huge shift. Moreover, based on what we have heard about Intel's 18A process, while it is fabulously efficient at lower power levels, it has major thermal density issues at higher power; this is why the story about Intel's 18A-P having some 40% reduced thermal resistance is a big deal.
Well-known Korean "PC Tech Fan" 포시포시 doesn't think much of the idea.
Now, it's entirely possible that KeyBanc is describing a decision made internally some time ago that is only now becoming visible through supply chain checks, or that the report refers to specific Nova Lake variants rather than the entire product family. A reply on the Xwitter thread notes that Intel originally planned to have more chiplets fabbed on 18A but those plans shifted; perhaps those plans have shifted back. It's not completely unthinkable that Intel could have been pursuing parallel development in the case of just such an eventuality.
Besides the talk of Nova Lake being brought in-house, the analyst firm apparently also makes several broader claims about Intel Foundry that carry enormous implications if they prove correct. Besides the confirmed Apple deal, KeyBanc says Intel has secured additional foundry design wins with AMD, NVIDIA, Marvell, Microsoft, Micron, and OpenAI. None of those customers have publicly announced such agreements, but if even a portion of those claims are borne out, it would be a dramatic validation of Intel's foundry ambitions and a remarkable turnaround for a business that many observers thought Intel should sell off not long ago.
