AI needs more power. But maybe the bigger problem is how we use it. $NVDA, $GOOGL and Emerald AI just launched the AI Energy Management Alliance. The idea is actually pretty simple. An AI data center might have a 1 GW connection, but it doesn't necessarily need to consume the full 1 GW every second. Some workloads can be slowed down or shifted when the grid is under pressure and ramped back up when power is available. Why does that matter? Because getting new power connected to the grid is becoming one of the biggest bottlenecks for AI data centers. If utilities can trust that a data center can temporarily reduce its load by, say, 100–200 MW, they may be able to connect it without waiting years for the grid to be expanded. That's the interesting part for me. We're not just talking about building more power. We're trying to make compute itself more flexible. And if regulators start giving real value to that flexibility, this could become a pretty important part of the AI infrastructure stack.