Pod Insights Catalyst with Shayle Kann More 2025 trends: DeepSeek, plug-in hybrids, and curtailment About this episode Didn’t catch last week’s episode on Nat Bullard’s mega slide deck on energy transition? Start there.
This is the second half of our extended conversation with Nat, the former chief content officer at BloombergNEF and current co-founder at data insights company Halcyon.
In this episode, Shayle and Nat dig into topics like:
Rising solar installations and stagnating wind
Why we’re wasting so much renewable power amid skyrocketing load growth
The rise of Chinese plug-in hybrids and exports
Whether DeepSeek’s efficiency will temper or turbocharge load growth
The woeful state of transmission buildout, despite demand for it
Why one quarter of Virginia’s power demand comes from data centers
Recommended resources
Latitude Media: Does DeepSeek call the data center boom into question?
Latitude Media: To get data centers online, one Virginia co-op is proposing a new business model
Latitude Media: A dizzying year at the AI energy nexus
Catalyst: Demystifying the Chinese EV market
Reuters: Exclusive: Global solar capacity hits 2 TW on path to climate goal, data shows
Credits: Hosted by Shayle Kann. Produced and edited by Daniel Woldorff. Original music and engineering by Sean Marquand. Stephen Lacey is executive editor.
Catalyst is brought to you by EnergyHub. EnergyHub helps utilities build next-generation virtual power plants that unlock reliable flexibility at every level of the grid. See how EnergyHub helps unlock the power of flexibility at scale, and deliver more value through cross-DER dispatch with their leading Edge DERMS platform, by visiting energyhub.com.
Catalyst is brought to you by Antenna Group, the public relations and strategic marketing agency of choice for climate and energy leaders. If you're a startup, investor, or global corporation that's looking to tell your climate story, demonstrate your impact, or accelerate your growth, Antenna Group's team of industry insiders is ready to help. Learn more at antennagroup.com.
Transcript Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.
0:00 Speaker 00 You listen to this podcast for incisive coverage of the tech, business and market forces shaping clean energy and climate tech. You can get them in your inbox, too. Latitude Media's flagship newsletter, The Latitude, offers B2B analysis, commentary and curated news from across the industry. It's the perfect complement to this show. Subscribe to receive The Latitude daily or weekly in your inbox, or get our AI Energy Nexus newsletter on Wednesdays. Go to latitudemedia.com/newsletter to subscribe today.
0:31 Speaker 01 Latitude Media, podcasts at the frontier of climate technology.
0:37 Shayle Kann? I'm Shail Khan, and this is Catalyst.
0:43 Speaker 03 A training hour is GPU use, which is electricity. So that's it. And if you can do all of this with less, you know, what does that mean for building gigascale data centers?
0:53 Shayle Kann? Coming up, part two of my conversation with Nat Bullard.
1:05 Speaker 00 Catalyst is brought to you by Antenna Group, the communications and marketing partner behind the most influential companies in energy, climate and infrastructure. For three decades, Antenna has helped ambitious companies stand out, build market leadership and become category-defining brands. From breakthrough startups to global industry leaders, their team brings the expertise, relationships and track record to accelerate your momentum. Learn more at antennagroup.com.
1:32 Speaker 04 The power plant for the modern grid isn't a building, it's a network. Energy Hub's derms platform connects batteries, EVs, thermostats and commercial and industrial sites into virtual power plants that over 230 utilities dispatch as reliable grid capacity day in, day out. VPPs deploy in 60 to 100 days at 40 to 60% less cost than traditional infrastructure, and they perform like conventional generation from day one. Learn about the flexibility platform for the modern grid at energyhub.com.
2:03 Speaker 00 Demand flexibility is now an essential part of the data center's speed to power playbook. But how do data centers actually curtail load? Grid Beyond is the end to end solution for flexible energy management. Grid Beyond's advanced DRS and VPP platform is tailored to what data centers care about most. Maximizing reliability, intelligently meeting curtailment obligations, reducing costs, and managing all your distributed resources in one place. Ready to transform your distributed energy management? Get started at gridbeyond.com.
2:37 Shayle Kann? I'm Shayle Khan. I invest in revolutionary climate technologies and energy impact partners. Welcome, or welcome back, rather. If you haven't listened to last week's episode, go back and do that first. That was the first half of my conversation with Nat Bullard, where we run through his panoply of slides and data sets on the state of global energy and climate. Here's the second part. We cover plug in hybrid vehicles, China, electric vehicles, data centers, data centers, data centers, of course, AI models and energy consumption. It's a good one. Here's Nat. We're on to slides, what, 78, 79, 85, 100? Okay. This is a sort of a Tale of Two Cities situation if you're comparing solar and wind in terms of what's going on right now. It's not entirely true because as you point out,
3:22 Shayle Kann? the 2024 was a record year globally for both solar and wind. But certainly, if you look at the US, they are not trending in the same direction. I think people need to appreciate that. So talk to me about what's happening with solar versus what's happening with wind.
