How AI Data Centers Can Go From Villain to Hero with Varun Sivaram

“Everyone hates data centers.”

That was the subject line on the email newsletter from Heatmap Daily the day before I sat down with Dr. Varun Sivaram, co-founder and CEO of Emerald AI. Communities see huge new loads coming onto the grid, hear about billions in new infrastructure, and worry that their bills will go up.

It doesn’t have to work that way.

Varun argues there are two paths. On the villain path, AI data centers drive up power bills and increase the likelihood of outages. On the hero path, they become flexible grid assets that help us use existing capacity better, absorb much of the cost of new grid infrastructure, and help residential and small commercial customers pay for distributed batteries, heat pumps, and more.

Texas and ERCOT are at that fork in the road.

Two futures for AI data centers

Varun calls this a “critical juncture.” If ratepayers have to pay more and grid reliability takes a hit, communities start pushing projects away and the U.S. falls behind in the global AI race

The alternative is the hero path, where data centers show up as flexible partners:

Data centers in this hero path are going to contribute to grid reliability and help us to avoid rolling blackouts. I think we can get there, but we’re not on that path right now and folks are right to worry. And this is the moment where we switch from the villain to the hero.

Texas has a chance to innovate — both technologically and with policy. Regulatory innovation is as important as technological innovation — maybe more so.

Turning AI load into flexibility

Emerald AI is a software layer that makes AI workloads flexible. Varun breaks it down into four kinds of flexibility:

* Temporal. Once you know what can move, you can shift it in time. Training a big model at 6 p.m., when ERCOT is tight, is very different than running it at 2 a.m. when prices are low and resources are abundant.

* Spatial. Many jobs can move across locations. If a Texas node is stressed and another region is fine, traffic can be shifted without changing the user experience.

* Resource. Some tasks truly need instant answers, others can wait minutes, hours, or days. Emerald deploys and optimizes onsite resources when necessary.

* Adjacent. Data centers can purchase flexibility — putting money into the pockets of residential and small commercial customers — from distributed batteries, HVAC systems, and other controllable equipment.

Put together, these layers make a data center behave less like a rigid block of demand and more like a flexible grid asset when conditions require it.

The Energy Capital Podcast is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

ERCOT’s stakes and the Texas choice

Varun shared a conversation with ERCOT CEO Pablo Vegas. Vegas said he did not just want a tool that jumps in during emergencies. He wanted something that keeps the grid from getting to an emergency. Don’t want for the flashing red lights; have data centers contribute flexibility when the lights are flashing yellow.

That is the heart of the hero path.

ERCOT was already dealing with intense load growth from industrial projects, crypto-miners, traditional data centers, increasing population, hotter temperatures, and now AI data centers. Texans will not accept anything less than high reliability and lower bills. If the PUC and ERCOT treat AI as inflexible, we will need to build a lot more capacity and infrastructure than we might otherwise need.

If we require and reward flexibility, we can serve more load at lower cost, then add new infrastructure when truly needed.

Final Thoughts

The hardware and software inside AI data centers means they are already some of the most controllable loads connected to the system. With the right tools, incentives, and market structures, AI factories can act as shock absorbers instead of stress multipliers.

Texas leads on gas. Texas leads on wind. Texas leads on solar and storage. We can also lead on making AI an ally to the grid, not a villain. That will take work but it is possible. It’s a choice we can make.

If you enjoyed this podcast, please share it with a friend or colleague or family member or neighbor. The more Texans engage with these decisions, the better chance we have for a grid that is reliable, affordable, and cleaner for everyone.

Energy Capital is produced by ClarityForge Studios.

Timestamps:

* 00:00 – Intro, Varun bio, Emerald AI

* 02:15 – The villain and hero paths for AI data centers

* 05:30 – Phoenix pilot as a tangible example of the hero path

* 09:00 – California simulation of 2020 outages

* 10:00 – Possibility of doing a pilot in ERCOT, Pablo Vegas’s comments

* 12:00 – What exactly does EmeraldAI do?

