CPU-inference argument gains a second voice: Notion's "CPUs underrated" case
From the AI digest for 2026-10-17
What happened: On October 17, 2026 at ~2:30 AM ET, nichochar amplified a clip of Notion AI's lead Sarah Sachs arguing that "a lot of things should be served on CPU that are served on GPU," citing "a lot of benefit and determinism in executed code," with a follow-on about agents; his own framing: "CPUs underrated. GPUs overrated. people are dramatically underestimating how much software wants to be running in the future."
What's genuinely new: A development of the Anthropic CPU-inference thread (the Akamai deal storyline), which so far rests on one company's CPU-based distributed-inference arrangement with no detail on which workloads moved. This is the first independent, named second voice in the record arguing the same direction — served-on-CPU over served-on-GPU, on determinism grounds — from a company actually running agent workloads.
Watch: Whether any named Notion workload actually moves from GPU to CPU; whether Anthropic's Akamai arrangement discloses which models or tiers run on CPU; any follow-on numbers from either side.