The trillion-dollar collision between AI and the grid

About the episode

The rapid rise of AI has sparked a multi-trillion-dollar collision between our aging power grid and massive, energy-hungry AI "factories." 

Unlike traditional cloud data centers, AI training workloads are incredibly dense and volatile, capable of spiking or dropping power demand in milliseconds and threatening grid stability. Dr. Varun Sivaram, CEO of Emerald AI, says the solution lies in load flexibility. 

By using software to dynamically shift AI workloads across time (temporal flexibility) or across the country (spatial flexibility), data centers can adapt to the grid rather than overloading it. It’s an approach that avoids the need for expensive, off-grid microgrids that drive up equipment costs and starve public utilities of revenue. Instead, implementing modest flexibility can immediately unlock more than 100 gigawatts of stranded grid capacity, ensuring American geopolitical competitiveness, lowering consumer energy rates, and creating a dynamic demand buffer to integrate more clean wind and solar energy.

Critical Capital is a co-production of Crux and Latitude Studios. Learn more about how Crux is financing the future of energy.

Episode transcript

Varun Sivaram: AI data centers can be extremely volatile, and they're extremely sensitive. You could have a cascading blackout if too many of these super sensitive AI data centers join the grid. All of this is to say, this is not your grandmother's load growth story — the grid, which has been stable over more than a century, needs to adapt at light speed to these AI factories.

Alfred Johnson: We've been building large data centers for 20 years. Over that period, the electricity system could handle these warehouse-scale computers, often hundreds of megawatts, for a couple reasons: we had spare grid capacity, and the load profiles of the data centers were predictable, so utilities and grid operators could plan for them.

But that logic has changed in the AI era. The rise of multi-gigawatt campuses is testing the limits of our grid capacity. The computing inside those campuses is very different from cloud workloads. They're volatile. A training run can spike and drop in seconds, and it can cause big fluctuations in demand for power, which means the old playbook no longer works. And in the race to scale AI, how you operate a data center matters as much as where you build it.

Varun Sivaram: There's a hundred gigawatts or more of latent power capacity just sitting on our existing grids that we could use to power data centers. But If they can be flexible, just modestly flexible a small percentage of the year, we can fit more data centers immediately onto power grids. 

Alfred Johnson: This is Critical Capital. Welcome. I'm Alfred Johnson, the CEO of Crux. We're the capital platform for the clean economy. This show explores the intersection of technology, capital markets, and policy for clean and critical infrastructure.

Our guest this week has been in the middle of all of it. Dr. Varun Sivaram is the founder and CEO of Emerald AI. He's a physicist, a former executive for two different clean energy developers, and he served as a senior advisor at the US Department of State working on global climate and energy policy. Varun calls himself an AI maximalist. He believes AI is one of the most important inventions in human history and that winning the AI race is the defining geopolitical challenge of today. But he also understands the physical constraints standing in the way. His answer is Emerald AI, a company built on the premise that the solution is hiding inside the AI factories causing the energy bottleneck in the first place.

So I brought Varun on the show to talk about what data center flexibility looks like in practice, the global stakes, shifting public opinion on AI, and pathways for commercializing the technology.

Varun Sivaram, welcome to Critical Capital. It is an absolute thrill to have you on the show. You and I have been talking about these issues for a very long time, and one of the things that I am struck about you is that you bring this really unique mix of experience that crosses energy, international diplomacy, economics and finance, AI, and you've talked about this historic collision that is happening between these two multi-trillion-dollar systems. On one hand, we have the grid, the most complicated machine humans have ever made. On the other hand, we have AI, one of the most promising technologies we've ever encountered. What history, assumptions, and realities have shaped the grid that we have today?

Varun Sivaram: Thanks, Alfred, for having me on, and congratulations on the new podcast. This is the heart of why there is both this technological and cultural collision. You asked about the history of the grid. Our grid is what I consider — and many consider — to be the technological marvel of the 20th century. It enabled the US economy to expand at breakneck pace, and across the country, populations expanded and economies dynamically expanded. Then for the last couple decades, load growth paused, loads kind of stopped growing — it was flat — and very recently, loads have grown at a breakneck pace yet again because of electrification, population growth, economic growth, and most importantly, as we're talking about today, data centers.

