# Data Center Energy Oversupply and Credit Risks

**Podcast:** Odd Lots
**Published:** 2026-02-02

## Transcript

UKG.
Their HR pay and workforce management tools help business leaders empower their people.
Because when work works, everything works.
Learn more at ukg.com slash work.
Donald Trump is rewriting the Washington rule book and reshaping the global economy.
If you're trying to connect the dots behind the headlines, Bloomberg's Trumponomics podcast is here to help.
I'm Stephanie Flanders, head of government and economics at Bloomberg.
Every week I'll bring you a smart focused conversation with reporters and experts from Washington, Wall Street, and beyond.
Listen to new episodes every Wednesday and follow Trumpenomics wherever you listen.
Bloomberg Audio Studios, podcasts, radio news.
And I'm Joe Weisenthal.
Joe, imagine you are a utilities analyst.
Yeah, fun.
And for years you are laboring in the utility analysis minds.
And you know, we like talking about utilities.
We like talking about energy.
We find pretty much anything interesting.
That's right.
Equal opportunity uh interest people we are, for most of our careers, I think if you were a utilities for a while, some people would say it was a little boring.
A really big part of your job, and maybe I'm wrong, but I just didn't the popular discourse was like talking about yield relative to treasuries, right?
They were seen as sort of bond-like instruments, et cetera.
Maybe a little bit of growth, but roughly reliable, safe haven-ish dividend plays, I guess.
Totally.
And since I know where we're going with this conversation, one of the themes of the last few years has been what I would say is the old industries that were either stable or cyclical, becoming secular in the way they grow.
Mm-hmm.
I think that's right.
So what is happening now is if you were, I don't want to say a lowly utilities analyst, but you know, maybe a sort of forgotten utilities analyst outside of your sector.
Suddenly you are very in demand, right?
Because all you hear about nowadays is the AI build-out and energy constraints on that.
And so obviously a lot of people want to look at it from a utilities perspective.
Totally.
I always think like, what a great luck that some people have in their careers.
You know, you could be an analyst and learn modeling skills and all kinds of stuff, and then you get allocated, and someone gets allocated to uh, I don't know, farm equipment, and another person gets allocated to they wind up in utilities in 2022.
And it's like, man, they're on TV all the time.
My old boss at Business Insider, Henry Blodgett.
It's like he was there as an internet analyst in like the late 90s.
What amazing timing and luck.
And so it's like journalism in the era.
It's the same, it's the exact same thing.
I started out covering airlines, yeah, of all things, but those were interesting.
Anyway, I'm glad to say we do, in fact, have the perfect guests.
So we're gonna be speaking to a utilities analyst, someone who happens to have a very contrarian take on the data center build out and how much energy is actually required.
We've been hearing a lot from people who are very, very bullish on the data center build-out.
So this will be a useful cover point.
I love it.
Okay.
So without further ado, uh, Andy DeVries, head of investment grade credit and head of utilities and power over at Credit Sites.
Thank you so much for coming on All Thoughts.
Thank you.
The pleasure's mine.
So is it great to be a utilities analyst right now?
Even better.
Well, your your Bloomberg News reporter Josh Saul wrote an article about the how much it's changed for be a utilities analyst now that data centers are here.
But to push back, we did have the largest bankruptcy of all time in Enron.
Oh, yeah.
The largest LBO of all time in TXU, which then went bankrupt, and the largest private equity return ever in Calpine, 25 billion, which exceeds Apollo's Lionel trade and Blackstone's Hilton trade.
So we have had a lot of fun along the way.
Have you been a utilities analyst throughout that entire timeline?
How long have you been doing it?
I started with the first pack gas bankruptcy and then went to the second, and here we are with data centers.
So what 25 years?
Wow.
So you really have seen it all.
It is fair to say, so you're absolutely right.
There have been some disasters and home runs and utilities are, you know, um, they do get central in the news, obviously, with the fires that we saw, for example, in California several years ago, and the court trials about allocation of risk in those uh situations.
You mentioned Enron, et cetera.
But it is also fair to say that much of the discourse in day-to-day has been like these are sort of bond-like instruments.
Absolutely.
Before we get into that, just like sort of like talk to us about what a what a normal day is like and when you're thinking about utility before, you know, pre-data centers.
Yeah, pre-data centers five years ago, whatever.
Pre-data centers, you're looking at a lot of rate cases.
Okay.
And then on the federal level, you obviously have the renewables displacing coal, and that's you know, obviously having a big impact now.
So it's it's a lot.
I actually think it's a lot of fun.
My impression was always the policy aspect of it seemed kind of the most important thing to keep track of.
Is that right?
Absolutely.
And that's down to the state level, but also the federal as well.
Yeah.
We've done a few energy episodes, still trying to wrap my head around the sort of patchwork of rules that seem to cover our energy infrastructure.
But anyway.
It's like, okay, I don't I don't know.
You know, it's like, are you talking about a rate board or are you talking about market prices?
It's so hard.
Anyway, especially in a 40 minute podcast to try to generalize that.
But I love the people.
Don't get me wrong, those are my favorite people.
Anyway, Andy, what is the mood like at the moment among uh utilities people analysts, investors?
Wasn't there a conference recently?
Everyone's uh gung-ho on this.
So the biggest conference of the year is EEI in November.
And uh it was packed.
It was standing rolling remowling for some of these presentations.
Wow.
Uh your competitors at CNBC were broadcasting from the floor.
Is that the first time that happened?
Probably a sign of a top.
Uh so yeah, people are very happy.
