With electricity now the limiting factor in the race to build superintelligence, the tech industry’s response has been very Silicon Valley: move fast, break things, and relentlessly scale.
The result? An overtaxed grid, a wave of community pushback, and an obsession with jet engines, ship turbines, and small modular reactors that don’t solve today’s problems.
In this live episode, recorded at the Transition-AI conference in San Francisco, we stress test three Silicon Valley mantras against the reality of what’s happening to the grid.
First, “move fast and break things”: Data center bans are spreading, communities are organizing, and the backlash is bipartisan. Is it too late to rebuild trust?
Then, “first principles” thinking: Data centers are going off-grid, jet engines are being bolted to trailers, and the grid is being treated as a hurdle to avoid. Why is this approach shortsighted?
Finally, “10x better, not 10% better”: Every hyperscaler has a moonshot, but very few have planning cycles that extend past three years. So are we missing a critical window to get creative?
We’ll close with some audience ideas for creative solutions to the AI energy bottleneck.
Credits: Co-hosted by Stephen Lacey, Jigar Shah, and Caroline Golin. Produced and edited by Stephen Lacey, Sean Marquand, and Anne Bailey.
Want to watch this episode? Subscribe to our YouTube channel.
Open Circuit is brought to you by FlexGen, a leader in integrated battery energy storage solutions and energy management software. FlexGen helps owners and operators gain greater visibility and control across complex energy systems to maximize performance. Learn more at www.flexgen.com.
Transcript
Stephen Lacey: From Latitude Media, this is Open Circuit.
Stephen Lacey: Silicon Valley has a rich history in clean energy innovation. Many of the top players in the industry were born here and plenty have died here as well. But the Valley’s relationship with energy is getting a lot more complicated and a lot more consequential. Electricity has become the limiting factor in the race to build super intelligence and the tech industry is running a familiar playbook. Move fast, break a lot of things, and focus on scale above all else. So today we’re going to lean into those Silicon Valley instincts and stress test them against the reality of the grid. And we’ll close with the pitch session. We’ll give the audience a shot at solving the AI energy bottleneck that’s coming up live from San Francisco at Transition-AI. Hello, San Francisco. Listen to all those powered land developers. They’re multiplying it.
Stephen Lacey: I am here live at the Transition-AI Conference with my two trustee co-hosts, Caroline Golan and Jigar Shah. Caroline is the chief growth and policy officer at NRG. You may know her as the former global head of energy at Google, and she is someone who more than anyone I know really understands the intersection of electrons and bits. Caroline, good to see you.
Caroline Golin: Oh, that was very kind, Stephen. I know bits. I know bits of little things, I guess, is what you’re saying.
Stephen Lacey: So you’ve been at NRG for a little bit. You started co-hosting this podcast before you moved to NRG. Has Jigar gotten you in trouble with the comms team yet?
Caroline Golin: No, you haven’t.
Jigar Shah: I need to work harder.
Caroline Golin: It’s true. No, he’s been great. I mean, maybe with my family and husband for texting for far too long or something like that, but no.
Jigar Shah: But that’s not something I said on the Energy Gang, or I mean on the Open Circuit podcast.
Caroline Golin: No, it is something you and I just need to work on in our personal relationship.
Stephen Lacey: Jigar Shah is an investor, a ubiquitous energy podcaster and former director of the DOE’s Loan Programs Office. How are you, sir?
Jigar Shah: I am well rested and ready to go.
Stephen Lacey: And you are wearing a suit. This is a first.
Caroline Golin: Okay.
Stephen Lacey: It’s a historic moment.
Caroline Golin: Can I take credit for you wearing a suit because it is-
Jigar Shah: Yeah, I got so much crap after the Houston Live podcast and none of it from Caroline, to be clear, from people who Caroline talked to, who basically said to me, “Caroline really didn’t like your outfit on stage.”
Stephen Lacey: She has really helped us step up our game in many areas, including fashion.
Caroline Golin: Beauty is in the eye of all beholders.
Stephen Lacey: I’m kind of curious. We have been talking for a long time now through the course of the year, and we’ve been talking a lot about AI infrastructure, which is what we’re going to talk about today. But I’m curious if you guys have seen AI bleed into your lives in any unique or interesting ways over the last couple of years. Are you users? Are you seeing people around you use it in interesting ways?
Caroline Golin: I am a dabbler.
Stephen Lacey: Personally.
Caroline Golin: Yeah. I mean-
Stephen Lacey: What do you find it helpful with?
Caroline Golin: I planned an entire trip to Scotland-
Stephen Lacey: Nice.
Caroline Golin: With AI. TBD, if it’s going to be good, I’ll come back after Scotland and tell you this was wonderful or rubbish.
Stephen Lacey: Do you use an agent for that, an AI agent, or do you just-
Caroline Golin: I just did iterations and said, “Okay, I want a Scotland tour that’s focused on folklore and dragons for my 13-year-old son.” And I got-
Stephen Lacey: Amazing.
Caroline Golin: Yeah, it’s going to be amazing. Or it’s going to be terrible and I’ll come back and tell you afterwards.
Stephen Lacey: Oh, so it hasn’t happened.
Caroline Golin: Hasn’t happened yet. But I’m hoping. I’m hoping. But no, other than that, I’m not using it a ton. I mean, obviously we are at NRG and integrating it, but in my personal life, I’m still learning how to knit, so I’m on that side.
