We have Libby who has three billion under management. We have the head of supercomputing at Lincoln Laboratories coming on down. I teach a class on the history of MIT and the biggest budget is Lincoln Laboratories. People don't realize it. And it's all because of his leadership and what he's doing. And we have Chase Lochmiller who's zooming in, who has definitely gotten two Chiaetss. He's spoken at all imagination action and he graduated in three years from MIT, got a Stanford degree so he's a double double one. And Dave is such a thoughtful person, really cares about this topic, he will be leading us. Dave, take it away.
All right, here we go. So this section is called Building the Backbone: Energy, Compute, and the Future of Infrastructure. And we have actually an incredible panel. If we can get Chase up there, he'll be here in a second I think. But we have the head of supercomputing at Lincoln Labs. Like you said, the biggest budget at MIT. Chase just showed up. Hey, Chase. How are you? Can you hear me?
Hey, Dave. What's going on? How's it going?
And Libby is actually teaching here at MIT and Harvard and has a long... actually, why don't I let you guys introduce yourselves very quickly and we'll just go down the line. Starting with Jeremy.
Thank you. It's a real pleasure to be here. I've been a big supporter of this activity over the years and Media Labs. Fantastic. So I'm Jeremy Kepner. I have a lot of hats at MIT. We all do here at MIT. So I founded supercomputing at Lincoln Laboratory. And it's been focused on data and AI at Lincoln Laboratory for its entire existence. People often don't realize that Lincoln Laboratory funded the famous Dartmouth meeting where the term AI was first coined. And it was led by a person who was leading the machine learning group at Lincoln Laboratory in 1953. So we've been doing AI for 70 years and that's just been a core of what we've been doing.
Yeah. Hey everyone. It's great to be here. Libby Wayman. I'm a partner at Breakthrough Energy Ventures. Breakthrough Energy Ventures is an investment vehicle that invests in early stage technologies that can enable enormous growth and value creation while decoupling that growth and economic value creation from the growth of emissions. And we invest across the economy. We start at the early stage. We also have a growth fund and we've been investing at the intersection of AI and climate change and energy and industry since the very founding of our fund. My background came out of this institution as an alum for undergrad and grad school, started off working on computational models for the dynamics for energy technologies like wind turbines and early stage technology development in solar and translating early stage technology to manufacturing. Had the opportunity to serve as a technical appointee at the Department of Energy to help the department establish a manufacturing strategy based on that work and was a brief time at GE leading clean technology innovation across the company and have been at Breakthrough for about the last eight years. So great to be here. Energy and compute and AI are meeting right now at this moment and there are a lot of interesting technical solutions I'm excited to talk about. And last but certainly not least, one of the coolest guys that ever graduated from MIT, Chase Lochmiller, graduated early in three years, which is hard to do, while playing varsity soccer at the same time, but then climbed six of the tallest seven mountains in the world while being a crypto god and is now building, I think, the biggest construction project potentially in the history of the world in Abilene, Texas, the half trillion dollar Stargate buildout. But he's the founder and CEO of Crusoe now which is a hugely successful startup. Chase, you want to introduce yourself?
Yeah, thanks for having me. Hopefully you guys can see me. I can't see anything but...
We can see you clear as day and we kind of like that.
Perfect. All right, perfect. Perfect. It's like one way. I'm thrilled to... I apologize I couldn't be there in person. I always love getting to campus. But as Dave mentioned, I'm an MIT alum, course 8 and course 18. And initially started my career, I was actually working in plasma physics, working on plasma physics research at Los Alamos National Lab. And then I got into the quant trading world where we were a big early adopter of a lot of different machine learning techniques, initially using a lot of classical machine learning techniques and then shifted to a lot of deep learning tools to forecast certain security returns in the marketplace. And that's really where I started experimenting a lot with running workloads across GPUs. And I started experimenting with computer vision models and natural language processing models. And I really had this high conviction that intelligence was going to be embedded in every aspect of the economy. It really felt like AI was this super science that literally made everything better. And as I was sort of playing that out and thinking about the scaling of compute required for AI to transform the economy, I really felt like at scale energy became the bottleneck. So we started Crusoe with this very energy-first approach to developing computing infrastructure. And for those of you that are familiar with the data center landscape, a lot of the data center infrastructure today that runs the internet sort of lives in these hubs like Northern Virginia. And we fundamentally had this belief that the infrastructure of intelligence that would run AI would be in marketplaces that had low-cost and abundant energy resources. And that's really what led us to building this vertically integrated AI infrastructure business that works on energy development, data center development, has a large-scale GPU cloud platform called Crusoe Cloud that enables AI innovators to do their life's work without having to worry about the management of the infrastructure layer and really spans a lot of different things that frankly I'm very interested in and I think are helping accelerate innovation across the AI economy. So, I'm excited to be here to talk about the infrastructure layer of AI.