* 14:00 – Breaking down four flexibilities: temporal, spatial, onsite resource flexibility, adjacent

* 20:00 – Emerald AI’s focus is on onsite flexibility

* 24:00 – Real-world stress test results

* 27:00 – What excites Varun about AI

* 32:00 – How AI can help lower power bills: the central tenet of the hero path

* 36:00 – Why ERCOT is potentially the global model for speed to power

* 40:00 – Connect-and-manage for loads

* 43:00 – A reference design for AI factories from a pilot in Virginia

* 46:30 – The hero and villain path for AI and emissions

* 49:00 – Optimizing the system to buy time until nuclear, geothermal, etc. are ready

* 51:30 – Getting a win-win-win: on affordability, on AI innovation, and sustainable, reliable systems

* 52:30 – Final thoughts: the Emerald AI team

Resources:

Host, Guest & Company• Varun Sivaram - Linkedin • EmeraldAI - LinkedIn• Doug Lewin - LinkedIn, Twitter(X), Bluesky, & YouTube

Company News• Sharing Our Seed Extension - Press Release• National Grid and Emerald AI announce strategic partnership - Press Release• How AI Factories Can Help Relieve Grid Stress - Press Release

Books & Articles •The Worlds I See: Curiosity, Exploration, and Discovery at the Dawn of AI by Dr. Fei-Fei Li•The Country’s Biggest Grid Has a Plan to Manage Data Centers’ Power Use. Everyone Hates It. - Heatmap News •The mechanics of data center flexibility - Catalyst Podcast (Latitude Media) •How the world’s first flexible AI factory will work in tandem with the grid by Arushi Sharma Frank in Latitude MediaPolicy & Reports • Report on disorganized integration of data centers - Texas Reliability Entity • 2025 State of Reliability - NERC• The Worlds I See - Dr. Fei-Fei Li• Arushi Sharma Frank’s ERCOT Planning Guide Revision Request• Retail Electricity Price and Cost Trends: 2024 - Lawrence Berkeley Labs• Rethinking Load Growth - Tyler Norris and Duke University• ANOPR on Large Load Interconnection - FERC• Emerald AI: presentation to ERCOT Large Flexible Load Task Force • PGGR 135: Large Load Interconnection Queue Process Revision

Related Podcasts by Doug• How Data centers Strengthen the Grid - Astrid Atkinson

• Texas’ Load Growth Challenges – And Opportunities, with Arushi Sharma Frank

• How Load Flexibility Could Unlock Energy Abundance with Tyler Norris

Related Substack Posts by Doug• AI Data Centers Aren’t Causing Higher Prices • Demand Side Resources Could Enable Load Growth• Can AI Data Centers Lower Costs for Residential Consumers?

Transcript:

Doug Lewin (00:05.154)

Welcome to the Energy Capital Podcast. I’m your host, Doug Lewin. My guest this week is Dr. Varun Sivaram. Varun is one of the most interesting guests I’ve had in the three years now I’ve been doing podcasts, both the Energy Capital Podcast and going back to the Texas Power Podcast. He is the founder of Emerald AI, a company which is transforming energy-intensive data centers into grid assets and grid allies. We talked about all the different ways that data centers, if integrated right... The Texas reliability entity brought this up in a report: the possible disorganized integration of data centers into the grid is one of the biggest reliability risks. I would argue, and clearly Dr. Sivaram argues, the counter is true as well. The organized integration of data centers can actually make grids more reliable and spread costs out to more customers.

We got into all of that. Just a couple of notes on Varun: He was formerly the chief strategy and innovation officer at Ørsted. He was the chief technology officer of India’s largest clean energy company, ReNew Power. He was a diplomat at the US State Department. He is currently a senior fellow at the Council on Foreign Relations. He was named as one of Time Magazine’s Time 100 Next for the next 100 most influential people in the world. MIT Technology Review named him one of the top 35 innovators under 35. You get the idea. He also has a PhD in condensed matter physics from Oxford. This bio is kind of ridiculous. Clearly one of the smartest people out there in this space, and this company, Emerald AI, is really doing some super innovative things with some really high-level partners, including NVIDIA and others. I think you’ll enjoy this conversation as much as I did. Please leave us a five-star review wherever you listen.

And most importantly, if you are not already a subscriber at douglewin.com, please go there and become a subscriber today. Your support for the podcast really makes it possible. And with that, here is my conversation with Dr. Varun Sivaram.

Varun Sivaram, welcome to the Energy Capital Podcast.

Varun Sivaram (02:19.256)

Thanks so much for having me. It’s an honor.

Doug Lewin (02:21.514)

Hey, it’s great to talk with you. I have been reading article after article about Emerald AI. I saw your presentation to the large load task force a couple of months ago in Texas and have been meaning to do this for a while. So thanks so much for taking the time. We’re going to obviously talk about Emerald AI. We’re going to talk about Texas and data center growth. We’re going to talk about all of these things, but I just want to start from a very high level, Varun. We are recording here on November 7th. Yesterday, Heatmap News had a very provocative headline: “Everyone hates data centers.” I don’t know that that’s actually true, but I know what they’re trying to say. There certainly is a lot of opposition to data centers right now. You are doing a lot of work, obviously, around data centers, data center flexibility, just from a grid perspective, thinking about affordability, reliability, lower emissions—all these aspects of data centers. Why should people not hate data centers?

