Ben Bajarin: Agentic AI will spawn an 'agentic native' CPU tier in datacenters
BenBajarin · x · 2026-09-20
Ben Bajarin highlights his 'Secret Agent CPU' thesis from March, predicting a coming CPU supercycle driven by agentic AI.
The core argument: the shift from monolithic LLM inference to multi-step agentic workflows structurally changes datacenter compute mix. When agents call tools, query databases, wait for human approval, or orchestrate sub-agents, GPUs sit idle while CPUs do the work — an idle window estimated at 12–22% of total inference time, scaling with agent complexity.
His modeling: a fully loaded CPU rack costs $300K (up to $500K in bullish scenarios) vs $4M+ for a GPU rack, with 7:1 power draw ratio. Allocating 5% of power to CPUs in a 1GW facility yields 2% higher effective token throughput and 3.7% lower cost per token for only 1.7% more capex. Breakeven sits near a 10% allocation. He expects a new 'agentic native' CPU tier to emerge alongside GPU clusters.
More from AGI Musings
- Narayanan, Kapoor, Bommasani launch STABL to bridge polarized AI debate — RishiBommasani · 2026-11-12
- OpenAI, Anthropic and DeepMind job posts signal PMs must write evals and code — aakashgupta · 2026-11-12
- Architecture study sparks monitoring concerns over a multi-hop planning spike — ChowdhuryNeil · 2026-11-12
- PowerShell creator Jeffrey Snover at Ignite: AI revolution is an extinction event for those who don't adapt — dfinke · 2026-11-12
- Worst AI use: signaling you don't care about things you should — Afinetheorem · 2026-11-12
- Stanford HAI blueprint: 56 US agencies report 3,611 AI use cases, evaluation lags — StanfordHAI · 2026-11-12