The exponential growth of AI is colliding with the linear reality of building energy infrastructure — forcing a rethink of how tech companies power their ambitions.
Is the corporate clean energy playbook becoming obsolete?
In this live episode from Transition-AI in Boston, we dive deep into the infrastructure challenges of the AI era. We’re joined by Caroline Golin, who spent eight years building Google’s energy strategy before leaving earlier this year.
We examine how tech companies moved from a buyer’s market with abundant resources to a seller’s market with serious capacity constraints.
“The vast majority of procurement in this country looks like single-source renewable PPAs, and that’s what it’s been for years,” Golin explained. “That wasn’t going to get us to where we needed to go.”
Then we turn to a financing puzzle. We need to see more than $5 trillion in capex spending on data center infrastructure by 2030 to keep pace with computing demands from AI. But even with many different types of capital trying to solve the infrastructure challenge, they all want completely different things. How do we align them?
The just-in-time procurement model that served the hyperscalers for a decade must evolve into global-scale, integrated infrastructure development. Can regulators, utilities, and policymakers keep up?
Credits: Co-hosted by Stephen Lacey, Jigar Shah, and Katherine Hamilton. Produced and edited by Stephen Lacey. Original music and engineering by Sean Marquand.
Open Circuit is brought to you by Natural Power. Natural Power specializes in renewable energy consulting and engineering, supporting wind, solar, and battery storage projects from concept through financing. Discover how we’re creating a world powered by renewable energy at naturalpower.com.
Open Circuit is brought to you by Sungrow, the trusted provider of PV inverters and battery storage. With over 605 GW installed worldwide and a BloombergNEF ranking of “most bankable” in power conversion and energy storage, Sungrow provides solar tech you can count on. Learn more at sungrowpower.com.
Transcript
Stephen Lacey: From Latitude Media, this is Open Circuit. This week we’re live from Transition-AI in Boston. What actually happens when one of the biggest technology disruptions in history meets the reality of actually powering it? The exponential path for AI is colliding with the linear reality of building infrastructure, and it’s creating an opportunity to rethink the corporate clean energy playbook.
We’re joined by Caroline Golin this week, who spent nearly eight years at Google developing and executing energy strategy. She’ll share some insights on what worked, what needs to change across the tech industry, and what comes next for powering AI infrastructure.
I’m joined by my co-host, Jigar Shah, who’s a clean-energy investor and former director of the Loan Programs office. How are you, sir?
Jigar Shah: I’m great. I’m just noticing that two of us on this panel of four are former.
Stephen Lacey: Makes for more candid conversations. We always go after the former people. I checked your LinkedIn just to see if you had updated everything, and your LinkedIn profile still says podcaster, and your last job was layoff, position eliminated.
Jigar Shah: That probably was generative AI.
Stephen Lacey: Katherine Hamilton is the co-founder and chair of 38 North Solutions, a clean energy public policy firm. Katherine, you’ve come down out of the mountains to sea level to Boston.
Katherine Hamilton: I did. Thank you. Unfortunately, I can’t sit on the floor with my papers, but y’all have kindly given me a table. So I’m grateful, Stephen.
Stephen Lacey: And our guest made a career change as well. Until this spring, Caroline Golin was the global head of market development and innovation at Google where she built the company’s clean-energy strategy. Caroline, we’re so delighted to have you here.
Caroline Golin: I am ecstatic to be here amongst my friends really. I mean, yeah.
Stephen Lacey: After today, maybe you can add podcaster to your LinkedIn page as well.
Caroline Golin: I thought this was a job interview, actually.
AI’s impact on energy strategy
Stephen Lacey: So I want to start with a look at how much AI has changed the game for energy teams at tech companies across the industry, and there’s no better person to do that with. And to start, I want to play a clip from Sundar Pichai’s speech at Google I/O recently. That’s the company’s developer conference. It happened in May, where Google unveiled a lot of new AI products and AI adoption progress.
Sundar Pichai: And the world is responding and adopting AI faster than ever before. This time last year, we were processing 9.7 trillion tokens a month across our products and APIs. Now we are processing 480 trillion monthly tokens. That’s about a 50X increase in just a year.
Stephen Lacey: So, Caroline, this kind of expansion is really exciting for a lot of people in the tech industry. But when someone like you hears this, do you just curl up in a ball in the corner, thinking about the energy implications?
Caroline Golin: No. No. Actually, when I came to Google, I came with a real enthusiasm for what a large platform like Google can do and has done, and I think anyone who goes to a large tech company within the last 10 years to work on energy went there because they wanted a big stick to make big changes in the industry.
I think when the AI boom started to take place about two years ago, what we got scared about was whether or not we had a fit-for-purpose model to go to market, and I think we’re all still a little scared about that. There’s a huge cultural change that has to go on within every tech company right now in terms of how you build teams to execute on what is a sustainability goal versus building teams to execute on what is an infrastructure goal, and empowering the AI needs is an infrastructure goal.