3:36 Speaker 03 A modest 599 gigawatts of solar installed last year globally. That's a good number. That is a lot of modules. That is a lot of delivered energy once it's all fully commissioned. And the wind sector, which I think does get painted with this brush, that it's somehow like a huge laggard, did have a significant record year in more than 130 gigawatts that were installed. But nowhere close to the sheer volume of what's happening in the solar sector. I mean, it's just, it's quite something. And if you and I were to back cast this through our analyst selves in 2021, Two thousand and eight and give us this number, we would say like some something truly crazy has gone on or
4:20 Speaker 03 uh you know, we've we've ended up with like a world, a world global carbon tax of $200 a ton on all economic activity. And solar has also gotten down to only, I don't know, 50 cents a watt to install. Uh, you know, it turns out neither those things are actually true and but this is the market that we have and that we've been given. But in the U. S. It's definitely not that case. So uh, Rhodium Group has some great data that they pull in terms of the actual invested dollars per quarter in in in both industrial sectors, but then also in the actual deployed energy that they have and wind is way off where its peak was, which was a gap peak in like early, like mid mid 2020 or late 2019.
5:04 Speaker 03 Uh, solar is just on its way up. You know, it's more than $40 billion uh in the trailing 12 months. And storage now invests more than wind does in the States. And you know, partly that speaks to it speaks to a number of things, I should say. One is that it's just going to be simpler and faster to develop solar than to develop wind. And you know, in particular, in the absence of any particular reform to our transmission system in the United States. Uh, two is that storage is highly complementary to a high solar penetration market, but also meeting the demand of a of a more dynamic grid in general. Uh, and in multiple dimensions, you know, providing service in multiple different dimensions.
5:47 Speaker 03 And so we do see this like very differing trajectories that won't necessarily get solved at a pure market basis. Like they'll probably stay like this. You know, where we should be doing as much wind as we absolutely can. That's going to be a challenge given interconnection and it's going to be a challenge given current politics in the U. S. So we'll have to wait and see where that goes, but you have very much a tale of two markets.
6:11 Shayle Kann? Yeah, but I mean, that's the thing is right. It was already a tale of two markets prior to the new administration. And so solar was up into the right, wind is down into the right. And then the new administration doesn't seem especially favorable to renewables in general, but has it, I mean, Trump has always headed out for wind. That's like a you that's been a long term thing. It's clearly already the case here. So this is likely exacerbating an existing trend, which is unfortunate for the wind and okay for solar probably and storage.
6:42 Speaker 03 That's right. That's right. And like, and you can, you know, you can see it in, you can see it in the world's great uncovered, or the US is great uncovered power market, which is ERCOT in Texas. You know, this year, we're probably going to have more solar installed in toto cumulative than wind. And if you think solar is growing on a rapid trajectory, batteries are going even faster. So you can just see this in the markets where competition is as sort of honest and earnest as it can be, and as unconstrained relatively speaking as possible. You know, so we're just just in takeoff mode and the shape of the wind growth curve is looking not the good kind of asymptotic, like it's looking like it's flattening out. And you obviously want to reverse that. These are complementary resources.
7:26 Speaker 03 In particular, when you add storage to solar and wind in a grid like Texas, or really any place that's sunny and windy, it's highly beneficial.
7:35 Shayle Kann? Well, let's talk about one of the results of all that solar and wind, though, which is, which is interesting data that you put in here in slides 86, 89, 98 on curtailment and negative pricing. So add lots more solar and wind to the grid, you're going to get periods where you're oversaturated on them. That's how things work unless you add enough storage. And it just clearly in the places that we have a lot of penetration, we have not added enough storage yet, or for that matter, enough flexible load that can soak up that excess energy to the point where we continue to just dump more and more power literally into the ground, at least in California, and we see negative prices at an alarming frequency in Europe.
8:16 Speaker 03 This is one of our favorite slides. It's the hour. I think every year we talk about this one.
8:20 Shayle Kann? Yeah, because it's always bananas.
8:21 Speaker 03 Data, right? Every, every spring, every spring, you know, in usually in April, it's relatively cool temperatures, very sunny in California, demand is low. Demand is low, supply is high. And what you end up with is more than 750 gigawatt hours of, solar curtailment in the state of California.
8:42 Shayle Kann? Which like just, just to, I mean, this is obvious, but at the same time, we're, we're talking constantly about what an electricity supply crunch we have and how, how big a problem that's going to be with the growth of AI and all this kind of stuff. And yet we forget that we are curtailing hundreds and hundreds and hundreds of gigawatts of solar every year in California alone.
9:03 Speaker 03 And what's tricky is that we can't see the counterfactual in here. We don't know what this would be without the 10 gigawatts of storage capacity that's already been built in California.
9:13 Shayle Kann? Right.
9:13 Speaker 03 I just imagine what it would be like without that. So that's California as a case. That's the story we know.
9:19 Shayle Kann? Wait, sorry, before, before we move on from California, I want to bring in another of your slides, which is actually, this is an even more alarming statistic about California, which is that I think just as of last year, as of 2024, California has a, a negative net load for the entire spring. And then net load for, for people who are not familiar is, is basically total demand for electricity minus the intermittent generation that is produced. And so in other words, we're producing more solar and wind, mostly solar in California on in aggregate than all electricity demand for the entire spring.
10:00 Shayle Kann? Not at every moment in the spring, of course, but for the spring.
10:05 Speaker 03 Aggregated across, exactly aggregated across by to a very tiny degree. Then the, but, but nevertheless negative, the minimum net load is negative. Like basically California is generating more than it knows what to do with from its, from its intermittent resources.
10:21 Shayle Kann? Which would be great if it weren't all intermittent and generating when it's generating and as a result, getting curtailed.
