Anthropic reportedly allocates 72% of compute to training, not serving
abhiadesai · x · 2026-10-27
Cited data suggests Anthropic devotes 72% of its compute to training rather than serving customers, confirming earlier "75% to training" claims. abhiadesai argues this is rational:
- Better capabilities correlate with exponential revenue jumps — the shift from tab-autocomplete to "SWE in a box" agentic coding is the template, and similar dynamics should play out across verticals and R&D tasks
- So the optimal lab strategy is dedicating the bare minimum serving compute needed to hit revenue targets, pushing the rest into training
- The cost is leaving money on the table, but compute can be shifted back to serving anytime to bridge a revenue gap
More from Infra
- OpenAI's initial Vera CPU results: 51% more Python throughput vs leading x86 — firstadopter · 2026-10-27
- Berth: Open-Source Tool Lets Multiple AI Engines Share One GPU Without Fighting — Distinct-Pie2389 · 2026-10-27
- Cloudflare discloses cross-request KV cache leak in GLM-5.3 on Workers AI: 155 responses affected — michellechen · 2026-10-27
- Cloudflare discloses cross-request KV cache exposure in GLM-5.3 on Workers AI (155 responses) — wondersky · 2026-10-27
- One month fully local on a DGX Spark: daily tasks fine, code still needs Claude to check — michaelthatsit · 2026-10-27
- Cloudflare Workers adds spanContext, setStatus and updateName tracing APIs — irvinebroque · 2026-10-27