I didn’t curl up in a ball. I don’t think anyone’s curling up in a ball. I think it meant for me and it means for everyone that the menu of options that we needed to learn how to deploy got a lot bigger, and that gets a lot more interesting. It also meant that we needed to change our business model.
I think the vast majority of procurement in this country looks like single-source renewable PPAs, and that’s what it’s been for years. That wasn’t going to get us to where we needed to go. We needed to figure out how do you scale storage? How do you scale demand response? How do you do real-time arbitrage, like we were just learning about here? And then how do you scale things like nuclear?
That wasn’t in anyone’s playbook really. It was more so in Google’s than others, I believe, but I don’t think anyone two years ago was really thinking about firm capacity solutions. We were thinking about deployments of megawatt hours, and that’s a very different business model. But, for me, that was exciting because it meant that we would actually be able to change the utility industry.
We’d be able to change market structures. We’d be able to change the way we think about tariffs and valuation and deployment of capital. I can’t say I slept a lot, so maybe I curled up into a ball in my bed at night for like three hours. But no, it was exciting. It continues to be exciting, and the teams that I’ve built and that others have built are the best in the world.
Rethinking corporate clean energy goals
Stephen Lacey: Well, that sets us up nicely for the conversation. Let’s talk first about the state of play for tech companies. The period from 2015 to 2022 featured a series of increasingly ambitious, aspirational targets from the top tech companies, the most ambitious being Google’s 24/7 initiative, and Microsoft’s carbon negative moonshot.
Those goals are still in place, but they’re seriously challenged in the AI era, and all the major tech firms have seen large increases in energy demand and subsequent emissions increases. So how do we rethink the framework for decarbonizing the tech industry, and what has the industry collectively learned? That’s what we want to tap into with you, Caroline.
Katherine, first, how much pressure is the industry under right now?
Katherine Hamilton: I was born and raised and still live in Virginia. Virginia has 400 data centers, and I believe it’s 70% of the global internet traffic goes through the commonwealth. I just happened to be talking to the guy who helps us on our property out in the mountains, and he brought his son with them.
I was like, “Oh, this guy’s just going to help out in the shed and all this.” And I said, “So what do you during the day?” And he said, “Well, I am the guy who decides where the energy comes from for the Amazon data centers.” I was like, “Well, we need to chat.” Amazon has 93 data centers in Virginia. They’re putting in 12 buildings, 11 of which will be data centers in Fredericksburg.
I said to him, “So what are y’all doing? Are you putting in storage and solar?” And he said, “No, we’re building two substations, and we’re getting everything we can from Dominion because they have enough on the nuclear side.” So it gave me some insight into this is just about getting it done really, really quickly.
We’ve heard this over and over today. How do we get speed to power, and what does a data center or developer have to think about in that context? So when you go back and look at the goals, the goals of the tech companies, a lot of them are for 20/40. That’s totally workable. I mean, y’all met and exceeded your goal at Google, and we won’t talk about Apple right now. But most of them are actually doing okay.
If you saw yesterday, Talen and Amazon just announced the nuclear deal, which was actually turned down by FERC because it was considered behind the meter. They’ve changed it to front of the meter. So now they have a really good deal for nuclear power. They’ve done this. Meta has done this, as well, in Illinois. So I don’t think right now we have an issue of them not meeting their goal.
Now, that’s not to say emissions haven’t gone up. Between 2020 and 2023, emissions have gone up 150%, but that’s mostly indirect emissions. So that’s where you’re getting your power. That’s where you’re getting your steam, your cooling. So those are things we have to pay attention to, but I don’t think it’s not doable.
Stephen Lacey: Jigar, we’re heading into new territory clearly, but where?
Jigar Shah: Well, I think that we’re actually heading in the exact territory that we always were heading into. I mean, the weird thing about this is we’ve always had a plan to get to 2050, and last week, I think, was the halfway point to 2050. So we’re now in a period of valuation.
I think, as Carolyn suggested, Google and many of the other companies were instrumental in figuring out contracts for differences contracts, getting solar and wind set up, all these other things going, and then there was a clear realization starting probably 10 years ago from Google saying that solar and wind was not going to be able to single-handedly decarbonize the world and we needed 24 by seven matching.
There was a conversation about geothermal and nuclear, which people just laughed at and didn’t do anything about. And then sometime over the last four years, we decided that all these other tools that people had not been focused on were now important, grid-enhancing technologies, virtual power plants, demand flexibility, geothermal, nuclear, all these other technologies.
I mean, just to do a thought experiment for you, theoretically, if you just took all the major real estate companies in the country and just took two of their parking spots and put Tesla Megapacks on those parking spots and then just operated the grid differently, you could just run the grid. It wouldn’t be that hard. Most of our natural gas power plants are running at 32% capacity utilization.