Varun Sivaram (03:19.928)

Well, look, Doug, I think we’re at a critical juncture, and that juncture is between what I consider to be the villain path for data centers and the hero path for data centers. And I don’t think either one of them is preordained. I think that folks may be right to say that they’re worried about the impact of data centers in their community the way things are headed today, right? The average annual household power bill in Columbus, Ohio rose by $240 directly attributable to data centers in 2025. And you’ve seen NERC studies, for example, and other reports showing that the advent of AI data centers could cause grid reliability issues. So there is a scary villain path that I’m worried about in which data centers come to town and communities don’t want them. They raise rates, they destabilize grids, and as a result, you just have fewer data centers getting built. I think the villain path is not just bad for the AI industry. The villain path is very bad for America because America needs AI infrastructure and AI data centers to compete in the 21st century in the most important economic sector we’ve ever seen. And data centers can provide economic development, and they can help us to compete with China. So we absolutely need a lot more AI data centers.

The hero path is the one that I’m obsessed with getting us onto because I actually think AI data centers, far from being the thing that undermines the grid, can actually be the asset that saves the grid. And in that hero path, if we get on it, data centers come to town, they actually lower your rates, or at least they arrest the increase in rates because they’re more efficiently utilizing your existing system. We can connect far more data centers much more quickly to existing power systems and defer the massively expensive overbuild of infrastructure and more prudently expand our grid and expand our generation. And by the way, data centers in this hero path are going to contribute to grid reliability and help us to avoid rolling blackouts. I think we can get there, but we’re not on that path right now and folks are right to worry. And this is the moment where we switch from the villain to the hero.

Doug Lewin (05:25.878)

Yeah, I definitely want to talk a lot about both paths because I think we’re seeing elements of both of them. But obviously, to me anyway, the hero path is much more interesting. Maybe this is my sunny outlook on life or something, but I think that there really is an opportunity here when you see the scale of investment that you see, and we know that the grid has been underinvested in. This is an opportunity to bring a lot of investment to the grid.

Where I think I want to go next is I do want to ask you about Emerald AI. And I think the way I kind of want to bring that in here is this test you guys did in Phoenix recently. Can you talk a little bit about what you did there and connect that, obviously, to the hero path?

Varun Sivaram (06:10.528)

Absolutely. So Phoenix, Arizona: deployment one of now four Emerald AI deployments. In that first deployment, we went to Phoenix with a range of very credible and authoritative partners like EPRI, the utility association that runs DCFlex, Oracle, in whose data center we were operating, NVIDIA, who’s both our investor and partner in this demonstration, as well as the local utility Salt River Project. The goal was to prove that on a grid that faces summertime strain—let’s say you’ve got a peak moment sometime in the summer where a million air conditioners are straining in Phoenix, Arizona—that a data center can actually flex its power consumption. It can reduce its power consumption at that very moment to provide badly needed relief. And in doing so, that data center demonstrates the kind of behavior where you say, “This is one of those grid-friendly data centers that if one of these comes to interconnect to my grid, I would love to have this hero of a data center connect—not in seven years, requiring me to build out transmission lines and power plants, but right now in seven months because it can provide badly needed relief when I need it and it’s not going to raise my peak load unsustainably.”

And so we went out to Phoenix and we worked with our partners, as well as the chief scientist of Databricks, to design a representative set of customer workloads running on this cluster of NVIDIA GPUs—a representative set of workloads across inference, fine-tuning, training of large language models. We went ahead and said, “Is it possible that if we get a signal during the peak demand on that day from the local utility that we can then reduce the power consumption by 25% for three hours?” And those were the parameters set to us by our utility partners. The test succeeded. In fact, the test didn’t just succeed once, it succeeded many times. And we’re really pleased—I think this is the first time I’m sharing this publicly, Doug—we’re very pleased that those results have now been formally peer-reviewed and accepted for publication at one of the world’s top scientific journals, Nature Energy.

So we’re delighted that this is kind of an inaugural first demonstration of AI computational flexibility where the data center itself is changing the way it operates in a way that is supporting the reliability standards of the AI customers. We made sure that our AI customers and partners were happy with the performance of their AI workloads while at the same time, the grid got exactly the performance it needed—to see that reduction over a controlled ramp rate, that 25% reduction over a three-hour period, which is what the instruction to us from the grid was, and then a controlled ramp back and no snapback beyond the baseline energy consumption. That’s the kind of behavior that if you replicated across many data centers can save a grid from a blackout.