Stephen Lacey: What about you, Jigar?
Jigar Shah: I think you guys all know that I’m not interested. I think in my entire life, I have not given a single dollar to an AI company. I think I’ve used the free version of Claude and ChatGPT. I did use it for my Florida spring break trip. It was amazing. But I went to this dad’s whiskey night on Saturday night, and one of the dads said, “Ugh, I was just inspired. My son really likes license plates, so I used Claude to create this whole license plate game where he could see the license plate, stick them in the thing, whatever.” I went to bed at three in the morning. I was like, “How many tokens did that cost you?” He’s like, $240. I was like, “That’s ridiculous.” I’m sure the game was terrible. I asked him if it was multiplayer. He’s like, “No, Claude explained to me that you couldn’t make a multiplayer game without having to spend $200 a month for the server access and whatever.” And I was like, “This is why I don’t do AI.”
Stephen Lacey: I’m a pretty conventional user. The coolest area that I’ve used AI is in cooking actually. And I will often say, “I’ve got these ingredients. Can you help me come up with an idea for a recipe?” And it’s remarkably effective and it’s good at cutting back food waste. In fact, I think Google was piloting something around this.
Caroline Golin: I’m sure they’re piloting lots of things.
Stephen Lacey: But it’s actually pretty cool. I recommend trying it. I have all these random ingredients in my pantry and I’ve been able to cook up some really great stuff, some not so great stuff, and it’s a little hit or miss sometimes, but really effective.
Stephen Lacey: Okay. So to start off, I think I want to set up a very particular way of thinking about this moment. We all know the tech bros, the crypto bros, the nuclear bros, the finance bros. Well, over the last year or two, year or two-
Caroline Golin: All male. All male. Everyone in the industry is male.
Stephen Lacey: Over the last year or two, we’ve got a new class that I call the electro bros. This is a person often from the tech industry who is suddenly discovering electricity and they think they have an answer to fixing power part.
Jigar Shah: Like Ben Franklin?
Stephen Lacey: They have the answers for redesigning the grid, unlocking energy abundance. They’re a fixture on X and YouTube, and they’ve never participated in a utility rate case or built a project, but they are very, very, very confident. Jigar, do you know this character?
Jigar Shah: I do. I do. They have five million podcast views.
Stephen Lacey: Yes. You seem to interact with a lot of them on X.
Jigar Shah: Well, interact is a strong word. I call them dudes. I call them dumb, and then they respond, and then I stop responding. I don’t keep going. My whole thing is basically that if you want to light money on fire, it’s obviously your choice, but when you’re ready for a sober solution, we’re here for you.
Stephen Lacey: Caroline, what is it like seeing people like this discover energy for the first time with such confidence?
Caroline Golin: When I was at Google, I had a separate inbox in my email for people who were just emailing me exactly what I should do with my job from within the company. It’s like a 300,000 person company. And when you’re in charge of strategy, there are a lot of people who think that they have figured it out. And I mean, that inbox would fill up 100, 150 emails a day, just from within Google for people who thought they had cracked the nut that clearly the rest of us couldn’t crack with their one idea.
Stephen Lacey: There’s an AI tool for that.
Caroline Golin: There’s an AI tool for all of it. But when I was in grad school, I did my first big presentation at John Hopkins and I remember coming off the stage and I thought I was like, an we cuss, like the shit? I was so cool because everyone was so impressed with my research and my advisor pulled me aside and he was like, “Okay, you’ve discovered your hand. There is a whole universe. Sit down.” And I just think that that’s part of what’s going on right now, which is that everyone wants to think that they understand the universe that is this industry because they’ve discovered their hand. And as a parent, you see that every day. My 15-year-old thinks he is smarter than everyone.
Stephen Lacey: I’m sure everyone with a teenager can relate. Well, I bring this up because I think it’s a good opportunity to think about this character as we interrogate some of the dominant storylines in energy and AI right now. And we’re going to do that through some of the tech industry cliches. We’re here in Silicon Valley. And I think the first is the most well-known trope, which is move fast and break things.
Stephen Lacey: Uber threatened taxis, Amazon threatened retail, social media threatened democracy, and AI, depending on who you ask, is coming for everything else. And so while AI labs race to build what they’re calling Machine God, one that could upend the workforce and reshape societies in ways that we can’t predict, they are colliding with an increasingly skeptical public. Data centers now under some growing pressure, not just in the markets, but also on
News clip: Capitol Hill. It’s not just happening in Washington. In Arizona last week, city officials unanimously rejected a proposed AI data center in New Jersey and Virginia. Rising electricity prices played rules in statewide races.
News clip: Two emotions are at the forefront of this protest. Fear and worry.
News clip: No secret deals. No. Secret deal.
Speaker 6: So far, many of the companies behind these proposals, including Microsoft, haven’t said much.
Stephen Lacey: And this is getting really scary now. I mean, Sam Altman had his house attacked. There was an Indianapolis councilor who supports a data center, had his house shot up recently. I mean, it’s pretty dark and intense out there. And of course, this is having a real direct impact on infrastructure too. Dozens of data centers have been derailed in the last couple of years. Equipment constraints have played a big role, but local opposition, according to data center watch, has amounted to $64 billion in projects being canceled. And this opposition is remarkably bipartisan, very evenly split among Democrats and Republicans. So this move fast and break things mentality, is it breaking trust beyond repair? Caroline, what do you make of the backlash? Has the scale surprised you?