You would just run it three percentage points more, and then you would just load level the grid through the batteries. So it’s not that we can’t do this. That is pretty expensive, putting those batteries everywhere. So now what’s the cheaper way of doing it, et cetera? It turns out that building new natural gas and building new coal is the very most expensive way of actually accomplishing this goal.
So the fact that all of these companies have figured out how to do all of these cool things, and you’ve got this manufactured crisis because of AI, is now forcing everyone from governors to regulators to utilities to say, “We don’t want to culturally change anything we’ve done for the last 50 years, but if we don’t, then we’re going to keep having 10% rate increases for as far as the eye can see. And so I guess we’re going to have to do things differently than we’ve been doing it for the last 50 years.”
And then the data center companies, to their credit, when they go to these private meetings in Oklahoma, like at Harold Hamm’s private meetings, they’re saying, “We’re not abandoning our goals.” Even when Chris Wright and Doug Burgum stare them down and say, “Please abandon your goals,” they’re like, “No, we’re not abandoning our goals, and we’re going to make you do things in a way that’s most rational for all consumers.”
We’re not quite there yet. There’s a lot of regulators who are sticks in the mud like folks in Maryland, but we’re getting there, and we’re trying to get everyone on board. But these kinds of conferences, these kinds of things are critical to educating everybody around exactly how this gets to our 2050 goals, which remain unchanged. We’ve got to decarbonize the energy system by 2050.
Stephen Lacey: So, Caroline, around the 2020 timeframe when we saw the most ambitious targets come out, it was a buyer’s market. There was a lot of overcapacity. When did that flip, and how challenged are those targets now?
Caroline Golin: Yeah. So everyone went through this fervor of setting a 2030 goal in the early 2020s, and that really was in a space where, as Jigar alluded to, we were all wrestling with what is the best way to deploy our capital to change the system because we wanted to make a dent in climate change. So whether you’re Amazon or Meta or Microsoft or Apple or Google or whatnot, you went back with your team and you said, “What’s the best way for us to use our dollars?”
I think, from a philosophical perspective, everyone stands in their own parking lot and agrees that this is the best way we’re going to use our capital. Amazon is going to deploy renewables at scale, and that’s the best way they’re going to use their capital. Google said very early on, “We wanted to figure out how to scale the technologies that would fully decarbonize the grid.”
There was no valuation for how to do that. If you were working with a utility, you could go through a green tariff program. If you could get solar in and it avoided cost, great, you could get the recs for it. If you went on the open market, you could put out an RFP. If you could sign PPAs, if they could come in and you could hedge against natural gas prices, great. You were going to get that through.
Those were really only the two data points by which we were going to deploy capital or anyone could deploy capital. Of course, we all had budget, and we were going to pay premiums, and we were going to drive down the costs, but generally that’s what you were working between. Are you going to get recs and get something under avoided costs so it pencils with a utility, or are you going to get something that you think you can hedge against natural gas prices with burn overall portfolio in the open market?
Then about two years ago, and I was that person at Google who read all the IRP summaries, I’d have my team give me a whole presentation of where is utility planning going. I think we very quickly realized that the capacity that we’d all enjoyed and really not thought flippantly about, but not really integrated into an energy strategy, was going away, and it was going away quicker than we could develop the technology solutions to meet it.
That was when I got really excited about what we could actually do in terms of innovation because I don’t think that if we hadn’t had a capacity issue, that a lot of these tech companies would be running after nuclear in the same way that they are running after it now because it wasn’t in their original goals.
I’m not sure that any of us would have gotten to the point where we recognized we need to scale gigawatts and gigawatts and gigawatts of demand response to do this, or we need to focus on how you put distributed resources on the distribution system so you alleviate stress on the transmission system. I’m not sure we would have gotten there in the way that we’re getting there now.
I think you would have seen it on the margins. I think you would have seen fervor to do the new cool thing, but do I think we would have changed policies? Do I think we would have changed regulations? Do I think we would have created the Clean Transition Tariff? Do I think we created new structures that allowed us to invest directly in capacity and figure out a capacity value for things?
I’m not sure we would have gotten there. I’m hopeful we would have because it’s the right thing to do, and it’s the actual way to think about grid planning, but it’s actually the run on AI which has forced all of us to say, “We need to look beyond just recs,” which is what we were focused on for such a long time. That’s a great thing.
I think that’s the Pandora’s box that we get to open now, but it’s going to mean a very different way to do structuring, and it’s a cultural change within the procurement model in general, and it’s going to require much more integration and resource planning, to use a very utility term. We need to be doing integrated resource planning from a customer’s perspective. Yeah.
Katherine Hamilton: Yeah. I think that is a super important point. I’ve also been involved in a lot of IRPs and a party to a bunch of them, and I’m always screaming about we have to use our distributed resources better. You have to even include them in your modeling because, most of the time, utilities do not include them. They’re just this is the load side, and this is all the other stuff that provides electrons.