Just very briefly, I’ll say, Doug, we also simulated a real California event. A little over five years ago in August 2020, we simulated what happened in California where a 500-megawatt gas plant just tripped offline. Had we had Emerald AI deployed on data centers in that service territory, we could have avoided the rolling blackouts that ensued. And what we demonstrated in this trial—again, this was in Phoenix, but now responding to a CAISO emergency need—we demonstrated that we could first reduce the consumption by a little bit and then reduce by a further amount if that’s what the grid operator signals that we need to do. And so this kind of dynamic ability to respond to the grid’s needs as they evolve while protecting the performance of the AI workloads, the most valuable workloads in human history—that’s the dual optimization that Emerald AI enabled in this Arizona test. And it’s just the first of many deployments that we’re super excited about.

Doug Lewin (09:48.246)

Now you had, I think a few months ago, again at that large load working group, said you guys were at least considering doing some kind of a test in Texas. Is that happening or on the roadmap?

Varun Sivaram (09:58.996)

It is my deep desire to get that test up and running. You know, ever since the Arizona demo, we went ahead and did another commercial demonstration again with EPRI DCFlex, and we’ll be excited once we finish EPRI’s independent technical validation to present those results to the public. We have a test that’s been announced and that will be done in the United Kingdom, our first international expansion to London, with National Grid, the national utility there, and a large data center with the most advanced NVIDIA GPUs. And then last week we made a major announcement about our fourth deployment—happy to talk more about it—a commercial scale of nearly 100 megawatts in Virginia with NVIDIA. So there’s a lot of excitement for what’s to come.

I really would like to do this in ERCOT. And I’ll just share that one of the most impactful statements I have heard came from Pablo Vegas, the CEO of ERCOT, who shared with me, he said, “Look, I would like your technology not just to try and relieve the grid when we’re in an emergency moment—imagine, you know, all signals flashing red—but rather when they’re flashing yellow.” That’s right. When they’re flashing yellow and it looks like, you know, we might be approaching a scarcity event, that’s when our grid-friendly data centers can really help bring the grid right back to that green zone and therefore avoid ever coming into the emergency situation to begin with. It’s why I’m so enthused about this hero pathway. If you have grid-friendly AI data centers—and Doug, we should talk about why Nvidia calls them AI factories—if you have these grid-friendly assets, they help every day to keep you in balance and avoid you entering that emergency condition from which then you have to take drastic action to recover from.

Doug Lewin (11:39.542)

Yeah, so I think there’s a ton of applicability here. What I’m trying to think through, and I do want to come back to talking some about AI and why there are data factories and not data centers and all of that is really interesting. I just want to get grounded a little bit more in what the technology is. So I’m going to repeat it back and then you’re going to tell me where I got it wrong and/or expand on it. You’re certainly going to tell me where I got it wrong. I’ll get it wrong somehow.

But let me see if I can break this down. So with an AI data center or data factory or however you want to call it, you’ve got a lot of different things going on. You’ve got some inference that could be doing things like, for instance, routing calls to a 911 call center or helping an autonomous vehicle interpret that that thing moving across the street is a pedestrian and it needs to stop. So there are certain functions that you can’t shut off. Those need to run and they’ve got to run quickly. You can’t even necessarily move them to a different data center. Latency is a major issue for those kinds of applications. That’s kind of on the extreme end. On the other extreme might be a large language model that’s training over a long period of time. Could be done from anywhere, could be done almost any time. And then there’s like a whole lot in between there. And what you’ve done is developed a software that understands all of those different use cases and can, within those use cases, kind of move workloads around and even sort of maximize the efficiency of the chip performance. This part I don’t understand. This part seems like magic and maybe this is just the magic of the technology you’ve developed and maybe you don’t want to say too much about it, but it seems like from what I’ve read, you’re maximizing the efficiency of that GPU in that moment such that for those use cases where they can’t be shifted, you’re still getting the exact same output, but at less energy use, which is basically the definition of energy efficiency.

Okay, so if I got that all wrong, you could just start over and describe what you do. If I got it partly right, then you could correct the other parts I got wrong. What did I get right, what did I get wrong? Grade my paper, Varun.

Varun Sivaram (13:49.858)

Doug, you get an A. You did a great job there. I’m really impressed. We’ve got to bring you on staff here. Look, let me just go up a level just to explain the broad framework of flexibility here. I think of kind of four components of flexibility for a data center.

