Overview
- Stanford released its ninth AI Index, a 423-page report that tracks models, money flows, jobs, infrastructure, and public opinion across the industry.
- The report shows China nearly matching top U.S. models with a 2.7-point gap, while the U.S. led 2025 in notable model releases and private investment as Chinese state funds added substantial support.
- The U.S.-led data center buildout has reached 29.6 gigawatts of AI power capacity and is drawing organized local resistance that has stalled or blocked an estimated $64 billion in projects.
- Estimated environmental costs are large, including 72,816 tons of CO2-equivalent to train Grok 4 and annual GPT-4o inference water use modeled to match the drinking needs of roughly 12 million people.
- Studies cited in the report find strong task-level gains but little economy-wide lift, with Penn Wharton estimating a 0.01 percentage point boost to total factor productivity and an MIT study reporting 95% of firms saw no return on AI spend.