Regulations can take a decade or more to mature, but in the AI space, we wanna deploy gigawatts, tens of gigawatts, tomorrow. In the United States, 50 gigawatts wants to come online over the next three years, between now and 2028, but only 25 gigawatts can actually get built. So that's the collision we're talking about.

You used the words multi-trillion. It is a multi-trillion-dollar collision because there's trillions of dollars of demand for AI data centers. And if we can't meet that demand, from a policy perspective, we lose American national competitiveness. And if we do meet that demand but in the wrong way, then we just raise rates for everybody, and that's an affordability crisis.

Alfred Johnson: Let's talk about AI for a second. I am sometimes reminded of the beginning days of COVID when everybody became an amateur virologist. I am now watching a lot of people talk about the demand profile for data centers associated with AI, but I don't think that there is sufficient nuance in that conversation.

Can you just talk about what the actual profile of demand is for AI data centers? Start at the national level and then talk about the specific instances of AI data centers and what they require.

Varun Sivaram: AI data centers are a new kind of load growth compared to historically. So you talk about air conditioners, in the second half of the 20th century, air conditioners were a massive, multi-10-gigawatt load-growth phenomenon, but they were distributed across the country, and each of these air conditioners typically was a small load. Data centers are different. Data centers are large and heavily concentrated loads, and so they create this geographic heterogeneity. They're more like chocolate chips in the cookie rather than a very equally spread-out cookie batter. That's a terrible analogy. I made it up right now. It's probably not very good.

Alfred Johnson: I like it. I like it. Let's stay with it. I'm gonna hold you accountable to staying with the cookie analogy for the rest of the show. 

Varun Sivaram: Well, so let's talk about these chocolate chips, right? Because each data center, historically, back when data centers represented less than 5% of American electricity consumption, each of these data centers comprised these racks of servers, and the racks would pull 5, 10 kilowatts. Now each rack of servers with NVIDIA GPUs, the latest and greatest, are pulling hundreds of kilowatts and headed toward a megawatt for every rack. So you're seeing multiple orders of magnitude increased energy density. You're talking about gigawatt-scale campuses on a single node of your power system. So that extreme density is one difference in terms of AI's load growth compared to previous, more homogeneous load growth.

The second is the profile. AI data centers can be extremely volatile, and they're extremely sensitive. You've seen examples, for example, in Northern Virginia, where multiple data centers will all just trip off at the same time. They've caused NERC, the North American reliability watchdog, to say, "Look, this is potentially a reliability catastrophe. You could have a cascading blackout if too many of these super sensitive AI data centers join the grid." And each of them not only is ramping up and down in the span of seconds, or even milliseconds — because training can be a highly volatile use case — but, in addition, they can all induce each other to just drop off the grid if they detect a mild harmonic or spike in power quality, a fluctuation.

All of this is to say, this is not your grandmother's load growth story, right? These chocolate chips are very idiosyncratic, and the grid, which has been stable over more than a century, needs to adapt at light speed to these AI factories, as Jensen Huang calls them. We believe that AI factories can also be adaptive, responsive, flexible assets on the power grid to help the power grid out instead of the relationship only running one way where the power grid itself has to accommodate these extraordinarily different and demanding loads.

Alfred Johnson: And Varun, what could that look like in practice? I know that you recently ran this experiment in Phoenix where you were able to prove that flexibility can achieve much better outcomes. So what could flexibility look like if it existed across the grid?

Varun Sivaram: So you mentioned the Phoenix demonstration that we did. We've actually done five demonstrations at commercial data centers around the world, starting with Phoenix with Oracle; Chicago; Virginia; London in the United Kingdom with National Grid; Hillsborough, Oregon. Across these demonstrations, we've showcased that flexibility is possible, where these AI factories act as real assets.

And you asked, Alfred, what's the potential here? The potential here is in the United States, as Tyler Norris at Duke University wrote, there's 100 gigawatts or more of latent power capacity just sitting on our existing grids that we could use to power data centers. Around the world, that number more than doubles. There's plenty of existing stranded grid capacity that we just can't use today because AI factories or data centers are not able to be flexible yet, or haven't been given the right incentives to do so.