Component number one is what we call temporal flexibility. Within a data center, you might have, just like you said, Doug, you might have some really mission-critical time-sensitive workloads that you can’t pause or slow down. You might have some other ones that you can pause or slow down and everything in between. And temporal flexibility takes advantage of slowing or pausing certain workloads. And you can do that in many ways, and we take advantage of all of these different ways as we demonstrated in Arizona, whether it’s changing the clock frequency on the Nvidia GPUs or it’s rescheduling workloads, or it is changing the resource allocation—what’s called in the industry auto-scaling—the GPU utilization for particular workloads, et cetera, et cetera. So there are lots of different things you can do, but basically in one data center, temporarily over time, I can slow or pause to create flexibility, reduce the energy draw from the grid in time.

The second way is what I call spatial flexibility. This is a very unique trait that data centers have, but other economic users like electric vehicles don’t have. Data centers can move their workloads from one location to another at the speed of light over the fiber optic network. Again, this works for some workloads, but not other ones. It’s probably not gonna work for a large training run because there’s too much data to transfer, but it can work. You can move queries for an inference query, for example. You can move queries from one location to another so long as you had the model set up in multiple locations. So the second is spatial flexibility, where if you have a problem in Arizona, the grid is strained, you move your query over to Dallas, right?

Doug Lewin (15:39.438)

Before you go to the third one, just real quick on that. So, because I think that’s where a lot of listeners will have some familiarity and direct experience with AI. If you’ve done a ChatGPT or Gemini or Grok or whatever kind of search you prefer—all three—it will often say like on OpenAI, you know, it’ll say on ChatGPT, it’ll say “thinking,” right? And you get that little lag there, right? And sometimes you want to, you can, you can change that and optimize it to give you a real fast answer. Sometimes you want it to think more, but that thinking, you know, when you’re dealing with fiber optics, like that could go around the world multiple times in a second. Like you think 15 seconds is a long time or a short time or whatever you think it is. Like you could move that around the world a bunch of different times. So, and the reason I want to dive into that, Varun, is one of the things I hear so often in conversations with energy people about data centers is there just is no flexibility in these things. The data centers are paying so much for these GPUs. They’ve got to run 100% of the time, not even 99.999%. They just got to run all the time. And that is true for some use cases, but for some of them, and perhaps the ones that are the biggest use case, maybe you can expound on that a little bit, there is a lot of this spatial flexibility because somebody’s asking a question and 10, 15 seconds is a perfectly fine response time.

Varun Sivaram (17:01.462)

Yeah, absolutely. You make a great point, Doug. The latency or the delay that you’ll face if you move a query from Phoenix, Arizona to Dallas or San Antonio is not going to be a second. It’s going to be measured in milliseconds, right? You will not notice it. You absolutely will not notice it. And that’s important because AI workloads in many ways are different from the historical class of workloads that data centers used. And again, we should talk, Doug, about why we’re moving toward this paradigm called AI factories. Yes, a factory that’s optimized for converting electricity into tokens of artificial intelligence. Historically, data centers have done all kinds of heterogeneous or different things. Those include routing 911 calls, by the way. No one ever wants to mess with routing a 911 call. No one ever wants to mess with Doug Lewin trying to send a Venmo transaction to Varun. I know you haven’t done this yet, but no one wants to mess with—

Doug Lewin (17:51.446)

Do I owe you money? Are you trying to tell me I owe you money? Oh dear.

Varun Sivaram (17:57.848)

Sorry, inapt example. But no one wants to mess with a transaction like that, right? Going forward, the set of AI workloads, there are these new sets of flexibility parameters. If you’re fine-tuning a model, for example, you might have some temporal flexibility. It may be okay for that fine-tuning operation to pause for an hour or two. When it comes to inference, you mentioned these chatbots. Yes, the time to first token, which is the time you wait around waiting for that first response or that first word back to you from ChatGPT, that can take a second, it can take a few seconds. And so the few milliseconds of moving the query is not noticeable.

I’ll also say that although Doug, you and I interact with ChatGPT every day, that’s not the only use case for AI. There are lots of use cases. You might be a scientist, for example, and you might send a request for protein folding configurations and expect that request back in the morning or next week. Right? That is what we call a batchable request, which you might be able to pause again for an hour if the Phoenix grid is strained. So there’s a range of different ways we’ll use AI and almost all of them have some kind of temporal or spatial flexibility.

So again, we had four ways. I told you about the first two: temporal and spatial. They’re computational flexibility. The third one is what I call resource flexibility. This one’s intuitive. You might have some batteries on site and you’re able to reduce your grid draw because you’ve got fully charged batteries and for that limited amount of time that your batteries can run, you can locally power your data center or part of your data center. And the fourth kind of flexibility is what I call adjacent flexibility. Adjacent means not on site. You might have, for example, at the same transmission node, you might have neighborhoods with batteries or Nest thermostats, and you may be able to aggregate them all together and provide a little bit of flexibility to that substation, for example, and the utility might be willing to treat that as flexibility that counts toward the data center’s own flexibility.