And it does to quantify it.
Utilities have generally grown around four or five, six percent a year, and then that's moved to five to seven percent a year, and now certain names, which we'll talk about later, are up to eight percent a year, and that's driven by data center growth.
Talk to us a little bit more about that.
So, what is the I mean, I'm like looking at a chart of the XLU ETF.
I don't think it's like done insane, but talk to us a little bit about like maybe quantify the exuberance for us.
So it's like, okay, we we are no longer just in the business of measuring bond proxies and looking at policies, et cetera.
There is a secular growth driver.
Talk to us about like the bull case, and then also how we would see the bull case, sort of like how it's manifesting in the tradable instruments.
Sure.
And you pulled up the graph of the XLU.
Obviously, it's still a very interest rate sensitive sector.
Yeah, that's true.
People are playing dividends, you can find higher yields and other fixed income instruments.
Yeah, right.
So, you know, maybe didn't do so well last year.
Uh, but the industry argues that as this EPS growth rate goes up to the high single digits, mid to high single digits, that it shouldn't be as interest rate sensitive.
So that's the big debate going on with investors right now.
And it's not okay.
I mean, the XLU has done well.
It's gone up.
If you zoom out, it looks pretty good.
It's done well.
That's my impression of a Bitcoin investor, by the way.
Zoom out.
Well, right.
Think about it.
You've gotten the and you've gotten that coupon, right?
So like you get so like you might in normal times just be happy with a coupon.
You're getting coupon plus.
I mean, it worked out.
The Fed went zero interest rates for so long, so utilities are the place to be.
That's just math.
And then as soon as the Fed starts checking up rates, Chat GPT comes on the scene and all of a sudden there's data centers.
So now all of a sudden you're taking the leg up on growth when you don't need to be so interest rate sensitive.
Interesting.
So one of the reasons we wanted to talk to you is because you have that contrarian take on the data center build-out.
And we wrote it up in the uh all lots newsletter, which everyone should subscribe to.
It got a lot of attention.
Your analysis, interestingly, is just based on some pretty simple math.
So why don't you just to start out with, why don't you walk us through the calculations that you're actually making to try to analyze how much capacity the utilities are taking on to actually power data centers?
Sure.
So as as you said, it's pretty simple math here.
So utility it's so data centers now are consuming around 45 gigawatts of power.
And you can switch between capacity and throughput.
I'm gonna stick with capacity.
Okay.
So 45 gigawatts of power, and then there's lots and lots of third-party estimates for where they're gonna be in 2030.
And they center around this, you know, 90, 95 gigawatts.
So you need to add 50.
For 2035, there's a lot fewer estimates.
You come around 160.
Now, these estimates, they you know, they're all over the place.
They get come from sell side banks, they come from consultants, they come from everyone.
BNEF has one.
They're I think one of the best out there.
Uh so we use them a lot.
So that's on the demand side of where you're gonna come out in these.
And then you look at the supply, and everyone talks about the demand, right?
Oh, yeah.
But then you look at the supply, and all these tech bros are too cool to actually look at the supply and do utility analysis, right?
Who wants to be a utility analyst?
You were making fun of us before.
But if you look at this, you look at it.
We're pitying you, but and we realize our pity was misplaced, but we were not making fun of anyway.
Keep going.
So you look at the supply, and these utilities are tracking all these data centers connecting to the grid because they've got to do a lot of work, spend a lot of money on transmission, distribution, new substations, transformers.
It's a lot of work.
But it boosts their earnings growth.
So they're happy to talk about this.
And so you look at where they're at and where they're they see things coming, and they've got around 140 gigawatts of near-term supply.
Now, kudos to the utilities, they break out what's firm, committed, signed, contracted, versus pipeline behind it.
Because there's a lot of double, triple, quadruple counting.
So if you're gonna build a data center in the Southeast, you're gonna tell Duke, you're gonna tell Southern, you're gonna tell Dominion, you're gonna build one.
So that's the pipeline potential.
But looking just at the firm committed, whatever they want to call it, you're around 140 gigawatts.
Now that's you got a PUE adjust that.
So when you connect a data center, PUE.
Yeah, when you connect a data center to the grid, you've got lights, you've got cooling.
Those third-party estimates I gave you are just for raw compute.
Okay.
Why did you split those out though?
Because I mean, all data centers are gonna need to be cooled down, right?
What's the point of splitting it out?
I'm not splitting out, I'm just adjusting it downward because the third-party estimates are just compute.
So if you're connecting to the grid, you're gonna ask for the lights, the cooling, and everything.
Yeah.
So I want to go apples to apples versus the third party.
So what is PUE stand for?
Power usage effectiveness or power usage efficiency.
So they're at 140, so that PUE is down to 110 on apples to apples.
So just to go back, you only need 50 on the demand side between now and 2030.
And the utilities are working at connecting 110.
So the utilities are working on already connecting almost as much as you need by 2035.
So again, just to make sure on the same page, third-party estimates, 45 gigawatts for data centers now, going to 95.
That's 50.
Utilities are working on 110.
They don't give timing for that.
Some of it's gonna be past 2030.
What I'm trying to say is there is a lot of supply of data centers coming, and it's very unclear if there's gonna be demand for this.
So that's that's the issue there.
And then it might be worth pausing that and just saying how we're tracking these things.
Yeah.
So what we do for the demand side is we use the original AI agent.
Do you know what that is?
A Gmail alert.
Oh, the best.
So we anything that's not in our trade pubs, not on Bloomberg News, we get picked up by a Gmail alert.
And so then we get all that in a spreadsheet.
