1:07
The real bottleneck for scaling AI isn't chips or talent, it's electricity. Getting power to a new data center takes months or years, a metric the industry calls "time to power." That's what investors and developers now have to plan around.
The real bottleneck for scaling AI isn't chips or talent, it's electricity. Getting power to a new data center takes months or years, a metric the industry calls "time to power." That's what investors and developers now have to plan around.
Thank you for all being here early on Saturday, and I want to start with a question for everyone in this room and online. Everybody has used the AI model this week. How many of you know where the electricity for the model came from? My question is, how many of you know where the electricity for the model comes from? I guess most of you— I heard most of the US power systems come from nuclear power. Because until about 2 years ago, nobody needed to know. But today, the single biggest constraint on scaling AI is not the chips, and it's not talent. It is power. And more precisely, it is how long it takes to get power to the place. you want to build. In the industry, we call it time to power and has become the number every data center developer, every energy developer, and every investor is now trying to predict.