So these are the four types. The first three are onsite. The first two are computational, temporal, spatial. The third one is onsite resource flexibility. And the fourth one is adjacent flexibility. You put all those together, and I sincerely believe we can make AI a flexible resource. And Emerald AI is the software layer that sits above and enables this flexibility. We are orchestrating the temporal and the spatial computational flexibility. We are co-optimizing that with the resource flexibility, the batteries on site, for example, so that you can best harmonize what you’re doing on the compute side with what you’re doing with the resources on site. And my sincere hope is over time, we will also harmonize all of this with adjacent flexibility and all the other resources that are offsite.

Doug Lewin (20:38.456)

So is your software already doing this or set up to do the adjacent part of it? Because that part I’m really fascinated with, because you talk about a hero path, right? The potential for data centers to pay for reductions from customers and, to be very clear, and people will get tired of me saying this, but I’m going to say it every time I talk about this: on a voluntary basis, nobody will ever be required to do this if you just want to pay a lot for electricity and you just don’t care. That is your right as an American and I will defend it. But if you would like to lower your bill by participating, like you could actually have data centers paying for thermostats and batteries, and this is what markets are all about, right? There’s a price for that. You figure out what the price is. What’s the price to add another increment of battery on site? If the price is lower to put money into the pockets of the people at the neighborhood just across the way and be a hero and save money, right? The data center... I mean, you talk about a win-win-win all around. I’m really fascinated by that piece. I’m really glad that you laid it out with, because I was going to ask about it, but you gave it without a prompt. I’m curious though, is that something that is like part of the software or is your software more like one and two—temporal, spatial—and then that’s done by somebody else?

Varun Sivaram (21:55.02)

So look, right now, if I’m being perfectly transparent, Emerald AI is focused on flexibility at the data center. So we’re doing temporal flexibility, spatial flexibility, and coordinating with onsite resource flexibility, right? Emerald AI makes it possible to co-dispatch an onsite battery alongside with your computational flexibility. So we do the first three, but I’m a big fan of all the different buckets.

Let me just say a moment on this hero path. Look, data centers are already striving to be heroes. My friend, Chase Lochmiller runs Crusoe. He’s an investor in Emerald. And they’re a standout example of a company that, when they come to town to Abilene, Texas, for example, they invest in the community, they invest in the workforce. Chase actually has this vision of building even more generation than the data center will need so that you’re actually reducing overall power system costs. And by the way, to your point, Doug, it would be wonderful if data centers, through their high willingness to pay for their compute costs, are also willing to therefore subsidize flexibility in the adjoining communities.

And another good friend, I was recently on a panel with Justin Lopez at Base Power. Base Power is an example of a company that is putting together a range of battery resources in a neighborhood or in a community, and they’re able to bid that in or dispatch that as an adjacent flexibility to data centers. So I’m just delighted that all of these great innovators are coming with solutions. Emerald seeks to be this glue that makes it possible for the data center itself to flex and to play nicely with all that adjacent flex out there.

Doug Lewin (23:24.928)

And so, okay, that’s super helpful. I appreciate that. And so it’s obviously like, you don’t necessarily need a company to do all pieces of that. I had Astrid Atkinson from Camus on recently and they do a lot of that like aggregation of the JSON you were talking about. There are lots of different companies that do that. I was just wanting to be clear on what you guys do, but you do those first three and I’m interested in how you think about, and I guess maybe it’s not you thinking about it, it’s the software, but you’re sort of training the software like, what are the kind of trade-offs between those things? I mean, it’s got to still be early days, right? There’s not a ton of information yet, I wouldn’t think, but maybe there is, about how to sort of stack those things against each other. We’re going to switch a workload, do a different data center. We’re going to move the time, or we’re going to use the resource that’s on site, or we don’t have enough resource on site. We need to put more. That, I assume, is what is going on with the tests you’re doing and the early deployments, is you’re getting the information to train the model to continue to refine it? Is that accurate?

Varun Sivaram (24:26.262)

Yeah, absolutely. Doug, you asked, you know, “Hey, Varun, are you doing the thinking here?” You better hope I’m not the one thinking because... Far too slow, far too slow to pull this off. The magic, the secret sauce behind Emerald AI—the thing that takes my breath away—is the autonomous intelligence, the closed-loop functionality, the way the Emerald AI set of agents actually just operate at scale. And so there was a recent test where, you know, we were just watching it, but it was kind of epiphenomenal. We had no ability to change it. We were just like watching the results pour in. And one of our partners was like maniacally changing the workload mix, starting and stopping workloads. And at one point, one of the workloads failed because I think it was improperly prepared. And the Emerald system did some behavior we had never seen it... We’d never designed it to do, but it was fascinating. It enabled the overall power draw to look to the utility like we were still very smoothly ramping down by 20, 25% and holding steady, even while under the surface, there’s this churn of all of these different workload behaviors. Some are starting, stopping, some are even failing, which we had never encountered before.