By pushing on that planning, it is crucial from both the supply side where, as Rob Gramlich was talking about earlier about having to do transmission-planning, making sure your transmission-planning, your grid-planning is really strong, but also what are all those resources out there that we can build quickly that are on the customer side that we can then bring to bear because rates are going up everywhere, and it’s only going to be exacerbated if we don’t pull every electron from every resource we can find.
Caroline Golin: Well, on the rate piece, I will say that if rates are going up because of data-center deployment, then we’re not working hard enough with our regulators because the regulated model should insulate from that, if done correctly. There are tariffs, and there are ways to ensure that that cost shift doesn’t happen and that cross-subsidization doesn’t happen.
So I think that this is the conversation, and you hear it in the news, “Oh, AI is driving residential rates up.” But the reality is the regulated model and the utility model has space for that not to be the case, and so it’s actually on the onus of us working together to figure out what are the right tariff structures, what are the right cost structures because I don’t see, and in my tenure at Google I know we never went to market saying, “Oh, we want others to pay for our infrastructure.” Absolutely not.
I don’t think that any of my peers in the industry wanted that either. The problem is we haven’t evolved the business models and the regulatory structures to make sure that that doesn’t happen. But that can happen, and it’s incredibly important that we work on that.
Stephen Lacey: So what I’m hearing you say is that a lot of this work initially during that phase was symptomatic work. Jigar, any comment on that? What do you think is the most interesting, innovative stuff that came out of this era?
Jigar Shah: Well, I think we’re still going through it. I mean, to Caroline’s point, I believe her when she says that that was not their intention, but for all of you who are in the data-center business, you know that that’s bull. They are comparing five different data-center proposals against each other, and the one that’s offering $64 a megawatt hour is getting selected over the one that’s $79 a megawatt hour.
The one that’s says $79 a megawatt hour actually worked super hard with the utility to make sure there wasn’t a cost shift, and the one that was $64 a megawatt hour got one over on the local guy and got him to give you a cheaper deal. The CFO of Microsoft or Google or whatever is like, “Oh, yeah, I want $64,” right? Now, was it their intention to screw over poor people in Baltimore? No. But is that what happened? Hell yeah.
So I just think it’s important for all of us to recognize that we’re in this unique situation, and the reason why I pick on Maryland is just because I live there. All of them are friends of mine, and I’m so pissed off. But because, for instance, when PJM came out with all their transmission upgrades, it was all for data centers. There’s one transmission line going from West Virginia to Frederick, Frederick to Virginia, clearly for Microsoft data centers in Frederick.
Microsoft came to the Public Service Commission and said, “We don’t want a cost shift. We want you to charge us for that transmission line.” What did Fred Hoover and Kumar Barve do? Say, “Well, for 25 years, we’ve just rate-based all those transmission lines that come from PJM, and it would take us six months to actually do this rate case and do this thing that Microsoft can do. So we’re just going to rate-base it, and then that’s just easier. It’s a three-day process instead of a six-month process.” So Microsoft’s going to get much cheaper priced power.
So we’re just in this weird spot where we have all these technologies. We have all these liftoff reports. We have all these innovations around the Clean Transition Tariff. We have all these things, and then you have lazy-ass regulators and super lazy governors. I don’t think Wes Moore has any idea how electricity works, but I think Josh Shapiro actually has taken the time to understand how electricity works.
So we’re in this weird spot where you can absolutely figure out how to help people in a rising tide, lifts-all-boats situation, but it requires a level of dedication and intentionality to engage in the process, which we have not universally seen.
Stephen Lacey: Somehow I guess that in your regulatory work, you did not use the language, “Lazy-ass regulators.”
Caroline Golin: No. No. Or with anyone really. I’m much more polite than you. But what I will say is so there was just some legislation that was passed in Minnesota, and I think that this becomes an interesting model moving forward because Minnesota was wrestling with how they open up their data-center market. There’s a lot that’s been going on back and forth within that legislature.
But one of the places that we landed was a recognition that the utilities had to offer tariffs that allow data centers to invest in their own capacity and create an avenue for speed, but also ensures that this cost shift doesn’t happen. That type of legislation, if you could see that across the country, then you open up a different type of bilateral market.
You open up a market where developers, different stack technologies, and data-center customers can come together with a prepackaged solution, and it doesn’t even have to touch the regulated process that we’re used to because that process should be kept in place, and it should be kept in place for commercial and other industrial and residential customers.
But if you’re trying to fast track infrastructure and trying to do it in a way where everyone is more beneficial at the end of it, you’re going to need to create a separate process, and it’s through these types of tariffs. That’s going to require legislation across most of the country. Many of these utilities are like, “That’s great, Caroline. I’d love to do that investment in X, Y, and Z with you. Can you go get the legislature or the governor to tell us to do it? Because without that, it’s gray space. We don’t really know.”