But if they can be flexible, just modestly flexible a small percentage of the year, we can fit more data centers immediately onto power grids, so that enhances America's AI competitiveness. We can keep rates low because we better utilize our existing power system instead of triggering these exorbitant upgrades that cost everybody in the community additional rates, and we can keep the grid more stable and reliable because the AI factories can respond to fluctuations and needs of the power grid.

That's like three birds with one stone. Alfred, you and I served in policy. You never get a three for one. You barely even get a two for one. Typically, there's a tradeoff. There's almost no tradeoff here, which is a very rare thing, and the reason that I wanted to jump on this idea and found a company is you can get three amazing things for very little cost or tradeoff, and that's why this technology works so well.

Alfred Johnson: Yes, and so we have these multi-trillion-dollar systems colliding, right? We've got the energy system, we have AI, and that all exists within the context of a complex world, right? People consume power at different times. The sun shines brighter, the wind blows faster. Sometimes it's hot, sometimes it's cold.

Talk about temporal flexibility and spatial flexibility. Why is it important that we be able to control the loads from a time perspective and then also be able to distribute those loads across the country?

Varun Sivaram: So you remember those chocolate chips, and I told you how scary the chocolate chips are.

Alfred Johnson: Yes. I'm getting hungry.

Varun Sivaram: They're really different from previous load-growth episodes. AI factories act in a different way than any other power load acts. An AI factory is electronically controllable, which means, literally from a laptop, we are able to modulate the consumption of these GPUs. They don't have these massive operational drawbacks of other loads. If you tried to stop a factory from using energy, you might stop producing cars, there might be some ramp down, ramp up time, et cetera, machinery gets harmed. That's not necessarily true if you do it gracefully with software for an AI factory.

And most interestingly of all, AI factories can move their workloads from one location to another at the speed of light, which you can't do with any other user of energy.

You asked about temporal and spatial flexibility. Temporal flexibility means that you might have some jobs running in an AI factory — maybe it's fine tuning some models — that you can slow down or pause because the customer finds that acceptable. They're like, "I really don't need this model fine tuned right this very minute. I can wait an hour or a few hours." 

Spatial flexibility is where you can move AI workloads from one location to another. Serving inference, you may be speaking with a chatbot and the milliseconds that it takes to move from Virginia to Chicago, as we demonstrated with Oracle at their data centers, the milliseconds that it takes to move are tolerable for a user who expects responses on the scale of seconds, if that. I just got a ChatGPT query back right before this podcast. It took three minutes of thinking. I would be okay if it took 500 milliseconds more.

My point here is there is inherent flexibility that we can take advantage of in these AI factories and the network of which they're a part and taking advantage of that flexibility and putting it together with other kinds of flexibility, such as what I call resource flexibility. If you have an onsite battery that you can discharge if it's at an appropriate state of charge. You put all these together and these AI factories can become living, breathing, responsive entities. 

Alfred Johnson: You described a win-win-win earlier and how, if we're able to be more flexible, we can achieve much better outcomes. Do you see a win-win-win geopolitically? Is it possible that AI will be this really positive force that can play out in better geopolitical and global cooperation, or are we just in this present reality that is inherently more competitive? And how will that play out as the US and China both invest as aggressively as we are in AI infrastructure and energy?

Varun Sivaram: Alfred, you and I are both Council on Foreign Relations members, and so in some sense we are both committed to finding innovative ways for countries to collaborate and cooperate on the global stage. But make no mistake, I was an international relations double major along with physics i my undergrad. I am a realist through and through. That means I come from the standpoint of: nations exist in a state of anarchy competing against one another, and it is brutal geopolitical competition. On the margins, it would be great to find ways for them to cooperate, but I am under no illusions that there is a governance superstructure that constrains nations, least of all the United States.