So I love watching the system autonomously and intelligently make these trade-offs in real time by taking into account, hey, what’s the user okay with? What’s the user’s priority level for these jobs? What can be tolerated in terms of temporal or spatial flexibility? And then to your question Doug, how does this stack along with the battery, right? Because the battery comes with its own set of constraints. It has a particular state of charge. You can dispatch a battery but you then need to recharge it before you dispatch it again. What if you get two back-to-back events without a recharge time in between? This is why we’re going to need a combination of flexibility approaches. It’s why compute flex is so impressive, right? Compute flex is powerful because we can do multiple events in one single day. We can do long events, even if you haven’t sized your battery to achieve an eight-hour event. Let’s say you’re in PJM and there is a long event. This has historically happened. Compute flex can really bail you out if you exceed the capacity of your batteries.

I like to think about a supply curve. There’s a supply curve along many dimensions of different interventions you’ve got at the data center. You’ve got your temporal flex, your spatial flex, you’ve got your batteries on site, maybe you’ve got a fuel cell. You’ve got a diesel generator that you’re only allowed to run for X number of hours because of the air permit reasons. And you’re stacking all of these interventions and intelligently and autonomously, you’ve got to make good decisions in the moment because, and we should talk about this Doug, the utility and ERCOT are counting on you not to screw this up, right? They’re counting on you that if there is a curtailment signal, you better perform. And so for the data center, the goal is to make sure you perform while protecting the sanctity of these customer workloads, which are again, the most economically valuable workloads in history. Don’t screw them up.

Doug Lewin (27:21.08)

So before we were going to talk about ERCOT in just a minute, before we do that though, I do want to just linger on this for a minute. I think there are two things I want to kind of unpack a little bit more because I think they’re just fundamentally important on just a kind of a foundational level. Because again, back to that Heatmap headline, “Everyone hates data centers.” Like I feel like there’s a disconnect with the general public here. You’re obviously like, I could just see you just like light up as you’re talking about this. Would you talk about some of the use cases of AI that excite you the most? I just finished reading, it’s sitting over there, The Worlds I See by—I don’t know how to pronounce her name—is it Dr. Fei-Fei Li, who’s one of your investors, right? It’s a beautiful book. I highly recommend folks read it. She gives some of those insights into how AI could actually improve healthcare outcomes. I feel like sometimes, and it is an energy podcast, so of course we’re gonna talk about energy, but we sometimes skip over and I think that’s some of the disconnect with the public. So just take a minute and just like on a human level, like what excites you about AI? It’s a big question.

Varun Sivaram (28:22.626)

Everything. The best way I can frame it is, you know, just last week, the King of England’s Coronet Awards just awarded this major prize to both Dr. Fei-Fei Li, our investor, as well as to Jensen Huang, the CEO of NVIDIA. NVIDIA is another big investor in Emerald AI, as well as to NVIDIA’s chief scientist, Bill Dally. Jensen says, you know, there is this paradigm shift we’re seeing. AI is different from other inventions. Other inventions have been tools. AI can actually use tools. This agentic AI future in which AI agents are using tools that formerly humans would use opens a whole new world of discovery.

It’s why I love Dr. Fei-Fei Li’s book that you just mentioned, The Worlds I See. You know, fine, putting my science fiction hat on, I fully expect that it’s AI that will cure cancer. It’s AI that will enable the end of road fatalities. Yes. As we have far safer transportation, road transportation, autonomous vehicles. It’s AI that will end the rigmarole of meaningless work and open up far more leisure activities for everyday working-class citizens who don’t have to do things that we can now automate. And for those worried about job displacement, it’s AI that I believe will create almost unbounded economic gains. They will, I hope, help the United States become more fiscally sound through this incredible economic growth and revenue. And I hope that they will create enough of an economic bounty that even ordinary working-class citizens just get to share in those rewards and live meaningful, productive lives.

But we do not get there unless we invest right now. I do fear that if we don’t take this seriously—first, if we’re uncompetitive with other countries in the world, and second, if we just slow our trajectory compared with what we could achieve—we won’t realize these rewards. And every year we go that we haven’t cured every kind of cancer is just a year of unnecessary deaths. I know I’m exaggerating in some sense, but seeing AI solve protein folding or solve... these are fundamental advances.