So that's on the demand side.
And then on the supply side, we use Diego.
And that's not a large language model.
That's my junior sitting several blocks west of us right now.
So he tracks all this on utility calls.
Just yesterday, Next Era, moved another two gigawatts from the potential into the committed.
And these utilities are chopping at the bit to sign more and more of these deals.
And I think it's just going to be oversupply.
We're going to overbuild these things.
Just to be clear, the commit the firm commitments, those are signed agreements to actually build this capacity.
Yes.
Okay.
And so I was talking with the CFO of Encore, and they made these comments on their call as well.
They're owned by Sempra.
And I said, you know, no one really believes these demand estimates.
Texas is a walled off market, as you guys know.
87 gigawatt peak market.
That is the one thing I know about energy.
Texas is its own walled off market.
There you go.
So 87 gigawatt peak market.
And the demand estimates are they're going to add 30 gigawatts by 2030.
And I said to the CFO of Encore, I said, there's just no way.
And he said, it might not be 30, but it's going to be closer to 30 than it is zero.
And I said, I just the forward power curves don't reflect that at all.
And he said, then they're mispriced.
So just for the benefit of just for the benefit of your users, you cannot trade forward power in Texas on interactive brokers.
I know that's half of your audience.
That's too bad.
Everyone's like looking up there.
No, but this is great.
And this gives us a bunch of um technical questions to get into their HR pay and workforce management tools help business leaders empower their people.
Because when work works, everything works.
Learn more at ukg.com slash work.
Hey there, Odlots listeners.
As we come into 2026, we are realizing that one thing we're constantly thinking about on the show is how companies actually get built.
Not just like the headline version of that story, but the messy operational reality of it.
Right.
We love messy operational reality of things.
The never ending quest to dive deeper, how companies make it big, what causes one company to succeed, why others fail.
Lots of detail there, how it scaled, the ups and downs, and so much more.
Yeah, and we actually we had them on OddLots back in February last year.
We talked to them about everything from TSMC, NVIDIA, Mars, Hermes, scale, capital structure, the importance of incentives, all of the different, I guess, ingredients that go into some of the success of these names that we talk about every day.
Also, their show actually turned 10 years old in 2025, just like us.
So we're uh I guess the same age in podcast years.
Big year.
Anyway, if you like odd laws, the way we get into various market dynamics, how the economy actually works under the hood.
You'll obviously appreciate and enjoy the acquired podcast.
They do similar work, similar ideas, all focused on the context of individual companies.
So go check out the acquired podcast.
You can find them wherever you get your podcasts.
Let's just keep talking about Texas.
Explain to us kind of what the forward power curves are and how you can back out the implicit assumptions that traders are making based on those forward power curves about how much demand there's going to be.
Sure.
So I mean, obviously, if you're gonna go from 87 to you're gonna add 15 or 30, whatever it is, you'd expect that curve to go higher.
What's the curve measuring?
So okay, you say there's a forward power curve.
So the forward power curve is around the clock, peak or off-peak.
There's three separate curves.
And the difference between peak and off-peak is actually, you know, yeah, narrowed because data centers run 24-7.
So it depends on your North Texas or South Texas, and those are in the high 50s.
But what would we be seeing in the curves for like are is there trading happening at the 2030 tenor?
30 might not be so liquid, but 27 and 28 certainly are.
But there's in other energy markets.
So if you look at NAT gas, for instance, although gas traders are really weird about the futures curve in gas, which I don't really understand.
But if you look at that, you point out that over the longer term, it's downward sloping, which suggests that there isn't going to be as much demand, or maybe there's going to be more supply out in the future.
I love it.
We're morphing into natural gas because that is the main driver of power prices, especially in Texas.
And the forward curve for gas is much more liquid than it is for power.
And the forward curve for gas is inverted.
It goes from 370 to 360 by the end of the decade.
So, as my energy analyst Charles Johnson points out, the bigger driver there isn't data center demand.
We're at six BCF a day there.
A lot of people are about 10, 12, we can get into that.
But LNG exports were exporting 18 BCF a day now.
We're gonna add another 12.
Like you'd think that curve would be at least upward sloping by 25 cents, 30 cents.
By the way, you can trade that in your interactive broker's account.
So that goes into is there gonna be a glut of LNG?
It starts getting outside of my expertise, but uh that's what we heard from a lot of clients last week when we went on the road all over New York City.
This is very interesting.
I actually want to ask another question about the the pure power curve.
But since we are on LNG and then we can get back to the power curve, just setting aside data centers, et cetera, intuitively you would think that what is a growth business in the United States, LNG exports.
And in fact, one of the sort of policy debates around the whole question of building out LNG terminals is it's gonna make gas more expensive for American consumers because now we're gonna be competing with European buyers.
Whereas when we didn't have LNG export terminals, we were just swimming in it because it had nowhere to go, nowhere to go.
So it's very interesting to hear that even with everyone acknowledging a booming domestic demand and B the expansion of international demand, that downward sloping gas curve.
Yes.
Maybe, I don't know, maybe they're reflecting world peace in Europe and Russia LNG is unacceptable to the rest of the world.
That could be a driver, right?
Then they would have to rebuild that pipeline though.
Yeah.
But but back to our original conversation on demand.
The reason I was talking to the Encore CFO and asking him about this is he said he's holding two and a half billion dollars of cash collateral postings from some of that demand.
And he's like, you're not some, you know, Joe Schno startup, I'm gonna build a data center and connect to your grid if you're posting two and a half billion dollars.
