Description du poste
Most people in the AI and rationalist communities don't understand the US power grid, and this misunderstanding leads to poor thinking around the AI power constraint and US/China competitiveness. The US is entirely capable of supporting at least 100 GW of AI data centers by 2030. In 2025, the US had over 1300 GW of operational power production capacity, and more than 200 GW of new capacity will be added before 2030. The power constraint is the product of regulatory malaise, not physical limitations.
Power constraints shouldn't be assumed to limit AI progress in a situation where AI development has strong political support. Furthermore, while AI safety might be best served by not building AI data centers at all, the second best alternative is concentrating construction under a single regulatory regime like the US. Domestic political action becomes both more appealing and more effective if compute is concentrated in the US. The more dispersed global compute becomes, the more difficult international compute governance will be to implement.
The AI safety community has a vested interest in ensuring that AI data centers are built in the US, and one of the biggest problems with building data centers domestically is access to power. Until the political nature of the power constraint is understood, it can't be remedied.