Doug Lewin (30:39.514)

I think about a lot, I haven’t talked about this publicly or, but my dad has Parkinson’s. It’s a devastating, just devastating disease that doctors just don’t, they just don’t know. Everybody just kind of like, it’s kind of a shrug. We don’t know what causes it. We don’t know really how to treat it. They can give you some medicine that does a little bit here and there, but my God, like if we could use AI to try to understand what causes that, I mean, just the amount of human suffering from Alzheimer’s and dementia and Parkinson’s. There was just a constitutional amendment in Texas to establish some additional Alzheimer’s research. Now, I don’t want to be Pollyannaish. We all know AI could be used for bad stuff too, right? But this is where, when you talk about hero path and villain path, there’s that on the grid, and then there’s that more generally. And all these things we should talk about, which is really important, is why we’re talking about it right now. But I think a lot of times people are not thinking enough about how much it can help.

You mentioned cars. 40,000 people die on the roads every year. I think the math is something like equivalent to like a couple of plane crashes every week where everybody on board dies. If that was happening, the public would be up in arms, right? We wouldn’t accept it. We absolutely would not accept it. But somehow with cars, we all just kind of go, “Oh well.” It is unnecessary. We’re already seeing with Waymo the crashes and they’re running around Austin all the time and they’re... the rates of both injuries and of even any kind of crash are down. I forget what the numbers are—70, 80, 90%, something like that. So anyway, okay. Anything else you want to say about that before we move on? So we’re going to go to ERCOT in a minute. The other piece I just wanted to, and this relates to the ERCOT discussion as well, but the other piece I wanted to dial down into just a little bit more, drill into a little bit more is affordability. So we were talking about that fourth bucket you were talking about—adjacent where data centers could pay for reductions. I’m real excited about that, but there is a more fundamental way that data centers actually can help lower costs, which is, and this was in that LBL study that’s gotten a lot of traction recently over the last couple of weeks that states that have higher energy use actually have seen their rates go down. Now that’s uneven. Rates are going down more for large users than for residential. And that’s something we need to talk about and work on.

But overall, like the math is pretty simple, right? You have a fixed cost of a system and the more you spread those fixed costs out over multiple users, the lower costs go. So I don’t know if you want to say anything about that, but it’s just something I want to put out there more and more for folks to think about. That’s sort of part of the hero path as well, is as long as we have the right regulatory systems in place and ERCOT will be working on changing some of the transmission cost allocation and all that, there’s a real potential for costs for everybody to go down just from that simple math equation before you get into any of the whiz-bang exciting things AI can actually do to make the grid more reliable or affordable.

Varun Sivaram (33:33.826)

Doug, you’ve nailed it, so I won’t spend too much time repeating what you said, but it is a simple equation, but it’s central to the hero pathway. Because if you have to build out infrastructure faster than you’re bringing on the revenue from new kilowatt-hours, in other words, if you have to pay for new kilowatts faster than you get revenue from new kilowatt-hours, then everybody’s rates go up. Today we have a cost allocation problem because peak demand is rising rapidly, you have to pay for all of these new pieces of grid infrastructure, transmission lines, substations, as well as generation. And there aren’t as many kilowatt-hours getting paid for in order to make the math work. And so we socialize the cost. And then we have arguments over, “Well, should data centers pay more or should communities pay more?” et cetera. You can sidestep a lot of that through data center flexibility that allows data centers to better utilize the existing infrastructure.

Look, I still think we’re gonna need more. We’re gonna need to build more grid infrastructure and to modernize it, we’re gonna need more power generation capacity, but we can build it out prudently. And so the pace of kilowatt increases is outpaced by the kilowatt-hours that we get productive revenue from and that pays for all of this. So you have less of a cost allocation problem because you don’t have to fund this and therefore, it’s less about “What should communities pay or should data centers pay?” It’s that, as you said, flexible data centers coming on the system should actually reduce costs for everybody because the new kilowatt-hour payments really ought to pay for more than their share of what the kilowatt capacity increases are. So that’s the simple equation that I want us to keep in mind as the central tenet of the hero path.

Doug Lewin (35:17.934)

Unless anybody think that that just sounds very futuristic, this has actually happened over the last two years in Texas where the peak demand, the highest peak demand we had in 2023, we did not reach in 2024 or 2025. But our minimums for every month, the minimum demand in 2025 is up every single month compared to 2024. Our electric use is up like 11, 12% over the last two years. EIA thinks it’s going up 14% next year alone. So usage, overall usage of the system is going up while the peak is not. So exactly what you’re describing, it’s not just a futuristic thing. It has happened over the last two years, whether we keep it going or not, is gonna depend on good policy and market structures and all that kind of thing. Which brings me t