Caroline Golin: No, not at all. I mean, we saw this when I was at Google, I saw this years ago and knew that we were in for a very hard go. And in large part, it’s because the entities that have made up the data center boom and infrastructure are not entities that ever had to think through massive political or community engagement work. They’re real estate developers who were used to assembling some parcels, selling it off and going on their way, and now you just 10X the size of those parcels and you 10X the size of those impact, but they’re using sort of the same mechanics in terms of the way they go to market. And at Google, I’ve felt like we did a really good job actually of going above and beyond. I think we did a really bad job of telling our story and working to be a local company.
Caroline Golin: I think the thing that’s important for everyone to understand about the tech companies is none of them ever wanted to be local companies. None of them ever wanted to be relevant in DC. None of them ever wanted to actually be engaged in policy writ large. For the most part, the tech industry is like, “Call us if policy matters.” It’s not been something that they have a muscle around because they were a brand new industry that hadn’t been regulated. It’s just now that we’re starting to see regulations around data and data use and privacy and stuff, but for a long time, they would just stay out of that. And that sort of trickled down and I think also impacted the entire space. And at the end of the day, even if you’re a Google or a Microsoft or an Oracle, because I have to remember to say that every time, right Jigar, there’s a-
Jigar Shah: They really want to be a player.
Caroline Golin: They’re a player. They still haven’t emailed me though. I mean-
Stephen Lacey: We have a running joke that Caroline always left out Oracle.
Caroline Golin: I always left out Oracle, but they never … We could be as good as you could humanly possibly be in a community and the least common denominator is still the worst actor, right? And so you can only control for so much of it. And I also think that the worst actors really are working on a three-year, five-year capital cycle. So if they get their parcel, sell it off and go on, why do they care? But yeah, this has been coming for a long time.
Stephen Lacey: Yeah. Jigar, you have been characteristically vocal about this. What do you think the hyperscalers, the data center industry generally has gotten wrong that has led to this extreme situation?
Jigar Shah: No secret deals. No, I think … So I was downstairs after I came off the plane and had drinks with our friend, Brian Janous, and several other folks that were down there. And I mean, as I learned from all of these bros who have podcasts, that were short chips, apparently. So on the one hand, you’ve got Shanu Matthew, who’s like, “Oh, 40% of data centers that were announced for 2026 haven’t started construction yet.” And then on the other hand, they’re like, “Even if you wanted chips, we don’t have enough chips for you.” And NVIDIA is now deciding which data centers get chips on a data center by data center basis. So they’re no longer selling chips to Google or Microsoft or whatever. They’re selling it to that data center because they believe that that data center is actually going to get built. And so they don’t want you to hoard chips anymore.
Jigar Shah: They want to make sure that those chips get a thing. So on one side, we clearly don’t have enough chips, but on the other side, we’re disrupting communities in all 50 states with fake data centers that don’t have chips. And so I’m like, “Okay, so on this side, you’re lying to me, and on this side, they’re just telling me that they don’t have enough chips for the data centers that they need to put in the ground.” So part of the problem, I think, in this no secret deals environment is that people don’t actually know the truth. And when you have a trillion dollar industry, which has been single-handedly bouying the stock market in this country, you would think that some analyst would give you truth, would give you information. How many chips worth of data centers do we have? How many chips is NVIDIA like shipping in 2026 and what does that translate into megawatts?
Jigar Shah: Oh, 16? Great. So we have 16 gigawatts worth of data centers that we can build this year. Then why is anyone telling me we’re going to build 32 gigawatts? Clearly they’re too stupid to have a podcast. I just think that these are basic things that a trillion dollar industry should be able to reveal to me if they’re going to buoy the entire US economy, which is the largest economy the world has ever seen.
Stephen Lacey: Okay. I mean, that makes sense, but I don’t think that gets to the reaction from communities.
Jigar Shah: Are you kidding me? In Ashburn, Virginia, there’s a data center developer who just offered $4 million a house to 173 houses. They all get $100,000 right now this week. And if they all agree to sell, then they get the full four million. But if anyone holds out, they don’t get the four million. When you have that-
Caroline Golin: Tragedy of the commons at its finest.
Jigar Shah: Right. But when you have that kind of crap that happens, then what the average American thinks is that these data center companies believe they’re God and they don’t have to come into your community and explain themselves. They don’t have to talk to you about water. They don’t have to answer your questions. They just throw their big fat wallets on the table and say, “Shut the hell up and just sell me your house.” And that is offensive to everyone. And I understand why all of them are like, just burn it down. I am tired of all of it. My rates have gone up, my electricity rates. I’ll give you one more example. Charles Hua, who was here today, still here, came out with his Powerlines report yesterday for all of the crap that the hyperscalers are saying that they’re doing, being nice. Google Tapestry made the presentation unlike the PJM interconnection queue, all that stuff.
Jigar Shah: The utilities came out and said, “You know what? We said that we were going to invest a trillion dollars from 2025 to 2029. Hold my beer. It’s going to be $1.4 trillion from 2026 to 2030.” So what are the hyperscalers doing to make my bills go down? Nothing.
Stephen Lacey: Well, I mean, I think there’s a lot of misinformation because historically data centers have not played a role in rising electricity prices.
Jigar Shah: You tell that to the communities that are shutting down data centers. Because every single utility-
Stephen Lacey: There are lies about it.
