Frontier labs spend ~0.1% of $360B capex on safety, analysis claims
AlexTensor · x · 2026-09-15
An analysis highlights the gap between frontier labs' rhetoric on civilizational AI risk and their actual spending:
- Only 3.4% of combined frontier-lab workforce is dedicated to safety
- Just 0.1% of combined capex ($280M against a $360B infrastructure buildout) goes to safety research
- For comparison: nuclear power spends 30-40% of capital costs on safety compliance, aviation 15% of revenue on safety/maintenance, pharma 25% on trials and QA
The author argues labs like Anthropic and OpenAI calling for pauses and international governance should either stop releasing models or scale safety R&D 100x — selling AI services and doing safety-conscious R&D are separate decisions.
Related event: Frontier AI Labs Spend Only ~0.1% of Capex on Safety, Analysis Finds(5 posts)→
More from Companies & People
- AI inference teams in wild talent war, quote-replying each other to poach — dhruv2038 · 2026-09-15
- OpenAI safety lead pledges to open source much of its alignment research 'in the next month' — eliebakouch · 2026-09-15
- Critics link METR's major funding to later alarm reports, questioning eval incentives — FinanceYF5 · 2026-09-15
- UK minister urges heeding AI warnings as Anthropic flags extinction risk and Jack Clark floats kill switch — nordicinst · 2026-09-15
- Brex CEO: Stop Building AI Agents, Build AI Employees and Sell the Work Itself — petergyang · 2026-09-15
- Anthropic and OpenAI spend under 1% of capex on cybersecurity and AI safety, critic says — AlexTensor · 2026-09-15