AUGUST 10, 2026 — FLI SAFETY INDEX · SEQUOIA $3T REVENUE GAP
- FLI AI Safety Index Summer 2026: Anthropic C+ (2.66) leads 9 labs. OpenAI C (2.28). Google DeepMind C (2.01). Meta D+. xAI, DeepSeek, Mistral F — one each from US, China, Europe. No lab above C+. Key finding: top 4 labs weakened or voided prior pause pledges. Existential safety worst domain — no lab above C-. Grades are institutional policy ratings, not product safety ratings. Full scorecard →
- Sequoia $3T revenue gap: David Cahn: $1.5T AI infrastructure spend in 2026 needs ~$3T revenue to pencil. Anthropic + OpenAI at ~$80B ARR combined. ~$2.9T gap. Apollo warns S&P 500 correction risk if hyperscaler cash flows fall short. Bull case: enterprise adoption early (Palantir Q2 +93%). Bear case: 54% token efficiency gain compresses prices faster than adoption grows. Not investment advice. Full analysis →
Story 1 — FLI: The AI Safety Report Card Nobody Passed
Per the FLI official report, seven independent experts reviewed nine AI labs across 37 indicators and six domains. Nobody scored higher than C+. The most significant finding is not which lab ranked first — it is that Anthropic, OpenAI, Google DeepMind, and Meta have all weakened earlier commitments to pause development when safety thresholds are approached. This lands one week after OpenAI paused Astra because its Preparedness Framework's Critical threshold was triggered. OpenAI acted on its framework in that case. The FLI finding says the broader framework architecture is softening across the industry even as capability thresholds are being reached. The practical use of the index for enterprise buyers: Anthropic leads on documented safety governance. xAI, DeepSeek, and Mistral failed. Use it alongside your own audit — it grades labs, not deployments. Full scorecard →
Story 2 — Sequoia: The $3 Trillion Question
Per TechCrunch's Cahn interview, the methodology: $1.5T infrastructure spend × 2 (for full data center TCO) × 2 (for 50% gross margin) = $3T required revenue. Anthropic and OpenAI sum to approximately $80B ARR combined — a meaningful start, not a solution. The enterprise adoption curve — which Palantir Q2 (+93%), Anthropic's Ode JV ($1.5B), and the entire AI agent deployment wave represent — is Cahn's own bull case variable. The bear case: 54% token efficiency gains mean prices fall faster than enterprise adoption grows, compressing the revenue side. Apollo's Torsten Slok adds the systemic risk: if the four hyperscalers (Google, Meta, Microsoft, Amazon) miss their 2028 cash flow projections, the S&P 500 concentration in those names makes it a macro problem, not just a sector problem. For information only, not investment advice. Full analysis →