Why Anthropic spending 75% of compute on training raises rivals' bar
menhguin · x · 2026-10-21
Reacting to the claim that Anthropic allocates only 25% of compute to inference, the author argues the split actually works in Anthropic's favor.
If $75B goes to pre/post-training and $25B to inference, any company wanting a comparably competitive model must spend heavily too. With the reverse allocation, the pool of potential competitors would be far larger.
Related event: Analysis: Anthropic's Compute Split Raises the Bar for Rivals(2 posts)→
More from Venture
- French AI firm Pigment raises $250M at $2.3B valuation, with Nvidia joining the round — dinabass · 2026-10-21
- Musk claims SpaceX ARR could hit $500B–$1T by end of 2027 — XFreeze · 2026-10-21
- 4.5M TikToks analyzed: going viral is neither luck nor format, data shows — letandrewcook · 2026-10-21
- Super Micro guides FY27 Q1 revenue to $18.5–19.5B on AI demand, crushing prior guidance — firstadopter · 2026-10-21
- Digital pathology AI startup Proscia acquired by Agilent — HealthcareAIGuy · 2026-10-21
- By Elon's own math, SpaceX trades at a forward 2027 P/E of around 5 — JOBhakdi · 2026-10-21