An article explores direct current (DC) data centers powered by solar and batteries for AI applications. It argues that eliminating legacy fuel-based power components can be faster and cheaper. The core concept involves connecting a DC solar array to a DC battery bank, which then feeds a DC GPU rack, streamlining the energy supply chain.

This approach promises several advantages, including no grid connection, moving parts, or emissions. The authors contend that the key metric is tokens per dollar. They propose optimizing solar and battery farm size to achieve nearly 100% uptime. This model offers a more efficient and cost-effective way to power next-generation AI data centers, potentially driving down costs.

By leveraging solar power, these data centers could potentially provide electricity and heat to neighboring consumers. The article suggests that AI compute will increasingly rely on renewable energy sources. This shift could reshape the economics of AI development, making it more sustainable and reducing reliance on traditional power grids.

Ultimately, the vision is of vast solar arrays directly powering GPU farms. It proposes a future where AI compute is directly coupled with renewable energy sources. This will require significant land and investment. The authors provide a rough heuristic on performance, offering insights into the viability of this innovative approach.