10:27 Speaker 03 Right. I mean, like, like we would, it would be wonderful to think what, what, you know, what would this look like if, what would this look like? The curtailment and the net, net load look like in a world where we had an uncapped. Storage resource, the four-hour storage artifact in California is purely a function of written regulation. You know, what would it look like without that? What would the market be doing in terms of longer duration, storage or different business models to use that energy? We don't, you know, we don't know because there's no incentive to know. There's no incentive to even give it a try right now. It's hard to, it's hard to do.
11:03 Shayle Kann? Yeah, it's like a market inefficiency that's just staring you in the face. Like you gotta, you gotta fix this somehow. You could fix it with policy. You could fix it by figuring out the business model for more storage. You could fix it by figuring out a business model for intermittent load. Like there's all these different things you could do, but let's just remind everybody, there are hundreds and hundreds of gigawatt hours of completely wasted electric, completely wasted clean electricity every year, growing every year in California, some of the time.
11:31 Speaker 03 And let's think about the related priced version of that, so to speak, which is in Europe where you've got frequent occurrences of negative hourly electricity prices. So PEXA Park has this great data that has how many instances of negative prices there are in, in one hour basis across Europe. And it was 500 to 550 hours in 2022, and it was more than 9,000 hours last year, which is pretty extraordinary. I mean, that, that's basically, remember there's 8,760 hours in a year. So like, like, you know, an entire
12:13 Speaker 03 year's worth of energy across Europe is priced negative, so to speak. Obviously not all of it is being priced negative for all the different member states and all the different grids and whatnot. But this is pretty significant. And as ever, the question is like, what is going to come along and solve this? So ask a European power developer and they will tell you that hybridization, putting together plants that are built at the plant level as wind or solar plus storage is one of them. And there will be new contract structures that come along to do it. But what I would say is, is that if you ask a good developer, they'll tell you that this, this is going to go away because the signal was there to do something with it. You know, why would it, why, why would I not take advantage of
12:55 Speaker 03 all this, not only free, but I'm paid to take in electricity at some point and use that in some way, if only purely in the power market to basically do some arbitrage and make some money. But yeah, that's really, I mean, 9,000 hours of negative, in which there was a negative price, it's pretty striking.
13:16 Shayle Kann? It's crazy. Yeah, wild. All right, before we move on from battery stuff, this is just another bit of data in slide 94 that I just thought was cool and had not seen before, which is speaks to the availability of grid-connected batteries in various weather conditions. Now, like, I just want to frame this as, you know, there is a question that grid operators have been asking themselves for years as to how much reliability value, how much capacity value, for example, do you ascribe to intermittent resources like solar and wind, and then increasingly now to batteries. And so, and you can compare the reliability of this stuff up against the reliability of thermal generation, which
14:00 Shayle Kann? we've seen coal stockpiles freeze and all sorts of other issues in weird weather. So, how do batteries perform when it's hot? And how do batteries perform when it's cold?
14:08 Speaker 03 Generally, quite well. This is data from Modo, which has a high degree of visibility, again, into the Texas market. But grid-connected batteries have, like, if you look at the cluster of the dot plot of all of this, you know, in the mid to high 90s availability as a resource, whether the temperature is below 32 degrees Fahrenheit or above 105, there is a little bit of tailing off when it becomes very, very hot, but not so much that you'd be, you would call this unreliable. And like, the lowest plot point that they've got for, for the ultra-high temperatures is still above 75%. And look, above 105
14:49 Speaker 03 degrees Fahrenheit, all kinds of things start to go wonky. So, all kinds of things in the power grid are starting to, starting to come off at that point. Not just generation, but power equipment as well. Lines sag when it gets that hot. Like, all kinds of things happen. So, I think it's an, it's an important set of data points in terms of saying that, like, no, this can, this is a resource that can be relied upon within its confidence interval, so to speak, at these, you know, at these times of days and in these types of markets.
15:25 Speaker 00 If you work in energy or climate, you already know the company's Antenna Group has helped build. For three decades, they've been the communications and marketing partner behind the industry's breakout companies, category creators, and market leaders. Today, their team is working across the most consequential areas of the energy economy, from data centers and grid modernization to nuclear, geothermal, storage, critical materials, and next-generation technologies. Antenna helps ambitious companies become the names everyone in the market knows, trusts, and wants to work with. They build reputation, elevate executives, drive demand, and turn momentum into market leadership. If your company is ready to lead its market, Antenna is ready to help you get there. Visit antenna group.com.
16:09 Speaker 04 When you harness the power of millions of DERs in a VPP and put artificial intelligence behind the dispatch, the outcomes exceed conventional generation. Energy Hub's DERMs platform has been trained on years of real dispatch data from over 2.6 million DERs, so utilities can plan with confidence. That means consistent load reduction across a four-hour event window, even during the widest, most uncertain peak periods. Utilities target, shape, and verify capacity down to the meter. Unlock flexible capacity that costs 40% to 60% less to deploy than traditional infrastructure in a fraction of the time it takes you to break ground on a peaker plant. Learn more at energyhub.com.
