Artificial intelligence feels weightless. You type a question, words appear on a screen, and the answer seems to come from “the cloud.”

But there is no cloud without power plants, transmission lines, transformers, cooling systems, land, water, chips—and enormous amounts of electricity.

THE NEW RABBIT HOLES:

https://www.youtube.com/channel/UC5TXgJVGVGU4sqKH7-EZU1w

RESILIENCE RADIO SHORTS:

https://www.youtube.com/channel/UCEVw1dCPnuR3tDoeNzyCagA

In Episode #363 of Rabbit Holes Deep Dives, we investigate the physical machine being built behind the artificial intelligence revolution—and the possibility that AI is becoming one of the biggest energy and infrastructure stories of our time.

The episode begins with a staggering idea: a proposed 10-gigawatt data-center development paired with 10 gigawatts of new power generation in southern Ohio, with at least 9.2 gigawatts proposed from natural gas. A computing campus with an electrical appetite comparable to millions of homes.

That changes the question.

For years, we measured AI by intelligence: How well can a model reason? Code? Generate images? Create video? Solve scientific problems?

But the next AI benchmarks may be measured in megawatts, transformers, transmission capacity, turbines, cooling systems, water, land, construction time—and money.

Data centers already consume a meaningful share of U.S. electricity, and projections examined in this episode show that share potentially rising dramatically by the end of the decade. But projections are not destiny. Proposed data centers can be canceled. Chips can become more efficient. Algorithms can improve. AI demand could disappoint investors.

That uncertainty creates one of the strangest infrastructure problems in modern history.

Build too little, and the electrical grid could constrain a technological and industrial boom.

Build too much, and communities could be left supporting expensive infrastructure built for demand that never arrives.

We follow the money into an AI infrastructure race involving hundreds of billions of dollars in capital spending. Then we follow the electricity.

Why are AI developers increasingly pairing data centers with dedicated power generation?

Why is “time to power” becoming a competitive advantage?

Could the defining AI companies of the future start looking less like software companies and more like industrial conglomerates controlling chips, land, cooling, electricity contracts and generation?

Then the grid itself enters the story.

Large AI workloads can behave differently from traditional electrical demand. Thousands of processors operating together can create rapid changes in load. Suddenly, software has an electrical signature. One engineer sees artificial intelligence. Another sees megawatts. Both are looking at the same machine.

Texas provides an even bigger warning signal. The episode examines reports of roughly 474 gigawatts of proposed new electrical demand—around 90 percent associated with data-center projects—and why that enormous number should be treated as a planning problem, not a prediction.

Which projects are real?

Which are speculative?

How much infrastructure should utilities build?

And who pays when the forecasts are wrong?

If technology companies require billions of dollars in new generation, transmission and substations, should households and existing businesses help finance it? Or should the companies creating the demand bear the infrastructure cost?

Then another resource enters the rabbit hole: water.

And that reveals the larger story.

AI began as computation.

Now it is becoming electricity, water, land, natural gas, nuclear power, transmission, industrial policy, local politics and geopolitics.

But there is another possibility.

What if AI infrastructure can actually help the grid?