Right. Let's start talking about scale because the scale of what's going on right now has never been seen before in the history of the world. So, I was at the Gigafactory, Elon Musk's Tesla factory in Austin a couple months ago and got a whole tour. It's three kilometers around the edge of the factory. And so, they'll take raw metal, melt it, put it into a mold, and turn it into a car in 90 minutes. And so, you can just walk down this three kilometer walk watching a car get made. So, as we were talking to Elon, we're looking at this other building across the street and it's billowing smoke, just huge amounts of smoke coming out of it. And we say, 'What's that building?' And they say, 'Well, that's the data center where we're training the AI to drive.' And it uses more energy than all of this car metal melting that you see right here for three kilometers. Like inconceivable amounts of energy. So that really for me put it into scale. But why don't we start with you Chase and work down this way. Just describe to us the magnitude either by investment or parts or just square footage, just the feeling of scale.
Yeah. So I think it's always important to put scale into perspective and I think AI is driving a multiple order of magnitude change in the amount of power requirement to run global computing infrastructure. And so to give you an example, Northern Virginia, that hub that I sort of talked about, end of '24, the total capacity in Northern Virginia was 4.5 gigawatts. Today, Crusoe as a business is working on more than four and a half gigawatts. And we're just an 8-year-old company. So, there is scale to this infrastructure. I mean, I think OpenAI was trying to contract for 10 gigawatts. They ended up contracting 8 gigawatts last year. So to put it in perspective, these numbers are very large and crazy. Our campus that Dave alluded to in Abilene, Texas, we just announced an expansion there to 2.1 gigawatts. Putting that in perspective, like I grew up in Denver, the city of Denver. Denver runs on about one gigawatt of power. So you're talking about two Denvers worth of power to run this large-scale data center. But why is that investment worth it? Fundamentally, it's creating a new technology that will change people's lives in a hugely beneficial and positive way. The new models that are going to be developed there will have new capabilities that fundamentally weren't ever possible in the history of humanity. They will enable new types of drug discovery, of energy development, of material science development that will fundamentally move humanity forward and improve quality of life around the world. And it's also creating a massive amount of energy abundance because it's catalyzing net new investments across renewables, across batteries, across gas infrastructure, across all of this energy we need to basically power this infrastructure. And when it's not being used by the data center, it's actually bringing down overall costs for local ratepayers. So it creates this huge opportunity to create an abundance of not just intelligence but also of energy.
Yeah, Libby, fill in the gaps there. So let's expand the thinking there just in terms of implications and also locations, you know, where is all this going to happen? What does it mean for those economies, those areas?
Yeah, I mean this is right now kind of a US and China phenomenon. In the United States, about 4.4% of US electricity consumption went to data centers by the end of 2024. So zooming out from the specific data points that Chase shared, the trajectory that we're on is that data centers could consume about 20% of US electricity, which is basically expanding the electric grid and electric consumption in the US by 20% by the end of this decade. Putting that into perspective over time, the United States electricity grid and overall energy generation consumption has not really grown for the past several decades. So, this constraint that we're feeling right now is really just from a couple percent of increase and to take that up to 20% increase is going to put a massive strain on the system. What that's looking like at individual local economies is you're starting to see some modeling that indicates that electricity prices will go up, emissions will go up. I think you probably saw that with your own eyes from the energy generation down in Texas. Because data centers are finding whatever power they possibly can and not all of it can be clean. It's really about speed to power. You hear that phrase a lot and people are employing even reciprocating engines on site behind the meter to solve these problems. A lot of developers are really thinking about how they bring their own power so that they're not putting that growth in capacity and the extra cost on the utility system and on the ratepayers. They're thinking about how they adopt advanced technologies not only for generation but also for reducing energy consumption within the data center. If you look at older data centers, the power utilization effectiveness is about two, which you should think about that as kind of the inverse of efficiency. So the amount of power for an overall data center over the power that just drives compute in older, smaller data centers, it was around two, which is about 50% of the electricity goes to the actual compute. And now in the newer data centers, and Chase, we'd love to hear how you guys are going about this, it's really almost close to unity. It's about 1.1. So that's really exciting. Some strategies that people are using there are really efficient power conversions, obviously advanced cooling, really low loss data transfer, which is actually a surprising source of electricity losses in data centers, and then what's a really exciting frontier is sparse models. And I'm excited to see where that goes.
Yeah, actually Jeremy, why don't you fill us in on like, you know, when we say massive data centers, huge power under the covers, it's Nvidia GPUs, you know, millions of them, some AMD, a little bit of Intel, but I mean, you've been living this transition for a long, long time. Put it in perspective, but then tell us where it's all going.
Yeah. So we observed this energy importance in computing I think around 2002 and it was fairly obvious to us here at MIT that energy is going to be very important. And so we started thinking about our next generation data center which we felt would be a model for the world and we think it is, in about 2005 and we had the luxury of time so we studied it for a long time with all the brains here at MIT.