16:50 Speaker 00 For data centers, connecting to the grid quickly means two things: building distributed energy and having a plan for demand flexibility. GridBeyond helps with both. GridBeyond's DERMs and VPP platform unlocks the potential of distributed assets and dispatches flexibility into the right market at the right time. And this is more than software. GridBeyond supports developers every step of the way, from modeling your highest ROI energy mix to orchestrating the full set of assets around a data center. The track record speaks for itself. GridBeyond manages over 1,400 sites and more than 3.7 gigawatts of load across Europe, the UK, and the US. Learn more at gridbeyond.com or click the link in the show notes.
17:35 Shayle Kann? So I want to talk about the topic that is the topic du jour and has been for a while in electricity world, which is load growth and what that looks like, putting it in context. And then is it, is it going to be real? So we're going to do slides 101, 105, and then 186, where we get to talk about deep seek, which is the topic du jour of this week. But first, let's talk about what's been happening in terms of the expectation of load growth. We have some interesting data there and in terms of what official forecasts have been and how that has trended.
18:05 Speaker 03 This chart probably took me longer than any other one.
18:08 Shayle Kann? So I picked one you worked on. That's good.
18:10 Speaker 03 Simply because, simply because NERC, the body that does these official reliability forecasts, simply from what I can tell, refuses to put them in any kind of usable format. So this is me going in and hand loading stuff from a bunch of old PDFs, putting it all into a data series, scratching my head for hours as to why it doesn't line up before realizing that they've just sort of had a mulligan in 2012 and put in no data at all. So for those of you who are looking at the chart, there's a little bit of an Easter egg there, which you'll see. But yeah, we used to have an expectation of summer and winter
18:50 Speaker 03 load growth of like one and a half to 2%, like go back 20 years. That was the paradigm that the grid was looking at. In 10 years' time, our, in the next 10 years, our compound average growth rate is going to be about one and a half to 2%, depending on the, you know, the season. And that steadily eroded down and down and down and down. So by 2020, the expectation for US load growth was it was going to be about a half a percent per year. And that has rapidly unwound back to the point where we're now above 1% CAGR, 10 year CAGR in the power sector in the US. And it doesn't sound like much in one SaaS,
19:35 Speaker 03 1% a year, except this is a very big system. And this very big system that we've done a very good job of making it hard to build into.
19:42 Shayle Kann? Yeah. I mean, what I think is interesting about it is it's useful to put all the fear and craziness today into historical context, which is
19:54 Shayle Kann? We're expecting all this load growth. It's causing all these problems. But you know, the current expectations, now you could argue the current expectations are like behind what's actually going to happen and it should be a bigger number, but the current expectations are a load growth of, you know, let's, let's bump it up a little and call one and a half percent annually for, for the next 10 years, um, which is up a lot. That's triple what it was in 2020. And that's why there's such a whiplash effect, but it's actually still well below what it was in the early 2000s. I think that's the key point in my mind, right? In the early 2000s, we were doing 2% annual load growth, right? And so we're just getting back to like the growth rate that we saw in the early 2000s. It's just the sector had grown accustomed over the past few years to being pretty stagnant,
20:39 Shayle Kann? growing at half a percent a year. And, and so that's the adjustment we're making. It's not like historic load growth. It's load growth that we did historically.
20:48 Speaker 03 And in fact, and in fact, it's, it's funny because these, this chart is a chart of expectation. The expectation was for half a percent of load growth and 1% load growth, but there was zero. The actual reality was that there was no growth in total, in, in total electricity.
21:01 Shayle Kann? And it may indeed be that the expectation now is one and a half percent, but it's actually three, although maybe not for the reason that we're about to discuss.
21:09 Speaker 03 For reasons.
21:10 Shayle Kann? For reasons. And they are DeepSeek. Um, so we're, we're releasing this on Thursday. We're, we're recording it on Monday, just to peek behind the scenes. Monday is the day in which, uh, DeepSeek caused the, the stock market to crash, basically. And every company that is either Nvidia and baking chips or all the AI providers or the power companies who had been riding high on the expected wave of load growth all crashed in the market today. The reason for that is, uh, this DeepSeek AI model that appears to be maybe a much, much, much more efficient way to both train AI models and also do inference off of them. So talk to me a little bit about what we know about the efficiency of the
21:52 Shayle Kann? DeepSeek model. And then, yeah, does it change your view of?
21:58 Shayle Kann? This load growth equation.
21:59 Speaker 03 You know, it's funny, I I've spent my time at this, looking at it actually for several weeks. They released this, they released the model that I reference here right before the new year. So I've actually had this slide made for weeks. I was looking at it before it was cool. Put it that way.
22:14 Shayle Kann? Congrats.
22:14 Speaker 03 Uh, but but I was looking at it mostly from a sort of an energy and energy perspective, which is to say, from a training hours perspective, how much time on the GPU did you need to do this? And DeepSeek is an open source, small model. It was funded with $5.5 million from a Chinese hedge fund. So it's not funded with billions of dollars of US venture capital. Uh, it is operating in, in multiple dimensions, a constrained environment, not a lot of money, not a lot of access to GPUs. Uh, but it is both a very large parameter model and it is very, very efficient.
22:56 Speaker 03 Uh, and the sort of benchmark to compare it to would be the Llama 3.1 model from, from, from Meta, which had 405 billion parameters and just under 31 million training hours. And DeepSeek is getting similar results to V3 with 671 billion parameters, but fewer than 2.8 million training hours. So just read that as a lens on efficiency, like, like, you know, a train, a training hour is GPU use, which is electricity. So that's it. And if you can do all of this with less, you know, what does that mean for building gig, gigascale data centers? Basically.