Under this backdrop of brutal anarchic competition, I think that winning the AI race is absolutely critical for the United States in this generational competition with China first and foremost, and with the series of other adversaries and competitors. China has 400 gigawatts of spare power capacity by 2030 that it can deploy toward AI data centers and AI factories. Energy is not its bottleneck. Chips are China's bottleneck. The position is reversed in the United States. Chips are not our bottleneck; we are not going to be fab limited in 2030, but we will be energy limited on current course and speed. That's why unlocking the energy capacity for AI deployment in the United States is the most critical question for geopolitical competition.

There's so many other issues, Alfred, that you and I could talk about. We could talk about the Strait of Hormuz. We could talk about drone warfare in Europe, et cetera, et cetera. The number one determinant of America's geopolitical success in the 21st century, in my opinion, is going to be how many data centers we can build in America. And that's a provocative claim, right? Others say, do we really need them in this country? Might it not be possible to have a network of AI data centers in allied countries, which I actually think is a great idea. But the ability to build AI infrastructure in the United States, I think, is a crucial determinant of our ability to stay ahead in the frontier model race, our ability to make our economy a low-latency, AI-driven economy and our ability to develop the AI services and goods that we can then use to make our economy the most technologically advanced in the world and export those services around the world in order to build lasting economic prosperity. Under the backdrop of a brutal anarchic military and geopolitical competition, there's no question — we gotta build them at home and unlocking power is the key to do so.

Alfred Johnson: Okay, so we are existing in this state of global anarchy. There are chocolate chip cookies that are on the table and can be accessed by various countries around the world. We need to build urgently: energy infrastructure, AI — we need to make that more flexible. What's holding us back?

Varun Sivaram: We started here, Alfred, with this historic collision between two systems that really weren't designed to work at the same speed or in the same way. The grid is historically extremely conservative and risk averse. Electric utilities will offer you a connection, for example, but before they do so, they'll run a study to make sure that in the absolute worst case over the next 10 years, they'll still be able to serve you at that worst moment when all the air conditioners are running in Phoenix and you, data center, are requiring your absolute peak capacity. And if they can't serve you in that study, they're gonna say, "Wait in line for 10 years while I build out my grid to serve you at that peak." That's kind of one cultural approach.

And the other cultural approach is, "Move really fast, sometimes break things." Those are the chocolate chips of the cookie that the data center's coming in and saying, "Connect me right now. I'm ready to go. And by the way, if you don't connect me, I'm ready to go off grid. I'm going to build my own onsite infrastructure. I'll even go into space."

We're at a critical juncture right now because if we make the wrong choice and we do in fact induce the data centers, the AI infrastructure, to go off grid, it's bad for everybody. This is counterintuitive. You might say, "Wait a minute, you're not connecting to my local power grid, so my local community is not gonna see rising rates because the AI data center just went off and did its own thing. That's great for us!" It's actually terrible. Every local community in America will suffer if AI data centers decide not to connect to their local power grid, because if they do that, they're depriving the local power grid, the local community, of the revenue that comes from these large and lucrative customers. And over the long run, if most of American GDP is driven by AI, which is what I believe, and most of that GDP goes off of the grid, that public grid is going to be left to serve its local community and be overburdened without the benefit of its best customers, which will have left.

It's also bad, by the way, for those customers, the AI data centers. These AI factories now have to build extremely redundant and expensive microgrids. These microgrids with onsite resources have to supply them power — and high-quality power — 24/7 around the year, and that just costs a lot of money, it's technically complex, and many folks can't do it as effectively as the best engineering invention of the 20th century — the power grid.

It's just bad for everybody because the cost of tokens of artificial intelligence goes up and America becomes less competitive 'cause they're building these expensive, redundant systems. And the public grid is burdened. By the way, if you build these islands off grid, you actually raise costs for everybody because they monopolize or increase the demand for rare equipment — natural gas generators, switch gear — and increasing the cost of these equipment for everybody happens if you build redundant, inefficient systems.

So I really wanna make sure that we connect AI data centers, AI factories, to the public power grids. I wanna make sure that utilities can serve these new, lucrative customers. And I wanna make sure we do it in such a way that everyone's rates go down because the data centers are willing to be at least a little bit flexible as grid allies.

So the biggest thing holding us back is the inertia of history, and it's time now to do things a different way.

