AI data centers are using more electricity than ever, and the power grid may be becoming one of the biggest constraints on how quickly artificial intelligence can scale.
In this video, I break down the real energy cost of AI and what’s happening behind the scenes inside modern data centers. We’ll look at why GPUs like NVIDIA’s H100 and Blackwell chips require so much power, why liquid cooling is becoming essential, and how rising AI electricity demand is putting pressure on power grids around the world.
Global data centers already consume roughly 2% of the world’s electricity, and demand is expected to keep climbing as companies build larger AI infrastructure. But adding more GPUs isn’t enough. These facilities need enormous amounts of electricity, cooling, land, transmission infrastructure and access to the grid.
We’ll also look at some of the solutions companies are exploring, including nuclear energy, small modular reactors, renewable power, more efficient cooling systems and even the idea of putting data centers in space. Amazon, Meta, Microsoft and Google are already making major investments in securing power for future AI infrastructure.
AI may feel like software, but its future increasingly depends on very physical infrastructure.