Anthropic's 2-trillion-dollar IPO collides with Dario's call to slow down AI
量子位 · wechat · 2026-09-16
Dario's September 12 essay "We Must Pace the Frontier" urged the industry to coordinate slower AI capability gains — and triggered a Silicon Valley backlash:
- Michael Burry fired back: the premise that models are already dangerous doesn't hold; industry-wide slowdowns favor front-runners like OpenAI and Anthropic; "powerful enough to be dangerous" is itself the hype an IPO needs, and pre-builds an excuse if growth stalls.
- Kevin Bass traced the money: early investor Dustin Moskovitz funds both Anthropic and AI-safety groups like METR and Redwood Research; METR — which Dario flagged as an independent evaluator — overlaps with that network (METR says it takes no direct AI-company money but uses copious free tokens); the same ecosystem funds the Tarbell Center, which trains journalists covering AI risk. Bass calls it an "AI Doom Machine" — a self-reinforcing loop of valuations, risk narratives and regulation that favors incumbents.
- The backdrop: Reuters reports Anthropic is preparing a record IPO of up to $100B at $2T valuation, with Nvidia in talks to invest up to $10B — up from $96.5B valuation in May. It has also floated a $30T TAM and committed $100B+ to AWS over a decade, plus $30B in Azure deals.
Related event: Anthropic Warns of AI Risks While Racing Toward Record IPO(2 posts)→
More from Companies & People
- Synthesia opens 50,000 sq ft NYC HQ as US drives half its revenue — synthesiaIO · 2026-09-19
- Daily AI brief: Grok Voice Transcribe 2.0 lands, Meta Muse opens to developers — testingcatalog · 2026-09-19
- Chipotle turned to Palantir after a record $25M food-safety fine over sick employees — aakashgupta · 2026-09-19
- PyTorch Conference to showcase CRCR, letting downstream repos plug into upstream CI — PyTorch · 2026-09-19
- iPhone upgrade cycles stretch to 4 years: Apple's incentives slow old phones without sabotage — aakashgupta · 2026-09-19
- First Cafe Compute Nairobi meetup demos Cerebras API, calls for an /explain interpretability endpoint — paw_lean · 2026-09-19