Episode summary
Dylan Jovine | AI vs The Power Grid: The Real Bottleneck to Growth
Summary
Dylan Jovine argues that the real constraint on AI growth is not model performance but electricity. He cites a Bloomberg report projecting global data center power demand at 1,596 terawatt hours by 2035, with 430 terawatt hours in the U.S. alone, and says those figures make the scale of the problem unmistakable. He also says more than 600 terawatt hours of additional electricity will be needed by 2030, enough to power about 60 million homes.
He says the bottleneck is already visible in infrastructure that cannot connect to the grid, pointing to a 48-megawatt data center in Santa Clara that has sat empty for six years. Dylan says Google has warned it could take about 12 years for some new data centers to get grid access, and he frames that delay as a major obstacle to U.S. AI competitiveness.
Dylan argues that China can move faster because of centralized decision-making, while the U.S. faces local politics, regulation, and grid limits. He says the emerging response is for AI companies to build their own power centers and electricity supply, and he presents that shift as a major investment theme with both opportunities and risks.
Key takeaways
- Dylan Jovine says AI growth is capped by power supply, not just chips or software.
- The Bloomberg figures he cites call for 1,596 terawatt hours of global data center demand by 2035 and 430 terawatt hours in the U.S.
- Dylan says a 48-megawatt Santa Clara data center has sat empty for six years because it cannot connect to the grid.
- He says Google has warned it may take 12 years to get new data centers connected, underscoring the scale of the bottleneck.
- Dylan argues the U.S. may need private power infrastructure for AI firms to compete with China more effectively.
Notable quotes
“Global data center power demand is expected to rise to a record 1600 terowatt hours by 2035”
“AI growth is contingent on generating more power, which again is something that we've talked about many times”
“they have to wait basically 12 years to get new data centers connected to the grid”
“the president's doing now is allowing these companies to start building their own power centers, their own grids, get their own electricity to supply this”
“energy and AI growth has really been something we've covered since 2022 when we first started to really understand this”
How the episode opens
Good morning. This is Dylan Jven with Behind the Markets. Happy Wednesday.
Today's Wednesday, February 18th. What a great day it is to be alive. Anyway, today I'd like to talk about a new report out from Bloomberg.
They outlined the coming power needs of AI because, you know, we talk about it, you've heard me talk about it here dozens of times, how much power that AI alone needs. Very, very important. But these new numbers put it in stark relief.
I mean, it's astonishing. It's almost funny actually how big some of these numbers are. So, let me go down the line here.
Global data center power demand is expected to rise to a record 1600 terowatt hours by 2035. 1,600 terowatt. It's actually 1596 terowatt hours by 2035.
What is a watt hour? What does that even mean in English? I must have missed that class in high school.
So a watt hour is the energy expended by one watt of power running for one hour. So think about it this way. When I go home later, I turn on my lights.
I have a 100 watt light bulb. I flip it on. A 100 watt light bulb used for one hour consumes 100 W hours.
Think of it that way. Simpler way. So what's a terowatt?
A terowatt, and this is funny, is one trillion watt hours. A trillion watt hours. One terowatt.
We are going to need 1,596 terowatts. 1,596 trillion watts of power. That's insane.
That's just global power center demand. In the US alone, it's expected that we will need 430 terowatt hours. So about a third of that is coming to us.
China is expected to rise to 397 increased carowatt hours and Europe is expected to surge 303%. Now why do we need all of these data centers?
From the episode’s captions: roughly the first two minutes, lightly cleaned. Keep watching on YouTube →
Topics
- ai power demand
- data center electricity
- power grid bottleneck
- energy infrastructure buildout
- china vs us energy regulation
Companies and tickers mentioned
- Bloomberg
Watch the episode
Watch “Dylan Jovine | AI vs The Power Grid: The Real Bottleneck to Growth” on YouTube
Summary, takeaways and quote selection by Octagon. Quotes are short excerpts; the full conversation is in the episode.
More from Behind The Markets Podcast
-
Nvidia CEO Out: 4 Reasons I’m Not Panicking About Nvidia -
The Mag 7 Sell-Off: My 3 Secrets to Picking Stocks -
Morgan Stanley’s AI Winners List Is Full of Bank Stocks -
Anthropic IPO: Why I’m Not Rushing to Buy at $1 Trillion -
Chip Stocks Sell-Off: Why the AI Build-Out Just Went From 3 to 5 Years -
Midterm Election Results: The Stock Market’s Winners and Losers
Behind The Markets Podcast, this episode and its artwork belong to their owners. Octagon is not affiliated with the show; this page is an independent summary.
