#72 AI's Real Bottleneck Is Not Chips. It Is the 100-Year-Old Grid.

You've spoken about AI data centers causing gigawatt scale load swings in seconds.

For people who are not aware of it, what's the impact?

Yeah, I think that's one of the biggest problems because we've always had load growth with electric vehicles came online.

You had renewable energy resources that were attached to the grid.

So over the last 30 years, there's always been load growth, but it has been like 0.5% annually.

So it hasn't been unprecedented.

But with these data centers, I call them AI factories that have came online, the load growth has shifted tremendously, 2.5%, 3% per year increase in the load growth for the utilities.

And it has happened in a short period of time.

And one of the key things is with commercial buildings that have came up in the past, electric vehicles, renewable energy, distributed energy resources, you can model that kind of load because it's still a stable load.

With the AI factories or data centers, that load is very spiky.

And that's because of the ramp up time.

The biggest problem is these data centers are hundreds of megawatts and sometimes gigawatt capacity data centers that are draining power that need that kind of demand.

And they ramp up really quickly, especially when these AI models that are now large language models, when you train them, they require a lot of compute.

So if OpenAI or Anthropic and all of these companies are training the models on a regular basis, there's a ramp up at these AI factories and that happens really quickly.

And that's causing a lot of problems for the grid operators because they cannot model that ahead of time.

So the training piece of the AI is more spiky as compared to inference, which is still manageable.

But the training piece is causing a bit of problem for the operators to deliver power.

And that's one of the key areas that my book also talks about.

And we try to help utilities as well.

In order to support the construction of these massive data centers, you need to have a stable infrastructure in place.

which means the grid infrastructure needs to be maintained, monitored.

Otherwise, things can go very wrong.

Things can start going up.

Utilities also need to create new or construct new infrastructure because you'll need new power plants to deliver power because they are struggling to deliver that kind of capacity because these data centers are coming online within two years.

But the problem is that gap with the power sector because it takes six to seven years to build a new power plant and it takes seven to nine years to build a new transmission line.