Alfred Johnson: Everything that you're saying is so practical, and it's actually optimistic — the idea that we will be able to integrate these multi-trillion-dollar, colliding systems and create a better reality through flexibility. We also exist within a complicated political reality, and the politics around this are changing pretty rapidly. I was looking at a new Quinnipiac poll that came out recently that has 55% of Americans thinking that AI will do more harm than good, with only 34% saying it'll do more good than harm.

How do we change that story? And what do you think the practical constraints of politics will be as we try to create a new and better system?

Varun Sivaram: This is the perfect question to ask. I think it's critical we start changing that narrative right now. I know we're recording this podcast weeks before it actually launches, but this week I am in Washington, DC with our partners at Silicon Valley Power. Silicon Valley Power, heart of Silicon Valley, operates a system with a large number of data centers from every brand imaginable. And they've gone ahead and made this pioneering announcement with us and with NVIDIA that we are going to launch a program to unlock capacity for flexible data centers where everyone gets what they want. The data centers get more capacity faster. The utility customers, local communities, get affordable rates. We get a reliable system, all because of flexibility. We're going to meet with all of the commissioners of the Federal Energy Regulatory Commission, we're meeting with the White House later today. The goal is to explain that there is a win-win-win to be had here, and there's at least one utility that stepped up to the plate and said, "We are going to pioneer this particular approach."

NVIDIA, for its credit, is the leader of this overall ecosystem. They call it the DSX ecosystem. It's their reference architecture for AI factories. And they've stepped forward. Jensen Huang has been public about it. He said, "Look, if the power grids, if utilities around the country are willing to offer us what we'll call imperfect power —we know that they have enough power 99% of the time, but in rare circumstances, they need us to ramp down our consumption. They need us to be flexible. We should be willing to take that arrangement."

And so now, you know to combat the polling, you mentioned from Quinnipiac, if we can prove with real use cases, real case studies — "Hey, here's a real site, an NVIDIA data center in Silicon Valley, that was able to increase its capacity, pay more dollars for megawatt hours in that local community without triggering expensive upgrades that everybody else had to pay for," and "Oh, here in Virginia, we're gonna launch the world's first power-flexible AI factory later in 2026 with NVIDIA and Digital Realty called the Aurora facility, and that is going to prove to Dominion on the East Coast and PJM that these facilities are really possible" — then we can create a track record, a quantitative track record. Here's how AI factories perform, and here's what they can do for your affordability.

We're doing this around the world as well. In London, we did a demo with National Grid to showcase that when there's a lightning strike, an AI factory can respond within seconds to stabilize the grid. When there's a tea kettle epidemic, or whatever you call a phenomenon where lots and lots of people turn on their tea kettles in the middle of a soccer game, the AI factory is able to reduce its consumption and help to stabilize the grid, and so we can fit many more of these AI factories on the UK grid. Creating this fact pattern all over the world enables us to combat the narrative and say, "No, no, AI factories can absolutely be heroes."

And the final puzzle piece here, Alfred, because yes, we live in a complicated world driven by policy in many cases. We do need regulators, governors, and utilities to take that first step and say, "Okay, we believe this capability is real. We will offer you a deal. If you're willing to be flexible, if you're willing to support our system and our affordability and our reliability, we will in fact give you faster power access and larger power connections." That's the trade, the carrot that really matters for the AI industry, and I'm confident that trade will be taken. We just need the inaugural offers, and that's why I'm so excited about Silicon Valley Power, for example. I'm excited about what ERCOT, Texas is going to roll out in a couple months to make flexible data centers able to connect much faster to the power system. That's where I need to see the dominoes fall.

Alfred Johnson: And as those dominoes start to fall, I imagine you'll be having a lot more conversations with utilities across the country, and utilities are not historically seen as being very fast moving in adoption of new technologies. When you bring a case study like the Silicon Valley Power case study to a utility in another part of the country that has less experience with this, what do those conversations go like? What's it like to be in the room?

Varun Sivaram: I've been hearing many of the same thoughts from utilities. And I actually think they are very far advanced in their thinking and really just need a little bit of validation. So utilities will kind of naturally tell you, "We don't think it's a great idea for data centers to go off grid and never connect to the system. That drives up the supply cost of all these components that we need for the public grid, and that deprives us of very lucrative revenue that we could enable smearing out to the rest of our rate base to keep costs affordable for communities."

