Episode Details

This short feature was produced by Daymia Rousseau, Humanities intern from UCSC, with editing help from Rachel Anne Goodman and Howard Feldstein. What follows is the transcript.

Hello, I’m Daymia Rousseau, a UCSC Intern reporting for KSQD. AI data centers are a booming business, and California is one of the leading states in building them. Along with concerns about the environmental costs, critics are saying the financial burden of these energy-hungry facilities falls unfairly on customers. One of its biggest critics is alternative energy expert, Amory Lovins.

Lovins : “I’m mightily puzzled why 38 states subsidized data centers with tax breaks, and why they’re thought to be economically advantageous. They’re good if you’re in the construction industry; they’re not so good if you’re looking for long-term jobs. These are highly automated operations, and enormous amounts of capital get sucked out of the rest of the economy. ”

Professor Lovins is a physicist, AI data center designer, writer, and lecturer at Stanford University. He’s written several critical pieces on the setbacks of Nuclear Energy & AI data centers, as well as many articles on the best ways to utilize alternative, solar & wind-power energy. He’s concerned about customers bearing the brunt of cost overreach.

Lovins : “There’s quite a bit of political unease around the country, and protests around burdens falling on other customers. Partly because the contract with the data center may not cover all the costs of data extensions and other expenses needed to keep the grid going with the big new load…And also because many of the data centers proposed will not be built, or when built will not thrive…in other words, it looks like they’re set to earn a tenth as much of the revenue they would need to support the enormous investments they’re making.”

So data centers are expected to earn less revenue than it takes to construct them? Currently, data centers consume 1-2% of global energy. However, research from investment companies such as Goldman Sachs project data center’s energy demand will grow 160% by the year 2030. And not even Bill Gates predicts a viable energy plan until the year 2050. So data centers are expected to operate for 25 years, functionally, at an over 100% increase in power usage. And that’s only if AI demand can finance it.

According to Professor Lovins, this financial strategy effectively cannibalizes itself.

Lovins : “One of the reasons the financial community is concerned, and sees AI as a bubble, is that it’s indulging in what’s known as ‘circular transactions’. My company invests in your company, which then buys my products, so we can both afford to do that. But it doesn’t really sustain an increase in wealth, so much as to conceal that we’re not making money the conventional way by selling a product people wanna pay a lot for. And it’s also concerning that 120 Billion dollars has now been financed off balance sheets, so it doesn’t show up in the financial accounts. Between that and circular transactions, it’s really hard to tell what condition the tech companies are in…and that heightens the risk… So I think the concern about unexpected burdens falling on localities is well founded. I would also suggest that you shouldn’t automatically assume that because AI can make various operations efficient and save energy, that means it’s a net good for climate.”

One big worry is the capacity of the grid to distribute heavy energy demand during peak hours. Who will pay for upgrading it? The AI companies or the public? What about blackouts?

Many tech companies such as Google & Microsoft, claim they can fund “green premiums” by 2030 to sustain energy demand through market returns.  However, the fragility of the “AI bubble” in global markets casts doubt on that strategy.

Lovins : “Initially, it was more about making the hardware more efficient. Now it’s more about software and system architec