Jigar Shah: No, no, no. They’re not lies. Every single utility CEO that did an earnings call absolutely said that we upped our CapEx budget because of load growth primarily due to data centers. Every single one of the transcripts says that. So on this side, they’re saying that they’re raising rates because of data centers. And on this side, you’ve got Lawrence Berkeley saying, “No, that’s never happened.”
Caroline Golin: Well, historically, rates weren’t being raised because of data centers, but if it’s done irresponsibly, rates will go up. I mean, the reality is though the regulatory tools exist, the bilateral tools exist to ring fence and sleeve cost of new build generation. The problem is when you take existing generation on the grid that’s already been cost allocated and divided up between different rate classes and try to increase or juice the cost of that in the market. But if it’s an existing power plant, that’s a problem. If it’s a new power plant, every single utility in this country can ring fence that power plant cost and directly sleeve it. It’s just math. So I think the entities that need to get yelled at are broader, perhaps.
Stephen Lacey: Well, I mean, there’s a question about what needs to be done. We have heard from a number of folks in the data center community here who have talked about the need to be in communities early, often, figure out what their needs are. So there are plenty of people really taking this very, very seriously. When you talk about … I mean, are there specific … I don’t know if you want to name names, but do you see that there’s a lot of bad actors out there, Caroline? Because when I talk to a lot of the folks in the industry, a lot of them realize it’s a big problem and they need to do something about it. So how much is the industry not caring and how much of the industry is really taking it seriously?
Caroline Golin: I think there are bad actors. I think that are also just lazy actors. I mean, there’s a difference line between bad and lazy.
Stephen Lacey: Like gas engines on tractor trailer trucks.
Caroline Golin: Yeah. I mean, I don’t think anyone is like, “Oh, let’s sit in our conference room on the whiteboard. This is how we’re going to solve this problem by lawnmowers.” No, I mean, it’s like a worst case scenario that they get squeezed into and someone comes and says, you know, like The Wizard of Oz, I have a solution for you. And they take it. I mean, I think I said this on the last podcast, they’re just doing what’s right in front of them.
Caroline Golin: I think that that happens when they don’t have a competent development partner. And I think a lot of the entities, powered land, what did you call them? Land bros, land bros?
Stephen Lacey: I didn’t have land bros but yeah we can call them that.
Caroline Golin: Land dude bro bros. Those folks, they aren’t competent on things like due diligence and understanding power dynamics and transformers and lines and interconnection and all that stuff. So they’re doing the bare minimum hoping that they make 10, 15% return on their parcel, selling it up to someone who does know how to do that. Those people do not care what happens, and the market is set up in a way that they don’t have to. And frankly, historically, the way we built C&I Load, it favored apathy. I mean, we created a regulatory system across the country that said C&I load is a form of economic development, therefore they shall not be touched. It’s why they don’t have energy efficiency standards like everyone else’s, why they don’t have demand response like everyone else does because they were considered economic drivers and keeping their rates low was good for the economy.
Caroline Golin: All the data center community went into that tunnel and now we’ve decided maybe they’re not great for economic development. And that’s where the rub is, is they just followed along in the same footsteps that smelters and steel plants and mining companies have created for a very, very long time, but the social contract isn’t the same. And that’s where you see the breakdown. The breakdown is in people do not trust that the AI boom is going to bring the jobs and the economic development that traditional C&I brought, and therefore they shouldn’t get the same social contract.
Jigar Shah: No secret deals.
Caroline Golin: Or no secret deals. Yeah. Sure.
Stephen Lacey: Let’s move on to the next way of thinking, which is first principles thinking. If you spend enough time on X, you might think we already have all the answers. We just need to get to first principles thinking. It’s the authentic Musk, Silicon Valley framing, strip away the assumptions, ignore how things have always been done, reason from the ground up and you’ll find a solution that incumbents were too captured to see. Here’s what that sounds like in energy.
News clip: There’s so much more we can do. There’s medium speed reciprocating engines, engines that spin in circles. So sort of like any diesel engine, right? There’s like 10 people who make engines that way. Well, actually automobiles, manufacturing’s going down. These companies all have capacity and could scale and convert that to for data center power. Oh, what about ship engines? All of these engines for these massive cargo ships, those are great. Any of these individually will do tens of gigawatts and in a whole, they will do hundreds of gigawatts. So we can add a lot more energy and the supply chains are just way more simple than chips.
Stephen Lacey: So there’s two big arguments that come out of this way of thinking. The first, which you heard from Dylan Patel, who’s the CEO of the research firm SemiAnalysis, posits that we have all the capacity we need from the aviation and automotive industries, and we just need to repurpose it. And then the second, which I want to address next, comes out of that argument, which is that data centers should be fully off grid in the race for scale, and that there’s a compelling economic case for just deploying a bunch of ship engines at this and keeping those data centers off the grid entirely. So let’s go to the first point. We’ve talked on previous episodes about this idea of repurposing old jet engines and ship engines for power. Jigar, you’ve been pretty blunt about this. What’s your reaction to Dylan’s comments?
Jigar Shah: It would be awesome to have the Ford F350 truck just ship Hemis and run a data center off Hemi. Oh my God, that would be awesome. We should do a whole conference about that. Gosh, I mean, I agree. Those engine plants, I mean, they probably are underutilized. I think we had peak ICE sales in 2017 or 2018, and so you’re right. Gosh, I mean, I could have thought of that. I should have thought of that. No, I mean, come on, seriously.