Utilities across the country also recognize that their systems are underutilized most of the time, and therefore, power plants aren't running at the highest max capacity they could, and power lines have plenty of spare capacity available almost all of the year. And so they recognize that flexibility can really enable them to use their systems better, bring on more customers, better serve their rate base, et cetera.

I just think that, as we create some proof points — someone had to go first, and I'm delighted Silicon Valley decided to go first. I'm getting notes from utilities across the country saying, "How can we be fast followers?" So that record of examples showing that, "Hey, this worked, someone took the risk and did something innovative," enables other utilities to step forward.

It also requires political leadership. Look, every governor in the country wants the economic development and tax boom of data centers without the rate-affordability degradation that data centers could potentially bring. To be clear, I don't think data centers have raised costs historically, but they very well could in the future if they continue to trigger grid upgrades. And so politicals, right, governors, the White House, need to step forward and say it's critical that across the country we bring in these grid-friendly data centers. We give 'em a real incentive to connect faster, we encourage utilities and regulators. There are processes underway at the state level and at the federal level to do so. We just have to continue all of this. And if we do, the rewards are just massive.

Alfred Johnson: Yeah, so again, you've talked about how collaboration across all these different kinds of parties can drive better outcomes for everybody. You got into this originally, into this space — you were an engineer, you were the CTO of one of India's largest solar companies, and then you worked in innovation at Orsted after serving as a lead climate negotiator.

Let's talk about the climate piece of it. Is it possible to have a win-win there, too? We're in this moment of energy addition. Are we able to use this moment of need and growth in AI to drive better and different climate realities?

Varun Sivaram: Absolutely. Look, this is a topic you and I care deeply about, and this is the fourth bird that you can kill with the same stone. I really hate the imagery of the birds. So instead I'll say, this is where the chocolate chips become green. And the reason for this, Alfred, is if AI factories can be truly grid flexible, what that means is they're able to modulate their power consumption on demand by slowing down some computations or moving computations across the country. They become a demand-side resource to integrate variable, intermittent supply-side resources. Solar and wind are unreliable, intermittent, and variable, but they're also the cheapest and fastest resources to build, on land at least. And if we are able to make our demand side a little more responsive to the intermittencies of the supply side, all connected, mediated by a grid that smooths out and averages some of the harshest fluctuations, then it's possible with 25% or more of American load becoming flexible, that you could have a power system that is substantially more supplied by clean energy.

Now, of course, other clean energy sources, from geothermal to nuclear, that are firm and/or dispatchable are just wonderful to have on the system. So I personally think that flexible AI factories can not only connect to the grid faster and better utilize it, lower rates make the system more reliable and stable. The fourth benefit is they can bring on more clean energy. And it's the reason — Alfred, you and I both served together — I care deeply about climate and my hope is, over the long run, we'll be able to achieve all four of these benefits at the same time.

Alfred Johnson: Yeah, so we see a lot of this out in the field. We've recently done a number of very large deals where the offtake on a large solar field has been a PPA to a hyperscaler. Tell me what you're seeing out there when you go see data centers around the country. What does the actual power mix look like? We hear so much about the backlog in gas turbines and the need for firm four-nines power. What do you see in reality when you go out there and you see these facilities and you talk to them about how they're planning their power?

Varun Sivaram: So there's two paradigms here. Paradigm number one is you connect to the grid, in which case the source of the power is abstracted from you. You may have some climate commitments, but really all you really care about is getting firm power from the grid, right? And the other paradigm is where you're so fed up with the grid that you wanna put some behind-the-meter generation, perhaps to bridge you over the next couple years while you wait for a grid connection.

In the first one, we wanna make it possible for the grid to serve you because you're willing to be a little bit flexible. But on the supply side for the grid, whether it's a solar or wind generator that is part of the mix, that additional variability from renewable sources connecting front of the meter requires at least a little bit of demand flexibility for the market map to work out. It's mediated by a grid and a market, but the point remains, which is: if you just claim that you are bilaterally contracted through a PPA with a renewable source, you're not actually powering your data center operations with clean sources, right? It actually takes real operational flexibility.

