Researchers question AI data centers' 'eye-popping' energy demands

People across the country have concerns over the electricity demand AI is expected to require

But is a skyrocketing electricity demand real or overhyped? Jonathan Cumi remembers the hype of electricity demand from decades ago

Before the dot-com bubble, people believed computers would end up using half of all the

electricity produced in the U.S. And I found that they these guys had overestimated electricity use by a factor of 2000

And what we found across the board was that the factors, you know, the exaggerations were big and almost universally exaggerations

Now, Kumi worries the power consumption hype is playing out once again with AI

The largest electric grid in the country expects to see demand increase by 70,000 megawatts in the next 15 years

That's slightly less than half of the current peak for that zone, which covers several mid-Atlantic states and Chicago

Meanwhile, nationwide, estimates vary. Power sector consulting firm Grid Strategies projects a nearly 16 percent increase by 2029

But how accurate are those forecasts? The forecasts that we're seeing right now are basically like what the tech industry wants to happen and what they're selling to their investors

So obviously they're painting a very, you know, rosy picture of how much AI is going to take off

But the reality is that their financials don't match that picture that they're painting

Those big numbers can also help utility companies get the funding to build more power plants

When a utility gets permission from the Public Utility Commission to build a power plant It means that that utility is going to be able to recover the cost of that power plant plus a commercial rate of return whether or not that plant is ultimately needed

or not. Meaning the bigger projections are good for both the tech and utility companies

There was a collusive arrangement between big tech companies and electric utilities

to intentionally create a sense of hype and panic around the energy consumption of AI

because both had a shared financial interest in doing so. One thing making those projections bigger, duplicate data centers, sometimes called phantom

data centers. When tech companies want a new data center, they file a request with a utility

company. Those utility companies report the number of requests to grid operators, who then use those

numbers to generate projections, but the utility companies don't usually compare those requests

with each other, meaning they might be counted more than once. Many data centers that are talked

about and proposed and in some cases even announced will never get built. It's not just a case of it

will get built somewhere. Many of them simply won't get built at all. Other critics of these

projections say AI has not yet proven to be profitable. For example, the company behind

ChatGPT lost $5 billion last year. If at some point in the next few years these models don't

actually prove their profitability, obviously investors are going to be much less enthusiastic

about them. For Straight Arrow News, I'm Kaylee Carey. For more of Keaton Peter's report on the

energy demand of artificial intelligence, download the Stradar News mobile app today or go to san.com