$2.5B ARR per capability point: Ben Todd rebuts the AI slowdown argument
ronbodkin · x · 2026-11-09
Ben Todd pushes back on Ramez's claim that Anthropic now needs 8x more money per unit of AI capability gain than two years ago, calling it misleading:
- Measure in dollars, not abstract capability: a Claude at AECI 130 could earn $300M revenue, 150 $10B, 170 $60B
- The latest 20 capability points were worth $2.5B ARR per point, vs only $0.5B per point for the 20 before that — marginal dollar value per capability point is rising
- This explains why Anthropic's revenue has grown faster than its costs
His conclusion: whether the AI flywheel continues should be judged by dollars in vs. dollars out, not dollars in vs. ECI out.
More from AGI Musings
- A week toward the singularity: Anthropic speeds up 2x, Claude does 3-day lab work in 90 minutes — rand_longevity · 2026-11-09
- Anthropic's amended S-1 warns of imminent 'dramatically accelerated' AI-driven R&D progress — TheZvi · 2026-11-09
- METR: Anthropic's AI R&D is accelerating at ~2x, recursive self-improvement possibly under a year away — Don't Worry About the Vase (Zvi) · 2026-11-09
- Perplexity CEO touts model-neutrality as crypto's 'credible neutrality' idea reaches AI — dnlkwk · 2026-11-09
- InstaDeep CEO: Anthropic's unreleased Models 3 and 4 sit halfway to AI2027's Agent-2 — connoraxiotes · 2026-11-09
- Nature review: AI has transformed maths — and may outgrow human insight — LucaAmb · 2026-11-09