And he says, and this is what he said on their earnings call as well.
You know, that's real demand that is coming.
It's it's it's it's material.
I'm sorry, just now to go back to the power curve, which I get is much less liquid out there, but there are trades that happen.
These are price, like why how do you infer volume from price?
Because these are price curves.
We don't get the volume, but it's a it's a yearly curve, and then right before the year starts, it splits into 12 month curves, and then it goes into weekly before.
Well, what I'm saying is how do you infer what expected volume in 2028 is going to be from a price curve?
Sure.
It's it's flat.
It goes up a dollar from here to 2030.
Whereas if you're gonna add 20% to your grid demand, then you'd expect it to go up several dollars.
All right, on the natural gas side, I'd want to see 40 cents, 50 cents.
I see what you're saying.
Okay.
And then just to go back to the fork of your upper 50s in Texas, low 60s for peak, and the data center companies are paying 95.
So Vistra just did a deal off 90 dollars a megawatt hour.
Okay.
For round the clock.
So Vistra contracted out its Comanche Peak Plant in Texas, 95 dollars a megawatt hour.
So big tech is paying a very pretty penny.
You can argue some of that's for the CO2 free aspect of it, and some of it's just a lock in the supply.
So just to go back to the math and your overall argument, I mean you're basically saying that utilities are already committed to building out, I guess, twice as much capacity as is forecast to be needed by 2030.
Yes.
The wildcard to me seems to be the demand forecast, right?
And we're already seeing those change pretty wildly.
I know you mentioned Bloomberg NEF, but you know, they've they've raised their forecast um because of the data center build out.
So they've raised their forecast of how much energy is actually needed.
How much confidence do you have in those demand numbers and how could they change over time?
Moderate confidence, but like look where we're at now.
Like OpenAI built all the Chat GBT using two gigawatts.
All the big tech hyperscalers, they haven't given their 2025 volumes yet.
But if you take their 2024 volumes and then double it, and this is output, so I'm gonna transfer it back to capacity, and you assume a 60% capacity factor, all the hyperscalers combined are around 15 gigawatts.
And that's gotta be over half the data center demand.
So to talk about 95 gigawatts, I mean it's a staggering number.
And then you get more advances in you know Nvidia chip efficiency.
Yeah, obviously, Jevon's Paradox kicks in, you've had numerous guests talk about that.
It's just a lot of power.
A lot of power.
Can you just remind us one gigawatt is enough to power what?
I like these comparisons.
Not only do I find electricity markets and so and market structure and electricity very difficult to wrap my head around.
Even after all of these conversations, I have built no heuristics or intuitions for what these a gigawatt, kilowatt, megawatt, like you say these things, and I know a gigawatts bigger than a kilowatt.
But like what this actually means.
And then the fact that even there we're talking about the difference between a gigawatt and a gigawatt hour, and the like I've yet to develop the sort of intuitions that I haven't seen back to the future.
Yeah.
One point one gigawatts.
Oh, there you go.
We gotta print out, you know, a little table.
Yeah, I need that.
Like the way we used to do for credit ratings for the financial crisis, we need that up there.
And actually, credit ratings are gonna be interesting from a utilities perspective as well.
Can I just ask?
Um, you know, obviously, one of the sensitivities in general, with all things data center and utilities is this view, and I think it's kind of overstated.
Is the average rate payer gonna end up paying for a lot of data centers?
Or we'll raise our electricity bill.
And I understand, like these are complex questions and the math isn't so clear.
And also, from what I understand, the emergence of a data center can actually lower the a consumer's electricity bill because there's just that simple math, which is if there's more buyers splitting the cost of the build-out, then actually your uh price tag can go down.
But in the scenario you're laying out, in which there's a bunch of upfront capital investments and everyone's very excited to build it out, and that crowds wires, and you have to buy transformers and gear and all this stuff.
If the demand is does not materialize as expected, that does sound like conditions in which we could see consumer rates go up.
Absolutely.
So that's what we're spending all our time on.
And it's state by state, and even within the same state, you've got numerous jurisdictions.
So is it legislatively mandated, or is it done by a rate case?
Or in the case of Northern Indiana, have the companies themselves, data center companies themselves, gotten ahead of it and said, we're gonna put in a solution where ratepayers are absolutely protected and get money back.
So you look at NYSOurce, which is a Midwestern utility, they own Northern Indiana Public Service, NIPSCO, and they've got a deal where they've got a inside rate base, they've got a separate Genko.
So they sold, and that Genko is doing a deal with Amazon, and they're gonna kick back a billion dollars over 15 years to ratepayers.
So rather than have a debate, oh, is who's funding what?
It's like done and you get 67 million dollars a year.
And that's the blueprint, that's the gold standard.
Yeah.
Now keep in mind, six months before that Genko was launched, NICER sold 20% of NIPSCO to Blackstone.
So you could argue Blackstone said, hey, let's go ahead and do this.
And then the utility right north of Indiana or northeast of the Indiana is Ohio, and the CEO of First Energy is an ex-Blackstone guy.
So maybe they look at doing a Genko or something like that.
That's pure speculation.
I have no idea.
PackGas, Pacific Gas Electric, they've got a deal where they've got rates in place that protect residential ratepayers.
Uh Amarin has, but a lot of utilities don't.
They don't have these protections.
And the point is, someone, if it turns out that there's an overbuild and there is not as much demand for it, someone's paying for it.
And it either it's going to be the customers or perhaps utility shareholders.
I mean, you just the political risk of having mom and pop bail out, you know, Mark Zuckerberg, Jeff Bezos is just you can't have that happen.