Caroline Golin: You’d be richer if you thought of that.
Jigar Shah: I mean, just the dumbest people you would possibly meet.
Caroline Golin: So nice. You’re so nice.
Jigar Shah: But look, I know a lot about this stuff. I don’t know anything about the bit-watt spread or tokens or whatever it is, but I also don’t lecture people about bit-watt spreads and tokens all day on my podcast. I talk about energy issues that I know about, right? They clearly know nothing about power. I just think that one of the real tells of how dumb all these people are is they talk about generation as if we’re short generation. We are not short generation in this country. We never have been. We are short wires. We have always been short wires, right? So the biggest tell is when someone says, “We just need nuclear power and that’s going to solve it all.” Or we need geothermal or we need Hemi engines or whatever, then you’re like, “Oh, you really don’t know anything poor boy.”
Caroline Golin: I’m glad it’s still all men in this entire conversation. Not a single female podcaster out there acting astonish. Yes, you can clap for that. That’s what I’m talking about.
Jigar Shah: I mean, the sad thing is that they’re Indian men. We’re supposed to be smarter than that.
Stephen Lacey: Caroline, why are these technologies, these repurposed ship engines not part of your growth plan at NRG, Caroline?
Caroline Golin: Because we actually know how to build power plants. I mean, that’s the short answer. Listen, I think that these massive off-grid data centers are going to get built. I think they will get built. I would say, you want to make a wager on how many people wants- Well, I’m going to get to off-grid data centers next. Well, I mean, but that’s why you do it. Let’s make a wage. Jigar is right. The short-term issue is still wires capacity and congestion, right? That is the short-term issue. Longer term, I do think we need generation in this country, and I think we’re going to need it at different levels, both on the distribution side and large scale, the transmission side. Why is it not in NRG’s platform? Same answer. We know how to do this responsibly. And at NRG, we have residential customers. We have to go serve those customers and faithfully serve them and look them in the eye and say, “We did everything we could do to maximize our value proposition to you and keep your rates low.” And that takes precedence.
Stephen Lacey: Can we actually, before we move on to off-grid data centers, just pause on this for a second and talk about why you wouldn’t use those pieces of equipment. We had a really interesting presentation yesterday from Flexgen showing how AI workloads can cause these serious torque problems for turbines. Just operationally, why wouldn’t you deploy a bunch of this stuff for data centers?
Jigar Shah: Well, I think you start with the fact that data centers are out of compliance with large loads on the grid. You can’t connect a data center directly to the grid because if you did, they don’t meet the ramp rate requirements, they don’t meet any of this stuff. A lot of them go from 20% loading to 80% loading five times a minute. So now you need batteries in the middle to actually make sure that those loads don’t take down the grid from harmonics and other things. So now you need to put in stuff like that, Terraflow Energy, lots of other folks sort of do that, or they’re attempting to do that. And then the question becomes to Tyler Norris’s paper and other folks, how do you actually get them onto the existing wires on the grid? And you saw Ahmed and Ryan’s work with GridCare and how he found 480 megawatts that Portland General didn’t know that they had just by unleashing the data within the silos, within the utility.
Jigar Shah: And Portland General is very appreciative because they want to sell 20% more electrons because it allows them to reduce rates for all of the other rate payers. So I think you’re just in this place where we all know how to help a data center be something that actually could be embraced by the grid and valued as a load. Today, they are not. Today, some data centers are being turned on and immediately being told by the utility company to turn off because they’re destabilizing the grid. And the data centers spent like $20 billion without realizing that they were destabilizing the grid. I just think when you make that kind of own goal, you deserve to be flogged.
Stephen Lacey: Is that true? I haven’t heard that. Do you understand that to be true?
Caroline Golin: Well, AI training has insane fluctuations, which we’ve talked about at the conference. And so yes, you need to integrate us stabilizing mechanics. But ultimately the reason why this sort of Frankenstein model doesn’t work is because you actually can’t control for the harmonics long term. So you are sort of jerry rigging an amount of capacity that you want to be able to provide, but capacity provision is different than the harmonics that you need to maintain training and to maintain stability within whatever electric generation that you’re trying to pull from. So the problem with the Frankenstein approach to me is that you can probably get enough electrons, but to be able to control those, integrate those and match those to a load shape that is incredibly erratic as it is, is very, very, very difficult. So you’re going to sort of burn through things quickly and the trip failure on that is huge as you start stacking things.
Caroline Golin: That’s a lot of the technical reasons why I don’t think it works. And then what happens if you blow 10 or 20% of some of these jerry rigged motors? Do you have the internal communication to pull down 20% of your training equally? All of those things haven’t really been worked out. So I know entities are trying to work that out. The question is, is that scalable? I don’t think that’s a scalable solution. I think that, like I said, it’s a Frankenstein one and done solution and will probably end up in a lot of mechanical failure and will be incredibly difficult to then interconnect to the grid.
Stephen Lacey: Well, you brought us into off-grid data centers, so here we go. You actually think there will be a bunch of off-grid data centers built. Are those data centers with an eventual grid connection or do you think that they’re islanded data centers permanently?
