Intel says AI demand outpaces wafer capacity, can only double output

The interview covered lessons from Intel’s delayed 10 nm node, which was slated for 2016 but did not enter volume production until 2018, and how those delays shaped the company’s approach to new technologies. Intel is now using High‑NA EUV (numerical aperture 0.55) for its 14A process, a tool that costs about $200 million per unit and is roughly twice the price of standard EUV machines, but it can print features 1.7 × smaller and reduce multi‑pass patterning. Backside power routing and gate‑all‑around transistors are also being introduced to improve density and performance.

The discussion also touched on Intel’s transition to a foundry model, the maturity of its process‑design kits (PDKs) for customers, and the ongoing “Hot Topics” meetings where engineers prioritize yield‑improving experiments. Executives acknowledged that, without additional equipment, scaling wafer output further will remain a bottleneck for meeting AI demand.