https://spectrum.ieee.org/state-of-ai-index-2026
Article
-
Stanford HAI AI Index 2026 visualized: capabilities, investment, emissions, adoption.
-
Training frontier models (e.g. Grok 4) generates 72,000+ tons CO2-equivalent.
-
China leads globally in robotics installations, trajectory unchanged since 2012.
-
Software engineer GitHub activity cited as proxy for AI adoption.
Discussion
-
illiac786: capability progress is clearly plateauing across most benchmarks, not accelerating.
-
tqi: 72k tons sounds alarming but is trivial vs. 38 billion tons global annual emissions.
-
eulgro: Claude 4 Opus inference estimated at 23W per medium prompt—10x less efficient models exist.
-
themafia: profits generated by AI conspicuously absent from all graphs.
Discuss on HN
