AI Needs So Much Power, It’s Making Yours Worse

3 comments

1. Maybe an electrical engineer can chime in here, but I think there are two potential causes of the bad harmonics mentioned here. One is increased demand, mostly from data centers, which the article focuses on. But the other is a transition away from spinning turbines to inverter based power- that’s solar, and batteries but also some wind. Spinning mass provides inertia, and with less inertia, the power becomes… fucky. This gets into esoteric concepts that can only be explained with calculus.

   This can be managed with capacitors, or with flywheels. “Grid forming” inverters on solar farms are a type of distributed capacitor bank. One idea, which I love, but which is probably impractical, is simply to leave coal fired generators connected, but stop burning fuel and chop off the turbine. The generator becomes a motor, and its inertia exerts a pushing or pulling force on the grid that smooths it out.

   I hope an electrical engineer can explain this properly. Bloomberg is among the best sources of energy news, but I’m not sure that even their writers are able to wrap their minds around active vs reactive power in three phase AC.

2. AI is the next DotCom bubble. When people say they need 1GW of capacity I can’t help but think what the hell could AI be doing that warrants that much energy.

3. Datacenters, the onshoring of key manufacturing, electrification (transport and industry), and natural gas replacement drive new electricity demand. Of course, this is happening in parallel with the transition from highly available fossil generation to low-availability green generation, which may put us into an over-constrained problem. The US will likely need another non-emitting, highly dispatchable generation source.

Comments are closed.