But again, six months ago, this was coming up on the tail end of conference calls.
And now these utility CEOs are having in their prepared remarks.
So I'm pretty confident they're gonna figure it out.
Uh you mentioned Blackstone just then.
I do want to talk about who is currently making a lot of money from um the data center build out, but just to stress test the thesis a little bit more, because it is a contrarian take, and so I think we should ask a bunch of questions about it.
But does it take into account time lags for projects?
So I think, you know, capacity build out in the energy sector is notoriously bureaucratic.
That is one thing that Joe and I do actually know about the sector.
Is it possible that a lot of these committed projects actually take much longer to um get working on the ground than currently forecast?
I think the delays will be on building the new generation, not the data center.
So data center takes two, three years.
Even if that slips to four or five years, the power plants take six, seven years, and as you know, you can't get a G Vernova gas turbine for years and years, which is obviously a bullish backdrop here.
Yeah, talk to us more about that element of it all, because building out the let's if you overshoot on production, then that's a problem in itself.
If you overshoot on production at a time when it's gotten really expensive because there's massive inflation in the construction sector, that's an even greater problem.
Talk to us just about like per any given unit of productive capacity on the utility side, how much more expensive has it gotten and what are you forecasting for that?
Sure.
So to build the combined single gas plant 10 years ago is a thousand twelve hundred a KW to build, then it got to two thousand, then utility analysts like myself is like, whoa, that's insane.
Now we're up to three thousand.
Then it's like, who who's actually spending this?
But that three thousand dollars a KW for a new gas plant compares to the data center itself, the cost forty thousand dollars.
So for big tech to spend another three to lock in their gas prices or their fuel source, it's like it's nothing.
Oh, yeah.
It's the minimum, which goes back to the output, forward power is fifty-five sixty and big tech's paying ninety-five.
In the grand scheme of things, for the cost of data center, it's nothing.
So that's why our utility analysts are just jaw-dropping on how shocking it is big tech's willing to pay these amounts.
What about um if you don't measure uh production of new plants in dollars, but new plants in time?
And again, if you're talking about, okay, well, if we can't get this turbine that is several years ago, we could have got delivered next month.
How much longer are these projects taking?
So if you can figure this out, I think you're alluding to this.
If you figure this out, you can make a lot of money.
Okay.
Because obviously the Inflation Reduction Act had enormous I'm gonna vibe code a uh thing to figure it out.
Keep going.
An Inflation Reduction Act, enormous tax credits for renewables.
Yeah.
And then the one big beautiful bill, obviously clip those if you're not online by a certain date.
So all this data center numbers are weighted towards the end of the decade.
Whereas the new solar is right here right now, crushing power prices.
So you actually want to be a little short power for the next few years and then flip to being long.
And if you could figure out when that flip is, you can make a lot of money in either the Ford power curves or the natural gas curves.
But as far as your original question, as I said, data centers two, three years to build new power plants four or five.
But then you don't need as many new power plants as everyone's saying.
So Constellation CEO said on a call the other day, he said he's the Texas market.
He said 87 gigawatt peak market.
You could add 10 gigawatts to Texas tomorrow, which would be the equivalent of sending every single Nvidia chip for an entire year to Texas and running them 24-7.
That's 10 gigawatts.
Because you could run it right now, existing grid, existing plants for all but 40, 50 hours a year.
We stress tested it.
There are some coal plants that can ramp up capacity factor.
There's plenty of gas plants that can.
So I don't know if it's 40 hours, 100 hours, 150 hours.
But it makes more sense to pay someone else not to run their chemical company, the refinery company for 40, 50 hours a year, rather than have the utilities go out and spend 10 billion dollars connecting far away wind farms.
That's the argument.
We're sort of come in the middle of it, but there is plenty of existing capacity on the grid that could ramp up to meet it.
And then of others, guests have pointed out on ObBots, you know, the peak demand of the grid is 850 gigawatts.
The overall size of the grid is 1200 gigawatts.
And then you're adding 50 gigawatts a year of solar, and then you're gonna start adding 20 gigawatts of gas.
I mean, we're gonna handle it.
I'm not really worried about any brownouts or anything.
Oh, yeah.
Talk to us about regional transmission, because this is something that we hear a lot.
It's not necessarily the power generation that's an issue here, it's the transmission, which the US, you know, seems to struggle with, to put it mildly.
So there's regional markets, MISO, Midwest's uh the mid-continent uh ISO.
These guys retired the most amount of coal.
So I think they're gonna be in the worst shape, and then Texas, and then it depends if anyone builds anything in New England.
New England's got the far away most expensive power prices, $70.
The rest of the country's, you know, five.
I am well aware.
Yes.
I have too, because I live in Connecticut.
So if anyone builds a data center in New England, they're gonna be the tightest.
But after that, it's really MISO and no one's building data centers in Vermont where they occasionally have to switch over to oil and wood, right?
I mean, the ISO New England app, which we all have on our phone, right?
They were getting 40% of their power from uh oil.
Yeah, yeah.
The cold snap the other day.
But anyway, it's tough to talk about New England power without talking politics, and we're not gonna go down that route.
So uh so anyway, transmission is very important because you've got to connect all these faraway renewables to the grid.
You said something that I think is actually kind of important.
There is this narrative meme, you know, people talk about the AI race, US versus China.
And that one of the things I've seen people say, China is going to win because they could just build out power more uh easily than we can.
It sounds like I know you're not an AI analyst, but sounds like from your perspective, we don't know like what it means or who's gonna win US versus China, but then from your perspective, power is not gonna be the decider here.