23:31 Shayle Kann? Right, so it's, it's roughly 10x more efficient in terms of model training. And from my understanding, it's about 30x more efficient in terms of inference. So of course, the question here, right? This is a fundamental question. Let's just assume for the moment that DeepSeek and the, the process that they use to develop this model becomes the standard and the efficiency that they attained becomes the standard as well. There's two possibilities that you can imagine come out of that at the high level from the energy perspective. One is, oh, wait, turns out we didn't need all of those gigawatts that we thought we needed. We need a 10th or a 30th of gigawatts we thought we needed. And that is exactly what the stock market thinks as of this moment that we are recording.
24:17 Shayle Kann? The other possibility, of course, is the like Jevon's paradox version of this, which is, oh, wait, now it's like really, really cheap to both train these models and operate them. And so now we're going to do 10x, 30x, 100x more of it. And it doesn't actually affect the energy equation at all, or if anything, it causes even more load growth, just because of the accessibility of both the training and the inference here.
24:41 Speaker 03 I never imagined that the CEO of Microsoft would reference the Jevon's paradox, but he did.
24:48 Shayle Kann? Right.
24:48 Speaker 03 Nadella mentioned it specifically.
24:50 Shayle Kann? He's got a certain interest in it.
24:51 Speaker 03 He has, as he said, in reference to the Stargate, let's call it plan for building all these data centers in the US, quote, "I'm good for my 80 billion." Like, he's good, he's good to devote the CapEx to the tune of $80 billion to build all of this sort of stuff. And I think, I think the answer will be, will land somewhere in between. The challenge will be, will be who's making the money and who's operating the model, you know, on what basis is that going, is that going to accrue value in a market perspective? And what does that mean for data center construction? You know, does it mean that if everybody can go do a very lightweight, but very effective model, are they going
25:33 Speaker 03 to go back to hosting it in third-party data centers? Or is it going to be, you know, is it going to be Meta that runs everything on their own?
25:40 Shayle Kann? Well, that and the scale of the individual data center is an interesting, from an energy perspective, is like a very salient, very important question. So the, the trend line has been, okay, we need to train bigger and bigger models. These models need to be in one site. So we need the data centers themselves to get bigger and bigger and bigger. Is why we're talking about gigawatt scale and Stargate, who knows, you know, like multiple gigawatt scale, individual data centers. And that itself is becoming harder and harder to cite from an electricity perspective. Well, data centers as circa, you know, five, seven, 10 years ago, you'd see lots of these like 20-megawatt scale data centers, not tiny, but, but way, way smaller.
26:21 Shayle Kann? If indeed you get the sort of optimal performance off of an AI model, if you can build it at 20 megawatt scale, it unlocks a very different set of opportunities for where you can cite this stuff and how you can power it. So that's like an important question here. Does the result of this efficiency mean we can build 30x bigger or more powerful models on gigawatt scale? Or does it mean we're building this level of performance, but at 20 megawatt scale?
26:47 Speaker 03 100%, and when it gets, you know, when it gets down to that, it's like, okay, 20 megawatts is a not insignificant, but also not insurmountable load interconnection request.
26:59 Shayle Kann? No, totally. I mean, we have a portfolio company, not to name the company, but we have a portfolio company that needed a new facility. They're, they make hard stuff and they need a new facility. The place that they found is an industrial property, you know, in the middle of the country that just happens to have like 40 megawatts of capacity. It's kind of amazing. They, they don't need all of that capacity. They're not making a data center or anything like that. But that, that stuff still exists at that scale. Gigawatt scale, oh, wait, we just found a gigawatt of capacity does not exist anymore.
27:32 Speaker 03 It no longer exists. And in fact, I, I wonder if it ever actually did. I mean, that's, that, that's unlikely to have ever been the kind of thing that could have been easily integrated. And if it, if it did, it happened very quietly in a way that was kind of designed to not get a lot of scrutiny. It was designed to make sure that not a lot of people found out about it.
27:54 Shayle Kann? All right, so let's move on from the, the question of how much load growth to how are we going to meet this load growth? Slide 106, I'm going to rename this slide for you. The new name of this slide is build some fucking transmission. That's the name of the slide now. Can you describe what the slide is?
28:10 Speaker 03 It's a sad snide whistle sound here. In 2013, the US built 3,200 miles of 345 KV and up transmission. So high voltage, long distance transmission.
28:25 Speaker 03 3,000 miles, whatever, not a lot, but infinitely, not infinitely, but like, shall we say statistically significantly better than the 125 miles of high voltage, long distance transmission that the US built last year, which is pathetic. Like, there's really no other way to say this. And if you look at the trend of demand growth, and you look at the trend of high concentrated sites of new demand that might be far from load center or correction are becoming their own new load center, and you look at the need to decarbonize that electricity insofar as possible, and you say that the answer is
29:05 Speaker 03 to build 125 miles of 345 KV enough transmission, it's not a very good answer. Like, it's really not an answer at all.
29:15 Shayle Kann? No, I mean, to the extent that I think that the curtailment data is wild and crazy, that every year, when I see the amount of new transmission that we build in the United States, that's the craziest number to me. It's just, it's wild to me how little new transmission we've been building for a while now. I mean, it's been going down generally from what was already a pretty small number, basically since like the mid 2010s. But we're not, we haven't fixed it yet.