On the other side, if you have a behind-the-meter setup — and by the way, Invenergy, NVIDIA, and Emerald just made a big announcement about how we're all working together, and there'll be cases for Invenergy where Invenergy is putting a large solar, wind, and battery installation together and seeking to power a data center behind the meter, either as a bridge or going forward, that that particular configuration works best if the load has at least some flexibility. Because there will be that dunkelflaute situation where there's two weeks where you have overcast skies and still winds and your battery can't retain so much charge, and you'll want a little bit of flexibility in order to make the behind-the-meter supply match up with the load while minimizing the redundant reciprocating engines that you might have on site, et cetera.

So there are many different ways that this works. I can only imagine that Crux has the best intel on all of the sites coming up and serving this load. And my hope is these suppliers recognize that if loads are flexible, it's a better deal for the loads, it's a better deal for the supplier, and it happens to be a better deal for the grid.

Alfred Johnson: Yeah. We just exist in this world where everything is changing all the time, right, and the reality of the power system is that we're seeing much more wind, solar, and batteries as additions to the grid for the foreseeable future. And so there's this additionally complex reality of, how do we make those variable sources work for a power-demand profile that is incredibly voracious in its appetite? And I think the way that you think about flexibility and using AI to control AI is a very optimistic take on what that could be.

Varun, the company is, I think, not even two years old. You have just been named one of Time’s 100 most influential companies alongside people like Nike and Anthropic, some of the most iconic companies of a generation. How did that happen? Tell me how you see this moment for the company.

Varun Sivaram: Thank you. It's a true honor, owed entirely to our fantastic team. Our chief scientist spent over a decade kind of inventing and developing this field of flexible data centers. Our head of engineering came from Amazon. A lot of folks left very lucrative careers at massive companies and hyperscalers to come and join this mission-driven team.

Our culture has been about, “How do we change the world?” and I'm just delighted that there's some recognition that just in the last 17 months that we've been around, I think there has been a sea change. Flexibility, when we founded this, when you and I were sitting in Compass Coffee, this was a crackpot idea, right? Data centers were never gonna be flexible. This is a terrible idea. And just over those last 17 months since we were founded, regulators across the country, at the Federal Energy Regulatory Commission, in the government of the United Kingdom, and elsewhere have all said, "You know what? This would actually be a really good idea. We would love for flexible data centers to come onto our grids." And NVIDIA, the world's largest company by market cap, has come out aggressively saying, "This is the way for us to deploy trillions of dollars more AI infrastructure much faster and turbocharge the AI revolution."

Look, we're a tiny part of it. We're just lucky and honored to be alongside such massive, massive companies that have so much to give the world. But if there's one thing that we are trying to do here, it's to change a fundamental narrative. A narrative that AI is a villain, AI comes to your community, AI raises your prices, you don't want AI anywhere near you. No, no. AI is critical for America's future geopolitical strength and economic success. AI can be a hero and not a villain 'cause it can actually lower rates and make grids stable. And the technology to make AI this grid-friendly neighbor actually is AI itself.

As you know, Alfred, the hard work's ahead. You and I need to keep our heads down and just execute.

Alfred Johnson: Varun, what a wide-ranging conversation. We covered everything from the history of the grid to the present realities of AI to the complicated geopolitical economic world that we are operating in, even touched on evolving domestic politics and public opinion around this stuff. It was a true pleasure to have you on. Thank you for your time.

Varun Sivaram: Thank you so much. 

Alfred Johnson: Varun Sivaram is the CEO and founder of Emerald AI, a technology company that enables data centers to flexibly consume electricity and bolster the power grid.

If you enjoyed our conversation, consider subscribing to us on Apple, Spotify, or wherever you get your podcasts. Critical Capital is a co-production of Crux and Latitude Studios. Our production team includes John Sheehan, Jenna Herzog, Stephen Lacy, Anne Bailey, and Sean Marquand. The show is mixed by Matthew Filler. Additional production by Emily Hughes and the excellent team at Crux, the capital platform for the clean economy. I'm Alfred Johnson. Thanks for listening