Mar 6, 2025 · 55m · catalyst
A skeptic’s take on AI electricity load growth
0:00 / 0:00
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of Catalyst, host Shayle Kann and data center energy expert John Koomey critically examine the narrative surrounding AI electricity demand, demonstrating through historical trends, forecasting methodologies, and physical grid bottlenecks why projections of explosive national power growth are significantly exaggerated.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Shayle holds 29.1% of the talking time here. How this is scored →
Shayle as informed peer 4.4 Guest teaching 4.7 Guest disagreement 2.7 Shayle pushing back 2.6
speaking balance: gold is Shayle, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 45:09 Dismissing long-term data center forecasts as fabricated
Koomey directly rejects the credibility of decade-long electricity forecasts, arguing that anyone claiming to know 2035 data center demand is simply making it up.
Hardest push from Shayle ▶ 46:17 Pushing back in defense of utility planning horizonsKann firmly challenges Koomey's skepticism of long-term forecasts by pointing out that utility infrastructure cycles inherently mandate 20- to 30-year capital commitments.
Biggest teaching moment ▶ 35:00 Explaining the physics of Dennard scaling and multi-core transitionsKoomey walks Kann through the history of chip architecture, explaining how hitting the one-volt silicon threshold around 2000 forced the industry to pivot from clock speed to multi-core parallelism.
Shayle holds their own ▶ 23:00 Correcting the role of electricity costs in data center OPEXKann quickly intervenes when Koomey compares data center power usage to average economy-wide GDP energy costs, pointing out that power represents a massive share of data center operational expenditure.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Shayle as informed peer | Guest teaching | Guest disagreement | Shayle pushing back | Why |
|---|---|---|---|---|---|---|
| Host Monologue: The AI Load Narrative vs. Skepticism | 0 | 0 | 0 | 0 | Introductory monologue where the host frames the current AI power demand narrative and introduces John Koomey's contrarian perspective. | |
| Historical Perspective: Dot-Com Fears to Hyperscale Efficiency | 4 | 5 | 2 | 1 | Koomey recounts historical dot-com overestimates and explains the decoupling of compute growth from electricity use between 2010 and 2018. Kann actively synthesizes these insights into two distinct historical phases. | |
| Baseline Data Uncertainties and Flawed Forecasting Methods | 4 | 6 | 3 | 1 | Koomey breaks down the wide uncertainty in historical baseline data and critiques mainstream forecasts for blindly equating Nvidia hardware sales projections with actual long-term electricity consumption. | |
| Debating Jevons Paradox and the Infinite Demand Assumption | 6 | 4 | 4 | 4 | Kann articulates the industry's Jevons Paradox defense of endless demand and presses on data center OPEX dynamics, while Koomey challenges the premise that demand for AI is truly infinite. | |
| Mid-Roll Sponsor Segment: Bloom, ENGIE, and EnergyHub | 6 | 7 | 4 | 3 | Following mid-roll ads, Kann presses Koomey on whether infrastructure bottlenecks will naturally cap demand growth regardless of infinite demand. Koomey delivers an in-depth tutorial on post-Dennard scaling efficiency levers, algorithms, and application-specific silicon. | |
| Bits vs. Atoms: Supply Constraints and Utility Rate Design | 5 | 5 | 3 | 4 | Kann defends utility planners who are forced to make 20- to 30-year infrastructure bets, prompting Koomey to explain how sophisticated utilities use rate design, upfront capital, and take-or-pay contracts to de-risk load forecasting. | |
| Assessing the 2030 Load Doubling Bet and Summary Outlook | 6 | 6 | 3 | 5 | Kann pins Koomey down on a personal wager regarding whether US data center load will double by 2030. Koomey puts the numbers into macro perspective, demonstrating that even a doubling translates to manageable incremental national demand growth rather than an explosive crisis. |
Statements from this episode (13)
Koomey: Data center compute grew 6x while power rose only 6% (2010-2018)
“We had about a six-fold increase in the compute output for data centers from 20 10 to 20 18, but electricity use went up a grand total of six percent over that period.”
Koomey: IEA data center electricity estimates varied by 50% in 2024
“So for the year, 20, 22, the International Energy Agency did these projections. They did two of them in, in, in, 20, 24, and they started by trying to do history, and they looked at twenty-twenty-two, and the first study they released in January had a number t…”
Koomey: IEA hides data and calculations for its power forecasts
“Whereas some of the other projections, even From pretty credible organizations like the International Energy Agency, you can't actually figure out what they did. They don't release the data. They don't release the calculations in any explicit way.”
Koomey: Analysts wrongly use Nvidia sales projections to forecast electricity growth
“And usually what they're assuming is NVIDIA's business plan is our growth forecast. Right? They're just saying, because for AI, they're just saying, well, NVIDIA has this projection for the next few years of how many AI nodes they're gonna sell, and they just …”
Koomey: Rebound effects from efficiency gains are typically only 10 to 20%
“Yes, the car's a little bit cheaper, and yes, people drive a little bit more, but generally, these effects tend to be kind of small. They tend to be kind of 10 to 20%.”
Koomey: OpenAI lacks an obvious path to profitability without massive growth
“Open AI right now is just a major, it's hemorrhaging cash like you can't believe, right? There's no obvious way that they become profitable without a massive increase in the demand for their services and the ability to raise the cost, and that is really unclea…”
Koomey: AI industry will pivot to massive efficiency gains under infrastructure constraints
“Between those various, ah, improvements that you can get beyond just shrinking transistors and making things closer together, there's orders of magnitude possibility for improving efficiency, and so my sense is that the, you know, the focus has been basically …”
Koomey: Software moves on one-year cycles while physical infrastructure takes decades
“The world of
Bits moves really fast. They have all the money in the world. The world of atoms is slower. So it's like a year for the world of bits, 10 years or 20 years for the world of atoms, and so it's like a mismatch.”
Koomey: Anyone forecasting 2035 data center power demand is making it up
“So I think anyone who claims to know what's going to happen in twenty-thirty-five, the data center of electricity use, I think is just making it up.”
Koomey: US data center power share rose to 4.4% in 2023
“The U the U S number from the report to Congress that just came out from LBL was something like on the order of two percent of the electricity in 20, 20, 20, 21 was data centers. And now in 20, 23, now, 2023, which is the last historical year we have, it's 4.4…”
Koomey: Total US electricity generation in 2023 merely matched 2018 levels
“Twenty-twenty-three was about the same amount of generation as twenty-eighteen.”
Kann: US data center capacity will more than double by 2030
“I was on the side of, we will see total capacity of data centers as measured in megawatts, or gigawatts, I guess, really double by twenty-thirty, and I was on the side that we will see it more than double.”
Koomey: US data center load will double once more then level out
“It would not surprise me if You know, we saw another doubling, but I would say, you know, probably that's about it. I think that's going to level out, and then people are going to be more efficient at what they're doing based on the history”