Nov 12, 2026 · 46 min
Apple silicon veteran bets on CPUs for the AI data center
Apple’s Former Chief Chip Architect on the AI Chip Race
The episode examines whether efficient, CPU-first architectures can challenge GPU-centered assumptions as AI moves from cloud infrastructure toward everyday devices.
- 1Gerard Williams III credits empowered teams and performance per watt—not lone inventors—with Apple’s successful processor development.
- 2NuvaCore is pursuing clean-sheet, CPU-first designs for AI data centers, while leaving its target architecture open to Arm, x86, or RISC-V.
- 3AI tools are already shortening chip modeling work and could accelerate verification and the long cycle of CPU development.
Don't miss
Williams explains how AI has already shortened NuvaCore’s performance-modeling phase and could accelerate verification and CPU development.
The brief
Gerard Williams III traces a performance-per-watt philosophy from his earlier ARM work through Apple’s A-series and M-series processors, framing chipmaking as a team discipline rather than a solo invention.
After leaving Apple in 2019 to found Nuvia, Williams turned to NuvaCore and a clean-sheet, CPU-first response to AI data centers—an argument that efficient CPUs still matter alongside GPUs.
The conversation links cloud-scale computing to smaller language models on phones and laptops, where on-device AI could broaden experimentation and complement larger models in the cloud.
Williams says AI is becoming an extension of experienced engineering teams, shortening NuvaCore’s performance-modeling phase and potentially compressing CPU development, verification, and production timelines.
NuvaCore’s full silicon production is likely two to three years away, but Williams expects advances in CPUs and AI to make language-model capabilities useful in ordinary daily devices.
Weekly email
State of the Pod, every Monday
The week's most-discussed names and stories from podcasts, with the moments behind them. Free, five minutes, unsubscribe in one click.