So I think we talk a lot about there’s not enough capital. There’s none of this. I think there’s absolutely enough capital to meet this challenge. I think what we don’t have is the lanes and the business models working directly with utilities who want to see this happen. I mean that’s how we got Fervo and geothermal with NV Energy is because we worked for two years, governor, legislature, utility to get that model put in place.
If you can put those in place across the country, the technology opportunities, sky’s the limit, I think, at that point.
New strategies for financing AI infrastructure
Stephen Lacey: So let’s turn to that more explicitly. This is an issue that you’ve continued to focus on. How do we actually finance the infrastructure at the AI energy nexus? We need to see trillions of dollars of capital spending on data-center infrastructure by 2030 to keep pace with computing demands, and there’s this sort of economic puzzle that emerges.
We have a lot of different types of capital all trying to solve the same infrastructure challenge, but they all want completely different things. And then meanwhile, AI is just changing the game for power investments and who pays for them. So as this handful of tech companies drive large power demand, there’s considerable debate over what they should pay for.
So let’s talk a little bit about the financing regulatory models that are fair and align these different types of capital. Jigar, do you agree with Caroline that we have the capital to support this expansion?
Jigar Shah: Oh, yeah. No, I mean the capital is easily found. I think the problem that we have is that the capital has come from the wrong places. So when you think about what our asset class looks like, it looks a lot more like real estate actually than it looks like tech. So when you think about tech, tech is 100 extra turns. Tech is these very high gross margins, all these other things. We don’t have any of that.
So it’s one thing to say, “We’re going to start with 5,000 business presentations, and there’s going to be venture capitalists who provide seed-stage capital or A-Round Capital.” That’s fine, whatever. But at the time at which you go from product-market fit and to successful commercial contracts where people really like your product, you’re now heading towards real estate.
The reason I say real estate and not infrastructure is, remember, the way infrastructure works is it’s really about long-term cash flows. So whether it’s parking meters or whether it’s toll roads or whether it’s other things, that’s what it looks like. When you look at real estate, I mean you guys all know what real estate looks like right now.
You’ve got 50% empty commercial buildings. You have all this stuff. That’s where this goes. So when you have dynamic line ratings and you have the ability to probably cover all of Texas with dynamic line ratings for $50 million, you unlock huge amounts of capacity. The problem in Texas is who’s exactly supposed to rate-base that? Is it Encore? Is it the Generation? Is it the retailers?? Who’s supposed to do that?
But I’m pretty sure if someone just came in Merchant and said, “I’m going to do it. Now we have all this excess capacity. Who wants to buy it?” I’m pretty sure they could make 3X the money off of that capacity that they opened up, but that is not how anybody in the dynamic line rating investment industry works. The same thing’s true for DERs and virtual power plants. The same thing’s true for electric vehicle charging.
So the way in which everyone is trained to think is, on this side, they’re trained to think 10X, 20X, 30X returns when they can barely hit 3X returns. And then on this side, they’re taught that we need rock-solid 20-year PPAs, fixed-price contracts when they absolutely don’t, and in fact, that they’re not there to be found, and so we actually need people to look at Merchant on an individual deal basis.
But when you look at an entire portfolio of projects in a diversified geographical area, they largely cancel each other out, and you can get actually pretty consistent cash flows across that entire portfolio. But again, where does that capital get allocated from? So when you talk to the pension funds, they’re like, “Well, you belong in this allocation, not this allocation. You’re in the wrong box.” I’m like, “No, I’m in the right box. You’re in the wrong mood. Let’s figure out how to solve this.”
So we don’t have a capital problem, and everyone is just long capital. When you go right now into the marketplace and say, “I’ve got 200 megawatts with the batteries, and I need financing,” all of those deals are getting done this week. There’s so much money that they’re like, “Send me more deals. I have so much money. I’ve already raised it, and now I have to put it out the door.”
So we don’t have a lack of capital, but we do have a lack of expectations management on both sides, from the allocators and then the folks who are developing the projects.
Stephen Lacey: So, Katherine, can you break down, we talked about trillions of dollars in capital needed to support data-center infrastructure, where does energy fit into that?
Katherine Hamilton: So you have to think of this as infrastructure includes the data center build out itself. It includes electricity. It Includes all the supply chains that provide electricity. It requires permitting, the land use. It’s not just the real estate. That’s a piece of it. And then on top of that, of course, sits all of your platform, whatever you’re doing with that center, and then the applications on top of that.
So infrastructure is the most important thing, first step. It was very interesting because Senate Commerce Committee had a hearing, and Brad Smith, who is the president of Microsoft, testified, and he said, “The one piece of infrastructure that is the biggest logjam for me is human capital. There are not enough people to build the infrastructure.” He said, “We need half a million electricians that we do not have now.”