Caroline Golin: I think it’s an eventual grid connection. But if you think about building a three to five gigawatt data center, which is the size of most munis or co-ops in this country or bigger. At that point, you’re just creating a special interest district. So to me, it’s a more of a political zoning question. How do you create an islanded entity that becomes its own muni, or its own cooperative with its own self-serving load and generation? And it’s a single meter. Smelters have done it before where smelters needed to be located just happened to have a backbone of the grid nearby. And I think that’s what will happen and they will build the single line tie-in as they go. But ultimately, I think it’s just as much a physics issue as it is a human issue, a political issue.
Caroline Golin: If they’re out in the middle of nowhere and no one cares about them and no one sees them, okay, you maybe get rid of that political human issue and then you have to solve the physics. And I think for most of the tech companies, like my earlier statement, their culture, their DNA is like, we can deal with technical problems. Humans kind of scare us and they’d rather deal with technical problems than human problems.
Stephen Lacey: Totally. And that’s why they want to take data centers into space.
Caroline Golin: Exactly.
Caroline Golin: To the moon, Alice, to the moon. No, but I think when you think about just these examples you’re talking about, smelting or refineries or whatever, I mean, for these companies, half of their cost of goods sold could be energy. A third of their cost of good sold could be energy. So you’re talking about a team of a hundred that work there that actually manage their energy system and make sure that it’s playing nice in the system and they’re filing regulatory filings with the regulator and all this other stuff. In these places, they don’t care about energy. They’re like, “Oh, I’ll pay 24 cents a kilowatt hour if it’s like speed to power.” And so when you think about just how much they’re under resourcing this function, we’re talking about just the initial startup phase. We’re just trying to get speed to power, right? We just want them to turn on, right?
Caroline Golin: That’s the entire conversation that we’re having here. Now they actually have to play nice with the grid for 20 years, right? When a one gigawatt load just trips off because somebody ended a training run and forgot to install some inference work, that takes down the grid.
Caroline Golin: When something fluctuates in a way that somebody forgot about, oh, there was a software glitch. So sorry, I’ll get clawed on it. I just think that in general, that requires a hundred people to actually work there to figure this out. That’s how a refinery works. That’s how a major smelter works. In this case, they’re like, “Oh, we have a software program that manages that.
Caroline Golin: I don’t think there are a hundred people at refineries working on regulatory interconnection dockets.
Jigar Shah: No, on their energy team.
Stephen Lacey: Also, the software is pretty damn smart.
Jigar Shah: Until it’s not, have you looked at what happened in Dominion and Loudoun County recently where they almost took down the whole grid, right? This is what I’m saying. Everyone is just so positive about how wonderful this is.
Caroline Golin: Stephen, there’s a lot of things you haven’t heard.
Jigar Shah: I just think that they accidentally tripped off two gigawatts of data centers in Loudoun County and Dominion was like, “Oh crap, this actually just destabilized the grid.” I feel like people are just like, “Oh, that won’t happen. Oh, whoops, it happened. But we have a software patch. It’ll get updated next week.”
Caroline Golin: Well, but right, but if they’re completely off grid, they’re not filing regulatory-
Jigar Shah: We just decided that there are very few of them that are going to be off grid.
Caroline Golin: We did. Okay. We decided that. Okay.
Stephen Lacey: I mean, this feels way overblown.
Jigar Shah: What’s overblown?
Stephen Lacey: That you’re afraid of these data centers collapsing the grid.
Jigar Shah: Oh, it’s not overblown. Talk to Tom Fanning who now has an entire operation in DC that does nothing but try to figure out how the data centers companies are probably going to screw up the grid. I just think that when you think about just how callous all of these people are, and yes, they have people like Caroline, but Caroline left for a reason, right? At some point-
Caroline Golin: I left because I had such a great desire to sit on stage and bicker back and forth with you on a podcast.
Jigar Shah: I’m just saying what 57 gigawatts of off-grid data centers mean is that there’s too much Mark Zuckerberg masculine energy flowing through the top of those daysers.
Stephen Lacey: I forgot to mention how effective Jigar is at deeply offending the people he’s in front of when we do these live shows.
Jigar Shah: These guys all agree. How many people agree with me that when someone proposes an off-grid data center, it’s because they probably weren’t smart enough to figure out how to build it on grid. Right?
Caroline Golin: Wait, but how many people have built a five-gigawatt data center on grid?
Jigar Shah: There isn’t a five-gigawatt data center in operation right now.
Caroline Golin: Exactly.
Jigar Shah: And even the ones that we’re talking about that are five gigawatts are like, maybe it’s 700 megawatts right now, and they’re thinking about getting to a gigawatt because there’s not enough chips to run them. Oh my goodness. We’re going round trip now. Okay.
Stephen Lacey: Hold on. Before we move on to the next one, just one last take.
Caroline Golin: Oh my goodness. Sorry.
Stephen Lacey: Do you think there will be a significant number of off-grid data centers? Yes or no?
Jigar Shah: No.
Stephen Lacey: And do you think that that means that those data centers won’t get … Those are data centers that will not get built off-grid in the first place?
Jigar Shah: I was just with all of the major financing companies in New York, they have said that they are not going to finance off-grid data centers. They will not finance them.
Stephen Lacey: Even data centers that eventually will have a grid connection.