29:41 Speaker 03 No, and, and, and this is, you know, again, to some extent, an artifact of, well, load wasn't growing, so why would you do this? I don't know. Although, if you ask any renewable developer, they'll tell you, I know why we do this. Do you want to change the nature of what we generate? Then we need new transmission.
29:58 Shayle Kann? Yeah, maybe load wasn't growing, but like, the interconnection queues were, I could tell you that.
30:03 Speaker 03 Uh huh. Yes, they were. Right. So it's, it's, it's like, there's no, there's no non-snarky thing to say about this, I feel. Like, it's bad, and it's, it's something that needs correcting. And it is not because people are unwilling to fund.
30:20 Speaker 03 Building good transmission. It's not because there's not demand to put things into a long-distance transmission network. It's simply because it's difficult to build, or possibly impossible to build. And, you know, you might need to reform planning and permission in order to do this. But that's a policy question, not a market question.
30:41 Shayle Kann? All right, moving on. I want to talk about vehicles. And one thing that I think is underappreciated about new energy vehicles, which is the rise of plug-in hybrids, particularly in China, predominantly in China, actually. But just like, interesting how strong a position plug-in hybrids are in.
31:01 Speaker 03 It's really interesting, because I think we need to do a big reset in terms of what we consider to be a plug-in hybrid. I think people's versions in the United States is a Prius with a plug, as opposed to a vehicle with 200-plus kilometers of battery range on its own, plus a small, highly efficient internal combustion engine. You know, what in my part of the world, here in Southeast Asia, people refer to as an extended-range EV, which is kind of a weird way to say it, but it's like a vehicle that is capable of running for something like 800 or 1,000 kilometers without needing to either, without needing
31:45 Speaker 03 to fuel up the gasoline tank. And these are predominantly a China thing, in terms of where they're made. But they're being now exported all over the place. Like, I was in Thailand last month, and there are ads for, you know, the extended-range EV pickup trucks from China all over the place. You know, they have the same ads for them in Brazil. Their BYD is getting ready to ship its extended-range EV ute, i.e. pickup truck, in Australia. So like, they're starting to pick up in places where the infrastructure maybe not be fully built out, the performance advantages of EVs and the economics of them make sense.
32:27 Speaker 03 You do need that, it's not just a range anxiety thing, but it's a sort of like, a range reality kind of thing, that you might be a long way between electrical infrastructure. And so, yeah, they're like, they're picking up, but they're not, they're not your, you know, they're not your grandfather's plug-in Prius, let's put it that way. These are very different vehicles.
32:49 Shayle Kann? Okay, well, you just gave me the perfect segue to the the other vehicle related thing, which I wanted to talk about, which is slides 121 and 124, which is here comes the Chinese exports, basically, and there go the Chinese imports for that matter. So it's it's both are happening at once, but but talk to me about Chinese automakers.
33:08 Speaker 03 I don't. So we're of the age when we remember the fear that Japan Auto and Japan Auto Inc. was going to take over the world's auto manufacturing capacity, you know, the lean production system, the the sort of Toyota method of doing things was going to be dominating.
33:28 Speaker 03 China exported 6.4 million vehicles last year, five and a half million passenger cars, 2 million of those were electric vehicles. This is far above any other country's export capability. You know, it's more than Japan exports. It's more than than Germany exports, way more than the US. We're not a big exporter of vehicles. And it's just like an extraordinary rise from a couple of years ago, like basically about a million vehicles a year exported for a very long time, and then boom, exporting 6.4 million in the course of just a couple of years, like four years of run up. The Chinese auto, there hasn't been this sort
34:12 Speaker 03 of machine for building machines like this in terms of pure scale ever. China in 2023 was just a little bit under 40% of the globe's automaking production, not capacity, but actual production. And what does that mean? It means like, you know, first you see that in the in the domestic market where the standalone, or even some of the JV foreign companies, like, you know, the Japanese companies or the German companies are having a hard time moving product. Then you see it, you know, sort of propagating outwards in terms of those companies' market shares in other markets where they're going. But yeah, I mean, like, a good example is in the GM in China used to sell, you know,
34:54 Speaker 03 close to four million vehicles a year. Almost all of them were made locally, for the record. So they were not exports, GM exports to China, but they were GM companies and brands of GM being sold there. And that was 2017. And now it's, you know, like, less than a million in 2024 through, you know, almost through the end of the year. So like, it's really like, a very, very, very rapid diminution of all of these other companies. And it's this incredible amount of capability that can go somewhere, you know, and the question is where is that going to go and how?
35:29 Shayle Kann? Yeah, clearly, I mean, you know, tariffs and ban, import bans aside, you know, the absence of that, the wave of Chinese vehicles that is going to wash over basically every other country in the world seems imminent and rapid. Like, no question about it otherwise.