So his biggest issue was let’s make sure that, and he calls it diffusive or we have to get it in every single part of society. We have to start changing the way we think about infrastructure and the way we educate around infrastructure so that we get people on the ground to even build these things. We don’t have a demand problem unless we can build them.
Caroline Golin: Yeah. I’ll just piggyback on that. I mean I actually am hopeful that the energy challenge can be solved, and I think it can be solved and unlock an array of new technologies that we could only have dreamed about would be commonplace five to 10 years ago. My bigger concern is that we have the capital and we get shovel-ready and there’s no one to dig because we treat all of these industries as silos.
We have workforce for data centers. We have workforce for the grid. We have workforce for generation. I see a lack of cross-cutting partnership around workforce that I think if we don’t start tackling that, we’re going to end up losing this global competitive race mainly because we just didn’t teach people, and we didn’t figure out how to do procurement of workforce at scale across multiple different sectors to build. That worries me a lot.
I’ll say to what Jigar was talking about earlier, I always say there’s these four quadrants. There’s traditional private equity. They’re all sort of the middle school dance, and they don’t really know how to dance with each other, but they’re there and they’re looking at each other. And then you’ve got traditional customer procurement, which for a long time, as I said, was an energy-only game, like sign me, energy plus recs. That’s the only value stream. That has to change.
Then you’ve got utility. That’s a 50/50 debt split. I don’t think people appreciate enough, even if a utility said, “Yes, I want to go build that SMR for you,” how are they supposed to raise the capital to do that? They can’t. So that has to change. And then you have a traditional developer model which says, “I want you to sign me a 30-year energy-only PPA. That’s what I built all of my wealth on.” That doesn’t work either.
So you have these four different models that actually aren’t targeting the same problem at the same time, and there are also four different models that could share workforce together and maybe figure out how to move in unison. But I think in the intersection between those four, if we can start changing our language and start working differently in the way we attach value and also in the way that we think about contract terms and the way we think about stacking technologies and an entirely different approach to when you think about procurement at scale, I think we can get there.
The data-center industry has been a just-in-time industry for 10 years. We now have to be a global-scale procurement industry. It’s a huge cultural shift.
Stephen Lacey: Can you unpack that a little bit more? When we were talking before, you outlined this single-product concept. So what does it mean to treat data centers and energy infrastructure as integrated development rather than separate procurement?
Caroline Golin: Yeah. I’ve said this before, but one of the things that I worry about in the energy space and in the data-center space is that we tend to treat it the same way we treat healthcare, which is we treat symptoms. We don’t see the overall health of the system and try to see how we holistically we do this. I think it’s in the utilities’ best interest to plan and know where these data centers go.
That’s going to optimize their grid build-out and reduce the headaches for residential customers, for siting, for permitting, and for cost. I think it’s absolutely in energy technology companies’ best interest to work at it from an integrated approach where you are integrated into the land as opposed to a last thought or an end note.
For so long, we would go to market. We would buy land. We would sign an energy service agreement. Then we’d come around, and we’d sign a PPA to either offset that energy service agreement or to capitalize on a new technology that we wanted to bring forth, instead of being at the very beginning where we say, “We want this land powered this way, and these are the five technologies that we’re going to go to market with across an entire holding company or across an entire swath of land.”
That’s a scale-procurement question. Everything in this industry has been just in time and to the quarter, and now you’ve got to say 10 years at scale across 15 states across these five different technologies across this sort of power structure and this type of reliability structure. How do you do that? You’re not going to be able to do that unless it’s all integrated in one package, and I think there’s a huge opportunity for the utilities to start directing this.
I think that a lot of us would be happy with that, and I think that there’s a huge opportunity for the technologists to create that full-stack capacity solution. Batteries alone aren’t going to do it. Solar alone is not going to do it. Wind alone is not going to do it. Demand response alone is not going to do it. But if you work together and you figure out those solutions, an on-site natural gas should be working with demand response. Storage should be working with SMRs.
Take that arbitrage and re-put that value back into how you finance a long-term gen-three. Those types of solutions are not going to happen within an IRP planning process. They’re going to have to happen outside and in connection with some equity and a data-center developer. And if you can do that, off to the races.
Stephen Lacey: Jigar, what do you think the current constraints, pressures will mean for positive change? So if we think about this integrated procurement model, do you see new models emerging for building infrastructure in this area?
Jigar Shah: Yeah. I mean you’re in a situation where we’re heading into an extraordinary political catastrophe. So when you think about one in six households in the United States today can’t pay their energy bills on a monthly basis.
You’re in a place where when you go into a state, and you talk to the Uber driver, the Uber driver just complains about their utility bill. You’re like, “You literally have nothing else that you hate more than your utility right now? This is what you’re talking to me about?” So the politicians are now being told, “Hey, we need to do this stuff.” So now all the education work we did, which they were completely ignoring, are now we’re re-presenting.
