Gavin Baker Predicts 100x AI Spend Growth and a Minimum Token-Spend Barrier

On August 24, Gavin Baker, managing partner at Atreides Management, argued that his firm's internal AI spend will be roughly 100x higher in August 2026 than in March 2026 and is still doubling monthly. He concluded that scale now determines winners in AI, that a "minimum token spend" barrier will emerge across knowledge industries, and that compute-poor firms will fall behind. His views were relayed by multiple accounts including @AccBalanced, @imnotharsh and @danielmac8.

Confirmed

Baker predicted 100x growth in internal AI spend from March to August 2026, currently doubling monthly (consistent across multiple posts)
He expects a "Pareto-optimal" equilibrium among compute-efficient humans, cheap open-source tokens and frontier tokens
He called attention/FFN disaggregation (splitting inference stages across different hardware/nodes) a "holy grail" architecture but an extremely difficult networking problem
@AccBalanced relayed the concrete Pareto-optimal serving stack: Agent CPU control/scheduling, tool calls, GPU prefill (KV Cache) and Cerebras ASIC FFN decode, claiming AWS is technically ready while other clouds lag

Unconfirmed

The "AWS ahead, others behind" claim comes from a secondary post without first-hand evidence
Whether monthly doubling can persist through August 2026 remains a forecast

Why it matters

If the minimum token-spend barrier holds, compute inequality translates directly into competitive gaps in knowledge industries, reshaping investment theses
If AFFN disaggregation and the Agent CPU/GPU division of labor materialize, it would reshape inference hardware choices and cloud vendor competition

2026-08-24 ~ 2026-08-24 · 5 related posts

Primary sources

Gavin Baker: AI spend to grow 100x, highlighting compute inequality in knowledge industries — GavinSBaker · 2026-08-24 02:15 (UTC+08:00)
AFFN disaggregation is the holy grail of inference networking, says Gavin Baker — GavinSBaker · 2026-08-24 04:05 (UTC+08:00)
AI spend to grow 100x as open-source and frontier tokens find balance — imnotharsh · 2026-08-24 02:47 (UTC+08:00)

[source] Gavin Baker: AI spend to grow 100x, highlighting compute inequality in knowledge industries — GavinSBaker · 2026-08-24
[source] AI spend to grow 100x as open-source and frontier tokens find balance — imnotharsh · 2026-08-24
Infrastructure shift: Agent CPU scheduling & GPU prefill define new 'Pareto optimal' — AccBalanced · 2026-08-24
[source] AFFN disaggregation is the holy grail of inference networking, says Gavin Baker — GavinSBaker · 2026-08-24
AI Spend Doubling Monthly: Fund Manager Predicts Pareto Optimal Balance of Humans and Tokens — daniel_mac8 · 2026-08-24