Jigar Shah: Even data centers … Well, eventually, meaning if it’s off-grid for five years. If it’s off-grid for two years or more, they are most certainly not going to fund them because data centers that have already tried to run off grid, they break a shaft in 10 months. If you take the seven Chevron like GE Vernova like CCGTs, you stick it in the Permian and you run them off grid with negatively priced Waha gas, right? If one of those turbines breaks, there are no spare parts for that turbine. You’re in the back of the queue with GE Vernova for five years before you get another turbine. And so then what are you going to do? Say, “Well, let’s just run it off at three turbines and we’ll keep the other four as spare parts. Hell no.”
Stephen Lacey: That does feel to me like a serious risk. Absolutely. So let’s turn to the last segment, the last way of thinking, which is 10x better, not 10% better. We all love the idea of moonshot thinking. I don’t want to disparage moonshots here. We need the people driven by the hardest problems with the biggest payoffs, fusion, SMRs, enhanced geothermal. If any of these actually deliver at scale, they could radically change the game and the excitement is real. I’m here for moonshots. I think Theo Von captured the excitement pretty well in his conversation with Sam Altman.
Theo Von clip: I think we need to get to fusion as fast as possible. Get to what? Nuclear fusion? I think that is the … Oh shit, what is it? Where you basically knock two small out of this together and it takes a bunch of energy, but no carbon, very clean, doesn’t really harm the environment and power can become abundant and pretty limitless on earth. And we get out of all the current problems we’re in.
Theo Von clip: You guys sell tickets to that or what do you think that would be like?
Theo Von clip: Yeah, I think we just
Theo Von clip: Watch that shit. I mean, yeah, people go to monster trucks. You don’t think they’ll roll up to watch those two things at each other?
Theo Von clip: The atoms at each other? Yeah. It’s pretty hard to watch at each other, but maybe somehow we can do it.
Stephen Lacey: Okay. I love that Theo Von is bridging the Monster truck enthusiasts and the fusion enthusiasts. But yeah, I don’t think there’s anybody in this room who doesn’t love the idea of Fusion, but there’s kind of a live question as part of this, which is a lot of the biggest players are hanging their hat on major tech breakthroughs right now. What are they not doing in this three, four year timeframe that they should be doing if they are looking out seven, 10 years for some of these groundbreaking technologies, right? Transmission needs to get built right now. There’s interconnection forms, interconnection reforms. There’s a ton of creative work to get new capacity across the system. So this 10% solution deployed at massive scale is kind of what will move the needle right now. And I want to run through some of the moonshots that I’m hearing that are pretty common.
Stephen Lacey: One, why can’t we just stamp out a bunch of AP 1000s and call it a day?
Jigar Shah: No.
Stephen Lacey: Why not?
Jigar Shah: I mean, why not? We just finished Vogle. All 9,000 people that we trained are now building data centers somewhere in 50 states, so they’re gone three sheets to the wind. If I wanted to restart VC Summer’s construction, I’d have to what, train another 9,000 people because they’re probably not capable of being brought back to VC Summer. And then when you look at all the utility companies are like, “We’re not going to bet our balance sheet on this.” And so now there’s this mythical fund that Howard Lutnick controls over a Department of Commerce that’s been funded by the Koreans and the Japanese who love us right now because of the Strait of Hormuz. So I’m sure that they’re all in on the $85 billion fund that they promised. And so what are we talking about? Seriously, no one’s building an AP 1000 in this country, certainly not for me.
Stephen Lacey: I read on a thread on X that we would, Caroline, why can’t we build a bunch of AP 1000s?
Caroline Golin: I think the biggest problem with massive capital initiatives like AP 1000s is that you have to trust that there is a long-term price signal. And I think at the heart of this whole conversation, there is a general distrust that the fervor around offtake is going to maintain past five years because if something else breaks through and it’s lower cost or easier to scale, then the ADD of the tech industry is just going to pivot and do that. And that’s like an institutional problem across this entire economic transition is that there is a general feeling of lack of partnership. I think there’s a lot of enthusiasm around initial seed funding, but are these entities really going to stay there for 10 years and build really hard stuff? I think that a lot of the companies in the industry will say they’re unsure. And so at the end of the day, AP 1000s have a lot of commercial cost over on development risk.
Caroline Golin: And so you have to have someone who’s like, “I’m in it for the next 10 years and I’m willing to do that.” There isn’t a single hyperscaler out there that has a planning cycle past three years. So even if they wanted to, they can’t section out capital development past 2030 right now. And these type of capital projects require a longstanding trusted partner, which historically has always been the federal government and it’s still not happening. That’s the sober answer.
Stephen Lacey: And why can’t we just build all new data centers with SMRs? What about that plan?
Caroline Golin: Well, we haven’t built an SMR yet. That’s the reason why.
Stephen Lacey: But I heard on X that we would.
Caroline Golin: On Twitter. Well, we will. I mean, we will. I mean, they will get built. I mean, it will all get built. I have complete confidence that fusion is going to happen, SMRs are going to happen. The question is what is going to happen between now and then?
Stephen Lacey: What’s a moonshot you guys believe in?
Jigar Shah: I think we can actually use the extraordinary companies here like Astrid’s in the audience, like GridCare, like Google, Tapestry, and we could free up 100,000 megawatts of capacity that we already have trapped in the grid next week. If we just freed up the siloed data that all of the utilities have and we said, “Oh, you could fit a hundred megawatt data center here. Let’s do that. Oh, you could fit a 300 megawatt data center. Let’s do that.” I would be super inspired by that effort, right? That human effort to get stuff done. I just think that, and I don’t know why we’re calling it a moonshot. We just went around the moon again. That seems like very easy to do. We got to call it a Mars shot.