Not in China, it's not.
But it doesn't, you said like, you know, that we could ship every current with existing capacity, we could put every NVIDIA chip in Texas today, and we could run them for 50 hours.
For with the exception of a few really hot hours in the summer, I think I like the imagery of all the NVIDIA chips going on a field trip to Texas.
But it sounds like to your view, that really isn't gonna be from the US perspective, that won't be the binding constraint.
It's it's gonna be a little tight, but I'm not one of these doomsdayers.
Oh, it's the absolute gating factor, it's all gonna stop.
I was in Shenzhen, China uh last year, and a robot got on one floor, and the elevator went up and it got off another one.
And I was like, what is going on here?
Hey there, Odlots listeners.
As we come into 2026, we are realizing that one thing we're constantly thinking about on the show is how companies actually get built.
Not just like the headline version of that story, but the messy operational reality of it.
Right.
We love messy operational reality of things.
The never ending quest to dive deeper, how companies make it big, what causes one company to succeed, why others fail.
Well, I have good news.
That is exactly what the acquired podcast does.
Ben Gilbert and David Rosenthal pick a company and then explore all the ins and outs of its trajectory.
Lots of detail there, how it scaled, the ups and downs, and so much more.
Yeah, and we actually we had them on OddLots back in February last year.
We talked to them about everything from TSMC, NVIDIA, Mars, Hermes, scale, capital structure, the importance of incentives, all of the different, I guess, ingredients that go into some of the success of these names that we talk about every day.
Also, their show actually turned 10 years old in 2025, just like us.
So we're uh, I guess the same age in podcast years.
Big year.
Anyway, if you like odd laws, the way we get into various market dynamics, how the economy actually works under the hood, you'll obviously appreciate and enjoy the acquired podcast.
They do similar work, similar ideas, all focused on the context of individual company.
So go check out the acquired podcast.
You can find them wherever you get your podcasts.
Is obviously private credit has been a big story for the past few years, but now private credit is getting in on the data center build-out as well.
Um, they're sort of, I guess, getting on your turf a little bit in the public bond market.
But what sort of activity have you seen there?
Sure.
So we think that's where the risk is gonna is gonna happen.
And and frankly, Bloomberg News broke the story on Pimco made two billion dollars on day one loaning to the meta data center in Louisiana.
So they priced 25 billion dollars that debt at 220 over Treasuries and then immediately started trading at 140 and handed Pimco 2 billion.
Great for Pimco, but then everyone else is nice to be Pimco, isn't it?
It's nice to be Pimco, especially the weather in Newport Beach.
But everyone else in private credit is like, oh, these guys just made $2 billion dollars.
We need to start lending to data centers.
And we all know how this ends.
Covenants start falling, rates start falling.
And again, if you're big tech, who cares if you overspend?
Like you think AI is the B all end all, you're gonna overspend.
It's when you get down to the second tier, the QTS's, the vantages of the world, and then you get down to sort of the ones below that, and you get like, you know, the core weaves and the Nebbiuses of the world, and you know, there's a lot of shorts going out on Equinix, and obviously your guest Jim Chenos is, and it's all about the chips.
I'm not gonna get into the chip debate.
But it's interesting, you look at a core weave and they got a $50 billion market cap.
That's a real company.
You're gonna be around for a long time.
But the bond market's saying we want a 10% yield to win you 2030 paper, you might not be a real company.
And if you look at our supply-demand outlooks, we're kind of in the camp of the bond market.
But timing, which you mentioned earlier, Joe, is so key because this data centers you're gonna ramp for a couple of years, and the oversupply is really a 2030 event.
So good luck timing that one.
You said something you talked about that Pimco meta deal, and this question has come up, and I still don't think a totally satisfactory answer to it.
Meta is a very highly rated company.
As you see it as a credit analyst, what is it about the private credit?
You know, they'll talk about oh, it's flexible, et cetera.
But you 220 spread over treasuries is not nothing at all.
And is that really is that 220 spread really like worth it for like oh a little bit more flexibility, et cetera?
Like, what are they paying for exactly in the private credit market that they couldn't get cheaper?
I would think, in the public bond market.
I don't know, but I could speculate.
There's a couple of reasons.
If the better question is why did you put this off balance sheet?
Yeah, all right.
So you've got this state of the art data center with the best NVIDIA chips out there, and you're a tech company and AI is the be all end all for everything.
Did you kick it off your balance sheet because you didn't want to damage your balance sheet?
But the agencies are imputing it.
But maybe quant funds running their screens, they don't impute that.
So maybe that helps.
Or maybe you didn't want the depreciation running through your income statement.
Maybe that helps.
Or maybe you want to walk from this thing in five years.
I don't know.
But one of those is definitely the reasons.
Cause why else would you pay that much bigger spread?
150 bips over their borrowing costs.
But so the key thing is here when you talk about that PIMCO meta deal.
Technically, this is not meta debt.
It is not.
It's a value sheet.
Right.
Okay.
So they create a vehicle.
They're not gonna okay.
That's uh I think that's an important element that they're not just arbitrarily paying a lot more for like a sort of no balance check.
Okay, and if you read the credit docs, they've guaranteed this debt.
Yeah, even if the data center shuts down, but our understanding of the docs is if they sell it, then the guarantee goes away.
And so that would create a little risk.
But back to my utility roots.
Please, what happens to that data center shuts down for rate payers, and they actually have an explicit guarantee from Meta to protect rate payers?
So they have that.
A lot of other utilities don't.
