I started paying attention to the power grid when I realized that the biggest constraint on AI isn’t compute, but electricity. The more I dug in, the less I understood, and the more I wanted to.

Then I saw that Leopold Aschenbrenner made $1.3B last week betting on Bloom Energy and its solid oxide fuel cells tech. Oracle signed a contract with Bloom to power their upcoming data centers.

It turns out that powering a new data center is a challenging game that combines both science and policy (and often politics). In this piece, I dive into a number of different ways to power data centers, some of which are scalable but logistically hard, others which are novel and promising but whether they are scalable or not is TBD.