Stephen Lacey: Caroline, what’s the moonshot you think we should be pursuing and that you believe in?
Caroline Golin: Okay. I’m going to say this a little differently. If research in battery storage and the cathode anode match, if that holy grail breaks through, the entire industry, power electronics to the grid, smart home, AI training radically changes. There’s a lot of money and focus going into this race, but if that was to break through within the next five to 10 years, the ability for the grid to become a micro transaction, commodity transaction opens up. The ability for integration and training, the ability for our old word, non-wire alternatives to breakthrough, it just radically changes everything. So moonshot is to me the getting over the technical barriers, the substrate barriers, the materials barriers, the electro transport barriers in the storage space. And if that happens, it is like hands down the biggest game changer across everything that we touch and know. Yeah.
Stephen Lacey: Well, we’re kind of running short on time, but damn it, this is our podcast and our conference, so we’re going to take a few more minutes to bring in some ideas from the audience here. And so let me go get my basket. So we’ve asked the people out there to share their ideas for solving the AI energy bottleneck, and we’re going to pull a few out of these out of the basket and get your responses. So here we go. These are totally at random. I haven’t reviewed any of these and clearly I have not because this one is stationary cycling fitness class. Have all the bikes generate electricity and pay the participants? What do y’all think? That seems like quite a moonshot.
Jigar Shah: Yeah.
Stephen Lacey: Would you fund it?
Jigar Shah: You tie it in with Ozempic?
Stephen Lacey: Yeah.
Jigar Shah: Stick it in.
Stephen Lacey: Ozempic is cheap enough now. I think that could be democratized.
Caroline Golin: No comment.
Stephen Lacey: Have every state or county pledge a quantity of acres as a percentage of total solar development require open lands with zoning change, include streamlined permitting? Sorry, that one. I’m trying to figure out how to ask that.
Jigar Shah: Well, I mean, the thing that I don’t like about all that stuff is just that I think landowners should have the right to do what they want to do with their land. Today, if you wanted to plant corn right now, you would lose $100 an acre. I think it is very sensible for them to sign a lease for $1,000 an acre with a solar developer and make a guaranteed profit for the next 20 years. And I don’t want the county limiting it to a certain number of acres in the same way that the county didn’t limit the certain number of acres for ethanol production.
Caroline Golin: I mean, I think that’s right. I don’t see that really working because then you are going to require county commissioners to understand the things that the dude bros don’t understand already. So no.
Stephen Lacey: Do nothing and wait for the singularity so AI can solve its own energy challenges.
Jigar Shah: Well, first they got to solve the Straits of Hormuz. I mean, if there’s an AI problem that’s right there, I feel like they should just solve it.
Caroline Golin: I feel like that was like an Ice Ice Baby song.
Jigar Shah: Going to start rapping.
Caroline Golin: No, I am not. But you could take that on. I think that’d be a good look for you. Maybe we should start doing that like musical time with Jigar.
Stephen Lacey: Quick and massive build out of DG solar plus battery energy storage front of meter assets. All right. How are we feeling about how that?
Jigar Shah: I think that the idea that I had suggested and Caroline suggested are related, right? To free up the 100,000 megawatts of capacity on the grid, you’re going to have to free up 50 hours, 75 hours, 100 hours, and storage is the way you’re going to do that, whether you put it on houses or churches or central.
Caroline Golin: Yeah. Massive DERMS platforms integrated with HEMS on the distribution system. Yeah, for sure. Absolutely.
Stephen Lacey: Okay. Last one here. Using existing rail line infrastructure to transport clean energy, maybe build transmission lines, solve the transition … Sorry, this one was a little hard to read, but basically use existing rail lines to build transmission infrastructure.
Jigar Shah: So we did this in the first day at the Biden administration, and so it still exists. So there is a framework by which people can use all of the highway right of ways to build transmission lines. It’s about one and a half times more expensive than what people think transmission should cost, which of course it costs an infinite amount since we’re not building any. So I think that that is available. And we did a study with NREL around which highways you would choose. I-95 would be difficult, but I-10 is super fast and all that stuff. And then the rail infrastructure is also available like Sue Green, which was that line that went from MISO to PJM was using rail infrastructure. So I think this is something we can do. I think the problem that we have right now is that there’s a governance challenge around who makes decisions around transmission.
Jigar Shah: I don’t think this is a cost allocation challenge as much as … There’s a lot of owners of existing transmission that don’t want new transmission to get built.
Stephen Lacey: Caroline, do you buy the concept?
Caroline Golin: I mean, from an engineering perspective, yeah, but it’s an inter-intrastate authority issue. So again, it’s a human issue. Yeah.
Stephen Lacey: Well, if Caroline hasn’t gotten tired of Jigar’s rants and the data center industry hasn’t come after Jigar and the singularity hasn’t come to take our podcasting jobs, you can find us next week.
Stephen Lacey: This is Open Circuit. Thank you everyone for being here at Transition-AI. Open Circuit is produced by Latitude Media, Caroline Golan and Jigar Shah, my co-host. The show is produced and edited by me, Anne Bailey and Sean Marquand. You can find all of our podcast episodes anywhere you get your shows and of course watch them all on YouTube and find out about our upcoming events at latitudemedia.com/events. We’ll catch you next time.