So a lot of states, Louisiana, Mississippi, Tennessee, Texas, they need to do better jobs protecting their ratepayers.
And by the way, that's just one line in the dock that could fall up away in in other new data centers.
And that's what we spend our time looking at.
You mentioned the credit ratings just then.
Um so the rating agencies, they they look at the off balance sheet vehicles, even though it's not officially part of the company's debt.
They impute the lease payments and include that as debt.
I see.
And for Meta specifically, they won't do that until the lease starts when it comes online, but everyone's doing it.
Um, and then I was just thinking, I don't mean to labor this analogy too much, but you know, you started out by talking about all the exciting moments in the history of being a utilities analyst, and one of those was Enron, which I assume means you know, you have some experience with circular deals.
But what do you think about all the uh sort of incestuous financing deals that seem to be happening between all the various players in the data center industry?
You mean we'll buy your equity so you can buy our chips?
Yeah.
Again, I'm on the side of bondholders in that one.
Just look at the market caps and look at the bond yields, and and it's uh explain what you mean by that for people who don't have a Bloomberg core weave goes.
People like me who have a Bloomberg but are too lazy.
So these names are going out and in either OpenAI or Nvidia is going out and buying equity in these neo cloud companies, so then they can go out and either supply the compute to open AI and buy the chips from Nvidia.
So it's all very circular.
And I think the example people used 20 years ago is Nortel was doing this and all the vendor financing.
So there's a little bit of skepticism on that.
Okay, so we can't do a utilities episode.
I know we've been focused on data centers, but we can't do a utilities episode without mentioning nuclear power.
What's it gonna take to actually get you know some capacity from nuclear?
Sure.
So obviously the Vogel plant was the last big nuclear plant came online.
It was supposed to cost 14 billion, it ended up costing 32 billion, it came online 10 years late.
No utility wants to take that risk.
Now everyone's talking about these small modular reactors, and I think that's what you're gonna start seeing is more talk of these.
The only way we think a small module actor goes final investment decision, FID, is if big tech agrees to do two things.
They agree to buy some SMRs and they invest equity in those SMR manufacturers to give them the CapEx to build.
Which sounds like something they would do, to be honest.
For sure.
And I think that's the only way you get one of these off the ground.
And I think if those stocks rally on that deal, they're all shorts because they're already reflecting several of those deals happening.
So the big ones are New Scale and Oaklo, and Sam Altman of OpenAI used to be the chairman of Oaklo, and then he stepped down so they could do a deal.
So something along those lines would happen.
That being said, Donald Trump has talked about doing work with Westinghouse and taking equity ownership to build another AP 1000.
And obviously, President Trump is all about taking equity.
Uh, but none of the utilities in my coverage are gonna build something without some sort of backstop.
We did we did an episode recently with an infrastructure investor, and I, you know, I'm a journalist.
I I look at the past, I don't talk about the future.
But I was put on the spot, and I said, I think if in the next 20 years, gun to my head, we will never have another Vogel.
Or we're not gonna have another project like that uh in America.
And sounds like you agree.
Uh I agree.
I do think you see some SMRs.
Okay.
I mean, frankly, our country's been making nuclear submarines for 67 years.
That's an SMR right there.
So I think that's the way it happens, is big tech goes in and does that for sure.
In 20 years, we'll have you back on to see whether or not well, both of you.
This is my only right or wrong.
I don't know nothing about the future.
This is my only one call.
I just don't think we're gonna ever get them.
Right.
That's we're not gonna get a bunch of those things.
That's right.
I'm I'm with you on that.
And in 20 years, hopefully I can dial in from the beach.
All right.
Or about me.
See you in 20 years then.
No, probably before, because this was a fantastic conversation.
Thank you so much for coming on All Box.
Thanks for having me.
Joe, that was a really fun conversation.
That was super fun.
I love that.
My the opinion or framing that I'm sort of coalescing around is that AI can be simultaneously underhyped and overvalued, right?
And actually, throughout history, that's kind of what we've seen with transformative technology, right?
Yeah.
Like think about the internet bubble.
The internet changed the world, but it was a bubble.
Think about railroads, railroads in the 1800s changed the world.
Yeah, but also a bubble.
So I think that's kind of it's kind of what I think.
The key issue, which Andy and you both touched on is the timing, right?
The timing, and yeah, I mean, I think it's like it's very interesting because of course his argument does it doesn't even, you know, doesn't even rest on any valuations right now.
It's like there's all of this expectation for build-out.
There is a, you know, as he put it, there is a number for the amount of uh the volume of data center demand.
There is an amount that's being built up.
And he's like, the second number looks bigger, and that's going to be a that's going to be a problem.
Yeah.
And the time to your point, like it, I thought, you know, really interesting observation he had is a little bit not tangential to his core idea.
But this idea that some of our energy policies are encouraging a lot of production right now, particularly the expiring solar credits.
Yeah.
At the same time, a lot of this demand is gonna come online in the back end, et cetera.
Um, I do think it, you know, seems like a really it does it definitely seems like a fun space.
It's very far from when we were just talking about like utilities as uh as rate proxies.
So I think for old people to get uh income.
Well, the other thing I was thinking about on the demand side is I think there's a tendency among AI bulls, vibe coders such as yourself.
That's right, to think that demand is just gonna go one way, right?
So there's gonna be more demand for AI because I don't know, every piece of software is gonna be replicated through Claud Code or whatever.
And so power demand is gonna go up as well.
But what we've seen so far is that these things are getting more and more efficient.
They're definitely getting more and more efficient.
