Log in
Today, we’re going through a panel discussion from the Brookings Institution on Monday 2 March on why electricity prices are going up across the United States. Brookings talked to four regulators and experts. [YouTube]
So what’s making power prices go up across the US? Here’s Joseph Bowring, who is the Independent Market Monitor for PJM Interconnect.
PJM is the electric grid operator for a large chunk of the northeastern US, from New England out to Chicago. And especially Data Center Alley in North Virginia, which has more data centres than anywhere on Earth. You can guess Bowring’s answer:
It’s data centers and it’s nothing but data centers. And the people who start trying to muddy the waters with EVs and electrification, that’s not true. Absolutely not true. There’s no data to support it. It’s not true. Our data shows that it’s data centers.
Last three capacity auctions, $23 billion in additional costs for everyone else were the result of data centers. And just to put that in perspective, that’s about 40% of the cost of capacity.
Data centres as a significant sector came out of nowhere. Here’s Ryan Wiser from Lawrence Berkeley National Lab:
None of us were talking about AI data centers three years ago. And so I think that to this point of how unique is this situation, one big aspect of the uniqueness is that we have a form of demand that none of us in the industry, hardly any of us in the industry, envisioned until very recently.
Everyone wants a data centre in North Virginia, it’s one of the best connected places in the world. And it just happens to hammer a multi-state consumer market with the costs. And that makes them angry. Here’s Abe Silverman from Johns Hopkins University:
We have the data centers coming in asking in the PJM region. So that’s the Mid-Atlantic region. We’re in PJM right now. Goes basically up to New York, down to the Carolinas, west to Chicago. And that region is seeing somewhere around 13 gigawatts of new load, mostly from data centers, wanting to come onto the grid a year.
And let’s talk a little bit about what that actually means, because I think it’s helpful to unpack this a little bit. The city of Baltimore uses about six and a half gigawatts of electricity every year. So we are talking about adding two Baltimores to the grid every year between now and the early 2030s, if you believe the projections. Put another way, that’s approximately 10 new large nuclear power plants a year, which we we are nowhere there.
And so what we basically are saying is there is tens of billions of dollars of new energy infrastructure that is required to serve this class of customers, these new data centers. And the question becomes, who pays for it?
To be fair, it’s not data centres causing power prices to go up in every state. Ryan Wiser says California’s nearly the opposite and wants more data centres:
California, we’ve had declining load for a long time. Our prices have increased the most. It’s not data centers. Data centers have played no role in increasing the prices in California.
Indeed, the CEOs of the major electric utilities in the state have noted that they would welcome data centers because they would be able to spread some of the fixed costs of their electricity system over more megawatt-hours, which mathematically reduces electricity prices.
In California, the story is largely one of wildfire mitigation and liability insurance and increased distribution and transmission expenditures. The PJM story is a totally different story and the data centers are a big part, not the only part, but a big part of that story. So I think it’s ultimately just that we’ve got states differ, because states differ.
Note that not all of California yearns for the data centres. For example, there’s a Bloomberg story on two empty data centre buildings in Santa Clara in Silicon Valley that anticipate being in the queue for years to get a power hookup. So it’s very dependent on local supply. [Bloomberg, archive]
In the PJM Interconnect area, though, the data centres are the problem. Local citizens may hate data centres, but there’s strong political will to keep the AI data centres around.
The federal government believes it’s in an AI arms race with China. I’m not convinced, but they sure are. The states like getting tax money. Even if sometimes it’s the promise of tax money. So assume nobody’s going to just kick the data centres out.
The biggest problem the grids have is: how do you plan for this? Any number you write down in 2026 will be wrong, whether it’s wrong high or wrong low.
Abe Silverman’s answer is: get the sudden new data centre customers to pay for it:
But it comes back to this question of we just don’t know how much electricity data centers are going to use. Maybe quantum computing is going to come around tomorrow and reduce electricity usage by 50%. Maybe there’ll be more efficient chips. Maybe there’ll be consolidation. Or maybe it’s all a bubble and they’ll all go bust. I honestly don’t know.
But we need to put, our job as regulatory professionals is to put in place a scheme that is robust and protects consumers under any of those various scenarios. And so when you talked about scarcity, we can’t have scarcity in AI, I don’t disagree with that.
But it’s not about scarcity. It’s about who pays. You know, Google, Microsoft, Meta, take your pick. They are all perfectly capable of financing and building new power plants. They are the largest entities on the planet. They can do it. It’s just an unwillingness to pay at this point.
The sudden marginal demand sends costs up for everyone. The grids can’t predict if data centres are going to keep going for years, or just drop offline if AI is a bubble.
So current recommendation by the regulators and experts — make the data centres build their own power plants and hook them to the grid. At least then we’ll have some infrastructure left over when the bubble pops.