They’re like, “Oh, yeah. Oh, what’s this AI thing doing? We can actually integrate loads within four minutes instead of 45 days by using AI and using the unstructured data utilities and making decisions more quickly. Oh, we have all this excess interconnection capacity with all these solar farms that are clipping DC power, and now you have new inverters that are being created by Solectria and others, so you can actually absorb all of that DC-clipped power and actually use it to provide storage.”
It’s amazing how much optimization is right there. It’s one-10th the cost of what they were going to do, and they were just not open-minded around listening to it because it was so boring. I mean, I hear myself talk. It’s so boring. I get it.
Caroline Golin: You’re so boring. Yeah.
Jigar Shah: But people are just so mad right now because think about it. 30, 40% rate increases over the last four or five years. It is so mind-blowing how expensive energy has gotten, and so it is now all this boring talk.
People are like, “Wait. What? What were you saying to me about virtual power plants and grid-enhancing technologies?” And, “Wait. What were you saying? How much heat is there under the earth, and how do we actually get it out of the ground? Oh, the fracking industry could do that?” “Oh, yeah. I told you that six years ago. I told you that last year.” “Oh, but tell it to me again. I want to understand it this time. This time, I’m actually listening to you.”
So all of this stuff is actually happening in a way that is just unlocking enormous potential for all of these companies that have, frankly, been wandering in the darkness since they got their first seed round in 2017 and 2018, and it can’t come at a better time. We are ready to go. All these companies have spent seven years with pilot projects, and they’ve been piloted to death.
They’ve got investors, and they’ve got this stuff, and everything’s ready to go. So we are so ready for this revolution, and it’s not just about this revolution, which I think is amazing and it’s definitely going to happen. But it’s about all of these other countries around the world who are saying, “Wait a second. What? We can leapfrog here, too?” So India now is going hard on nuclear power, and they’re deciding on which nuclear reactor to go on.
China has 30 nuclear plants under construction right now and probably has hit peak coal as of this year. You’re just like, “Wow.” The moment that we’re living in is so extraordinary, and all of the people that have just never gotten the time of day from decision makers, never been on the main stage at Davos, have never been in this place, are now on the main stage.
They’re the most important conversation in the room. Everybody now wants to hear what they have to say. It’s so exciting.
The net impact of AI
Stephen Lacey: Okay. I want to close now with a look at AI in service of climate solutions and clean energy. So I want to play a quote from Dario Amodei, the CEO of Anthropic, in a conversation with Ezra Klein when he is answering a question that Ezra asked about the energy demands of AI.
Dario Amodei: Maybe AI makes solar energy more efficient or maybe it solves controlled nuclear fusion or maybe it makes geoengineering more stable or possible, but I don’t think we need to rely on the long run. There are some applications where the model is doing something that used to be automated, that used to be done by computer systems, and the model is able to do it faster with less computing time. Those are pure wins, and there are some of those.
There are others where it’s using the same amount of computing resources or maybe more computing resources, but to do something more valuable that saves labor elsewhere. Then there are cases where something used to be done by humans or in the physical world and now it’s being done by the models. Maybe it does something that previously I needed to go into the office to do that thing and now I no longer need to go into the office to do that thing, so I don’t have to get in my car. I don’t have to use the gas that was used for that.
Stephen Lacey: Caroline, what’s your read on that?
Caroline Golin: Which part exactly?
Stephen Lacey: Well, let’s talk first about AI changing how we do jobs in the physical world and the emissions benefits from changing the nature of work and what we’re able to automate.
Caroline Golin: Right now at least, and I can speak from my experience, right now, the goal with training and to deliver an AI product is to integrate it into our cloud so that our cloud customers are able to work at a super-speed business level and accomplish things that they weren’t able to accomplish.
I think that what we’re going to see in the near term is a lot of fervor around cool AI tools. I mean, I don’t let my kids play with these things, but I’m sure once they can create their own movie with unicorns and rainbows and lucky charms and all that stuff, they’ll be excited about it. I think you’re going to see a lot of fervor around that.
But I think ultimately the application is going to be in traditional processing that is incredibly inefficient and costing money, and bringing down that cost is going to allow for reinvestment in other areas, particularly on the energy space. I mean, I don’t know who said this before at the conference, but there’s the same number of engineers at MISO as there were 20 years ago. That’s true, also a major workforce issue.
But integration of AI tools means that things like interconnection can go faster. It means that reading land permits are going to go faster. So then you’re asking, “Well, what’s the time value of money within energy?” I mean, that’s ultimately what we’re asking with AI tools. If we integrate AI, what is the time value of holding? What’s the time value of waiting in the queue for five years as opposed to three months?
I don’t have an answer for that, but it seems beneficial from my perspective. I think that when I was at Google, what I saw increasingly over and over and over again was excitement around AI as a tool within the medical space, what we were able to break through, the cancer research that we were able to produce at a speed that was unfathomable like 15, 20 years ago.