35:49 Speaker 03 I'm living in it. So the Land Transport Authority in Singapore, every month publishes details on the cars that got registered. Admittedly, it's not a huge auto market. It's like 40,000 vehicles a year, new vehicles. And in December of 2021, BYD had 0.2% market share, and of December 2024, it had 14.4% market share, making it the second biggest auto seller in Singapore, second only Toyota. And mind you, Toyota's business has got a hammerlock on all taxicabs, pretty much. So, like, that's pretty fast, to go from nothing to more than
36:28 Speaker 03 14%, and to go from de minimis to second in the league table in terms of sales in a market, in a place with not great electrical charging infrastructure, but great product at a good price, you know, new models constantly being refreshed. And there's others. I see routinely in my garage here, brands that you never see, and that people in the US have probably never even really heard of. Zeeker and Xpeng, you know, we're getting we're getting new cars from new companies all the time here. And they're all Chinese and they're all EVs.
37:02 Shayle Kann? Here they come. Okay, we already talked a bunch about load growth and AI and stuff like that, but you have a bunch of interesting data center related slides that I think have data I like. So we're just going to talk through them real quick. Just slides 166 to 173. So we're getting there. All right. First of all, I'm just gonna, the first one is just interesting. We spent more on data centers in terms of CapEx last year than hospitals. I didn't actually realize how much we spent on hospitals, to be honest with you. We spent 31 billion on data centers, 27 billion on hospitals. It's a lot of money to spend on hospitals.
37:34 Speaker 03 It is a lot of money. And remember, that's structures. That's not like mission critical
37:41 Speaker 03 computational stuff. That, you know, that's basically the buildings, the buildings in some aspects of the infrastructure that go with it. Basically, construction work, not GPUs. Let's put it that way. But yeah, I mean, we should be investing a lot in hospitals. It's a, you know, like healthcare is a major part of the economy, and it's something for which we have quite a lot of need in the United States. But yeah, the data center figures, which only really begin in like 2014, are like taken off, basically. And it's, it's one of these trajectories, too, that moves from being a kind of jagged low line to being a very smooth high line, like in terms, in terms of where this trend is going. Like it's smoothing itself out. It's something that really looks like it's got, you know,
38:23 Speaker 03 the beginnings of like a real exponential growth to it. We'll see if that actually prevails or not. But like, yeah, that's, that's just construction on buildings. That's not all of the other aspects of CapEx, which really are like the, the real significant expenditure and investment.
38:39 Shayle Kann? Right. All right. So anyway, we spent a lot of money on hospitals. That's interesting. We're obviously now spending more on data centers and that number is going to continue to go up. The other thing we talked about actually for a minute already, which is the portion of data center servers that are in either large co-located data centers or hyperscale data centers as opposed to small on-prem type of stuff. Now, I think like intuitively, everybody knows that number's been increasing, but it's interesting to see how much has increased and what the timeframe has been.
39:06 Speaker 03 It is, this is exactly my take on this. Intuitively, we all know that your average tech firm is not running bare metal in their basement the way that they did 25 years ago. But it's interesting to see where it goes because that has major implications for the infrastructure that gets built. Like, meeting the demand for Google's very first server, which was in a building in Palo Alto, like in a cage, basically, is very different from meeting the demand for its latest hyperscale data center in terms of what you need to build. But basically, a lot of the server load is now going into these locations that are built really big, and they're built with all kinds of efficiencies within them that
39:49 Speaker 03 are really important to note, that are different than being on-premise. And we'll see, does that tap out? Does that reverse? You know, it would be fascinating in the context of the DeepSeek conversation we were having, to see if that doesn't pull backward at some point. Like, does that actually, do people decide, actually, no, I can go on-prem now? You know, it turns out that this is something I can do in a chilled closet, as opposed to an industrial park on the edge of town.
40:15 Shayle Kann? All right, speaking of industrial parks, the other interesting data you have in here is state-level data on data center energy consumption in the United States. Now, I'm sure most of our listeners know that, you know, if you ask people, like, what state has been most affected by data center energy, or just data centers in general, I think most people probably know it's Northern Virginia. I wasn't aware that over a quarter of all electricity consumption in the state of Virginia already, as of today, is coming from data centers. That's a big number.
40:45 Speaker 03 As of 2023, like, we don't even have the 2024 data yet to see what this looks like. So, it is really, really significant. You know, and it's a clustering effect that makes a lot of sense in terms of logistics, in particular in terms of fiber access to what's there, and generally speaking, in a place where it's like we don't have a lot of industrial load, we are able to build, let's go ahead and serve this. But yeah, Virginia's, you know, 30, just under 34 terawatt hours of power consumed by data centers in 2023, and almost 26% of state power already going to that. And, you know, even more important, the designs that companies have in terms of what they want to do and how they want to expand.
41:29 Speaker 03 And it could blow the doors off of these numbers, right? Like, it could absolutely blow the doors off of these numbers.
41:36 Shayle Kann? Speaking of blowing the doors off of those numbers, you have another slide that I really like just because, because what the hell is going on in, in Rappahannock Electric Cooperative territory? Want to talk about that a little bit?
41:46 Speaker 03 So Virginia, Virginia's really fascinating for this because you've got these great, you've got a great big utility dominion, but then you have a lot of co-ops that actually are service providers to all these different communities. And, you know, co-ops are like, look, we handle rural areas and ex-urban areas. We have, you know, our main load might be like cooling, you know, in different times of year. We're not really built around big industrial consumption. We are not a giant financialized entity either. You know, we, obviously we have real finances, but we're not Dominion. And Rappahannock had this filing in Virginia where they basically are saying, look, like,
42:27 Speaker 03 we are now getting requests for interconnection that in a single location are bigger than our entire peak load right now. We have a peak load of 1.2 gigawatts. We have people asking us to interconnect one asset that is bigger than that. In some cases, four times bigger than that. I have to say that if somebody's trying to connect a five gigawatt data center in Virginia that may have low likelihood of actually going through, but it's indicative of the kind of trends that people are wrestling with. And what's really interesting about this and something I've spent some time looking at is what does this do to what Severn Bornstein at Berkeley says is the duty to serve? What does this do in the market construct of saying, sure, we'll provide power to you?
