US vs. China AI Energy Infrastructure
- The US is lagging significantly behind China in bringing new energy generation online.
- China added over 400 gigawatts of power capacity last year, compared to just several dozen in the US.
- By 2028, Frontier AI labs are projected to require 20-25 gigawatts for training alone, and at least that much for inference.
- The entire US AI sector may need at least 50 gigawatts of electric capacity by 2028, equivalent to the power consumption of one Netherlands, Sweden, Argentina, or Taiwan.
- China's rapid expansion allows them to power new industrial plants, AI development, and domestic electrification simultaneously without straining supply, a feat the US is struggling to match.
Geostrategic Value of Open Source AI
- Promote Open Source AI: Encourage open source and open-weight AI models to establish global standards and ensure America has leading open models founded on American values.
- Geostrategic Value: Open source models can become global standards in business and academic research, offering geostrategic advantages.
- Supportive Environment: The government should create a supportive environment for open models, while respecting developers' choices regarding model release.
- Policy Actions: Ensure access to large-scale computing power for startups and academics; partner with tech companies to increase research community access to private sector computing.