I know this is off the energy tangent, but the number of my friends that are in the medical industry space that think that perhaps that trial that was going to take 40 years could now maybe take four to seven years and what that’s going to mean, what that is going to mean in work-life productivity, what that’s going to mean in workforce, these are all the questions we actually have to answer. I don’t have the answer to them, but I think that we stop short of thinking about that in a thoughtful way.
But yeah, in short, I think it’s going to have a revolutionary impact on the energy space, but I think where you’re going to see the biggest breakthrough is in traditional processing, medical spaces where there’s so much waste and the time value of money is so incredibly important to people’s everyday lives.
Stephen Lacey: Katherine, what’s your read on what Dario Amodei saying here on AI for energy innovation?
Katherine Hamilton: So I would point to a report, which I love to do. It’s the ICEF out of Columbia University. They just released their second AI for climate change mitigation roadmap, and they point to just so many of these ways in which AI will be helpful. It’s improving power system development generally, improving siting, speeding up permitting, increasing output of better forecasting, I mean DLR is distributed energy, demand response, being able to manage that so much better, and accelerating innovation in batteries and chemistry in V2G.
There’s so many ways in which AI can be helpful in that, and that’s not even going into food systems, manufacturing, road transportation, aviation, buildings, carbon capture, nuclear energy. Just anything that has to do with energy, AI is going to help make that better. So I do think that if we can get some of this going, we’re going to be able to solve some of those problems before they occur, the problems that we’re talking about today.
Stephen Lacey: Jigar, what do you think the net environmental benefit of AI will be? Will it be a net benefit?
Jigar Shah: It’ll be horrible. We had this conversation, I think, exactly in 2018 when you asked me about the internet of things and what it was going to do, and I was like, “It’s going to be horrible.” I mean, the notion that all of these positive impacts are going to outweigh the negative ones … What people are going to do is create these photographs and then make it claymation and then figure out how to make it in a statue and figure out all this other stuff.
I’m like, “Such a fricking waste of kilowatt hours.” But that is what they’re selling. That is what we’re going to do. We’re going to create these things, and people are going to write op-eds that are just dumb because a human didn’t write it, but AI wrote it, and you’re like, “This is the worst op-ed I’ve ever written.”
I just think that I’m a big fan of AI because I think net-net you’re going to see enormous productivity enhancements, and that’s why our GDP is almost double of Europe right now because even though we were the same in 2008, they’ve had no productivity growth and we’ve had consistent productivity growth. We’re going to double them again because of productivity growth.
But let’s not kid ourselves. From the environment’s perspective, this is going to be horrible.
Stephen Lacey: I don’t know. I think that’s a bunch of horse crap. I mean, materials discovery, drug discovery, there’s all sorts of enormous scientific potential to AI tools, and I think it opens up a world of opportunity to speed technology development, speed time to market of technologies.
So I completely agree with you that there’s going to be a lot of stupid stuff that comes out of it, but you’re completely missing a whole category of innovation in AI that is going to-
Jigar Shah: You should go back to your internet of things predictions in 2018 and tell me how that went for you. I’m pretty sure-
Stephen Lacey: I’m happy to do that.
Jigar Shah: … that seven years from now, I will still win this bet. That’s not to say that-
Stephen Lacey: I have a lot of predictions I could go back and bring up to you.
Jigar Shah: I am happy to do it. I’m just saying that I think we all just need to be acknowledging that we are in the chosen billion people in the world, and there’s seven billion that are not chosen and that those people have to still dig up a crap load of stuff out of the ground to make sure that the stuff that we do works. They have to do all of this stuff, and that is going to accelerate 10X.
We’re not in a place where we have replicators and things that Star Trek tried to explain to me was going to happen where we got rid of money because everything was free. We’re not there. So I want everyone to be excited about this moment because I think AI is going to bring this next layer of human flourishing, but we should all acknowledge just how much pain and suffering other people are going to experience so that the top billion have human flourishing.
Caroline Golin: Well, what I’ll say on that is that that’s a personal responsibility, and we are not having that conversation. It’s all of our decision how we engage with this new technology, and if we use it for lucky charm cat videos or if we use it to make sure it’s deployed for cancer research, that’s a personal responsibility as much as it is a policy and regulatory responsibility.
I think you could apply that same mindset with any materials in this country and any consumption in this country. At the end of the day, AI is going to be a product we consume, and how we consume it, if we consume it thoughtfully and articulately and efficiently, I think the net-net is going to potentially take the people that Jigar is talking about and leapfrog them into a space where they don’t go through the Industrial Revolution decrepitude that we went through in the Western world.
If we don’t do it thoughtfully, I think we are going to be struggling with a human rights question for a very, very long time.
Stephen Lacey: Stay tuned for Latitude Media’s Internet of Things Conference coming to you … That was a great place to end it.
Caroline Golin, thank you so much. Jigar Shah, Katherine Hamilton, this was a lot of fun.