43:10 Speaker 03 If one company shows up and can basically blow up the demand profile permanently by just introducing two to three gigawatts of load in a place that has one gigawatt peak load? Related to that, though, and equally important, and it's nice to see that this is entering into the regulatory discussion, what happens if they don't show up? What happens if I'm Northern Virginia Electric Cooperative or Rappahannock Electric Cooperative, and I go out there and build on the expectation of getting five gigawatts of data centers? And it turns out that these things got deep sunk and don't get built at all? Like, what does that, what does that do to my ratepayers? And we've seen similar things happen with a nuclear buildout in places like South Carolina
43:55 Speaker 03 with the VC Summer Plant that was basically a $7 billion worth of poured concrete and nothing more. What would be the equivalent happening there? And on what market structure do you need to account for and accommodate this kind of building? Again, all back to policy and regulation questions. Like, this is not a finance question. It's not a technology question. It's a decision question about what, what our concept of the, of duty to serve is going to be.
44:20 Shayle Kann? Deep sunk. You just come up with that on the, on the spot? It's pretty good.
44:23 Speaker 03 I did. You heard it here first, Shale.
44:27 Shayle Kann? You know, I like it. Okay, um, we're going to wrap up with one more thing on AI and energy, which is data that I really like because it drives home an important point. An important point that actually we've, I've discussed with Brian Janis on this podcast before. He talks about this concept of the bit watt spread, and it's basically the idea that actually the bits that you're, that you're using, producing is worth so much more than the watts you pay to produce the bits. And that's where there's this like sort of limitless willingness to pay to some extent within the, the data center world. But you have data that is really interesting in here on the share of total cost associated with training an AI model that can be ascribed to the energy.
45:10 Shayle Kann? And it's, you know, I think people think of it as being such an energy hog thing to do, that it would be a big share of the overall cost, but it's not.
45:20 Speaker 03 So to be very clear, this isn't my data. I got this from Andy Lubrashane, your colleague, our mutual friend. He in turn got it from Epoch AI. But they have these great charts that's the breakdown of model costs for training and experimentation in terms of what goes into it. And, you know, like R and D staff, so and also in this case, they include, as they should, the cost of equity is between like 30 and almost 50% of running of training a model and of doing experiments with it.
45:52 Shayle Kann? AI researchers are expensive.
45:53 Speaker 03 AI researchers are expensive and they demand compensation.
45:58 Speaker 03 The energy component of that comes way after the AI accelerator chips, the other server components, the interconnection costs, and it's somewhere between 1.7 and 6.3% of the cost of training and experimentation. Now, it's possibly a bigger and it probably should be a bigger part of inference of actually operating the things that are up and running, and that's an important caveat. But, you know, what Andy says is that energy is everything and energy is nothing. Like, it's everything in the sense that it's binary. I either have it or I don't in order to run these processes and capabilities that I need. But once I got it, I am not breaking the bank
46:42 Speaker 03 paying for electricity. And this is unusual in the world of very large consumers. Like, if you are in the business of, say, making aluminum or making steel, you are acutely sensitive to the cost of your power input, like very, very sensitive to it. But you're not like, like, it's not breaking the bank of a model at all if you had double the cost of electricity. The more important thing is that can you get it right now? And can you get it how you want it? You know, am I able to get it reliably? Am I able to get it with the uptime that I'm expecting? Can I get it quickly? Like, is this something I can contract for now and have in six months?
47:23 Speaker 03 And that's, that's, it's, it's very different. It's a very different kind of buyer than I think we're used to seeing and that than grids are used to seeing elsewhere. You know, a big box retailer is very sensitive to the cost of electricity. At some point, that carries through to the very, very low margin business of selling goods that it does. But we're not in the scenario where people are jacking up the cost of models that in many cases are still being given away for free because of the electricity costs underneath it. It's a different paradigm.
47:54 Shayle Kann? That it is. And with that, Nat, we've reached the end of this year's 200 Slides of Glory. Thank you so much, as always, for letting me cherry pick.
48:05 Speaker 03 Shale, always a pleasure. Thanks again, and I look forward to hearing them live.
48:11 Shayle Kann? Nat Bullard is a co-founder at Halcyon and the former Chief Content Officer at Bloomberg New Energy Finance. This show is a production of Latitude Media. You can head over to latitudemedia.com for links to today's topics. Latitude is supported by Prelude Ventures. Prelude backs visionary is accelerating climate innovation that will reshape the global economy for the betterment of people and planet. Learn more at preludeventures.com. This episode was produced by Daniel Waldorf. Mixing by Roy Campanella and Sean Marquand. Theme song by Sean Marquand. Stephen Lacey is our executive editor.
Previous Are utilities ready to fully harness demand flexibility? [partner content] Next The case for colocating data centers and generation Page updated November 4, 2026 · transcript and summary are machine-generated