Like faster than anyone expected.
Yeah, you know, I this is a little bit um tangential to the point.
But I do think like one of the recurring phenomena that we're seeing across this uh industry is that every I mean, and this is I guess it's a bull case, which is that you know, even the optimists keep getting uh turned out to be too pessimistic.
The pace of, say, like efficiency gains for the cost of processing a token dropping faster than people expected.
This morning we're recording this January 28th, ASML, the big chip uh equipment company, way better than expected.
The evails, the benchmarks for the models, where it's like the optimists say, like maybe it could code at this level by 2027.
Turns out it hits there by like, you know, early 2026, et cetera.
So like if you want to just make I'm not making any case here, but if you just want to like make a bull case, it's like even the optimists um keep getting surprised to the upside.
On the other hand, it's fascinating to hear him say, look at what the markets are saying.
They're not pricing in any of these expectations.
And I was particularly surprised because I didn't realize this, that even like, you know, for all of the talk of LNG export terminals, et cetera, that gas is expected to be cheaper a few years from now than it is right now.
Uh very interesting dissonance uh between that and the popular narrative.
Maybe uh gas traders just aren't vibe coders yet.
That's pretty much then they would understand exactly how much more of this uh we're going to be able to do that.
Just from an energy perspective, though, there is a push and pull factor here, right?
So on the one hand, everyone could use AI and demand goes up, but on the other hand, maybe it gets super super efficient and then demand goes down.
I think that's that's the difficulty.
Or it's just there is a number that's out there with like sort of like some reasonable inferences about where it's gonna go, and it's very high.
And it was actually very striking listening to him talk about some of those super sc uh, the hyperscaler numbers, because it's like, where is it at up, right?
Like, as he pointed out, okay, chat GPT like came out in two gigawatts, etc.
Like, there aren't a ton of chat GPTs out there, right?
And that's one of the most computationally intensive things.
Like, maybe there are reasons to think this is all going to go great.
There's going to be a ton of money made in AI, but you can't really just like get to the number.
I don't know.
I think he um it was uh this was a very useful perspective, just sort of on some of the simple math.
And the math sounds like it's subtraction.
Like, this sounds like what my son is learning about.
The other thing I thought was really interesting was the response to your question about why would you finance these things off balance sheet if you're, you know, this massive cash-rich technology giant.
And the suggestion there was, well, maybe at some point in the future, like five years down the line, you need to get rid of this liability.
You don't want to deal with it.
This is why we did that episode with the guy who has uh, you know, the company doing the legal docs, right?
Et cetera.
This is why um it's pretty crucial to understand some of these things because some of the questions sound like Facebook's or Meta's option to walk away, right?
There's some call option implicitly to walk away, et cetera.
And they uh they how they what they could do, what scenarios and what they would allow to be do is obviously going to be pretty crucial for any investors in this off balance sheet paper.
I did not know until you asked this whole idea that um the ratings agencies, while they don't look at it as debt, they do back out a lease cost, and therefore it can inform their overall um credit sustainability.
Well, the other thing, you know, we touched on this, but there's more and more demand from investors for data center debt, right?
Like the space is getting more for some reason, the space is getting more competitive.
And so naturally, what you see in any other credit cycle throughout history is as demand grows and people are competing for deals, the documentation and the protections tend to diminish.
You know, I I find the existence of hype cycles for debt to be a little bit weird because I get like, oh, I really want to get into AI equity, right?
Because that could a hundred X next year, right?
And it's like, oh, I'm really excited about getting into data center debt because it might pay me 50 bips more.
I find it to be very strange or 100 bips more.
It's like if I have a fixed look, I'm a simple, a simple guy, but if I have a fixed income allocation, all I care about is minimizing downside.
And I don't really care like what sector it is.
I'm not participating in the upside.
You're not gonna get greedy.
You're not gonna get rich.
Like for fixed income, I just don't want to lose my money.
Fair enough.
All right, shall we leave it there?
Let's leave it there.
This has been another episode of the All Thoughts Podcast.
I'm Tracy Alloway.
You can follow me at Tracy Alloway.
And I'm Joe Weisenthal.
You can follow me at the Stalwart.
Follow our producers, Carmen Rodriguez at Carmen Armin, Dash O'Bennet at Dashbot, and Kel Brooks at Kelbrooks.
Now for more oddlaws content, go to Bloomberg.com/slash oddlots for the daily newsletter and all of our episodes.
And you can chat about all of these topics 24-7 in our Discord, Discord.gg slash oddlots.
And if you enjoy all thoughts, if you like it when we talk about the data center build out, then please leave us a positive review on your favorite podcast platform.
And remember, if you are a Bloomberg subscriber, you can listen to all of our episodes absolutely ad-free.
All you need to do is find the Bloomberg channel on Apple Podcasts and follow the instructions there.
Thanks for listening.
Every week, we bring you fascinating conversations with the people who shape markets, investing, and business.
CEOs, fund managers, billionaires, Nobel laureates, traders, analysts, economists, everybody that affects what's going on in the market, whether you own stocks, bonds, real estate commodities, crypto, you really need to hear these conversations.
Sometimes it's behaviorists like Dick Faylor or Bob Schiller.
Sometimes it's fun managers like Peter Lynch, Bill Miller, Ray Dalio.
Sometimes it's authors, Michael Lewis, author of The Big Short, and Moneyball.
Regardless of the conversation, these are the folks that move markets each week.
That's the Masters in Business Podcast with me, Barry Rittholz.
Listen on Apple, Spotify, or wherever you get your podcasts.