Inside the Factory Using Rocks & Sunlight to Fix AI's Power Problem | Exowatt

“The AI race is so intense. It's now time to power, right? I need power, I don't care how you make it. If you talk about a gigawatt data center, that's almost a million US households of energy. The grid is just really not built for that. What that means is the data centers now have to bring their own power online because it's also important that when you plug in a data center, it consumes a lot of power from the grid.”

“This essentially focuses the light coming in from the sun onto the battery material. It gets very hot and then stores that energy. So, essentially, our battery is a heat battery, so what that means is we're heating up rocks, right? And then storing that energy in formal heat, which is very cheap as compared to doing lithium ion or any other type of electrochemical battery, and you don't have to get lithium or magnesium or cobalt or any of these other fancy chemicals.”

“What's new about our approach and our hypothesis is making this in a modular fashion, making this in a factory, and then scaling the production to millions and ultimately billions of these, which ultimately allows us to go down the cost curve and generate electricity at our ultimate goal of one cent per kilowatt hour. And so that's why we started the company. That's kind of been the goal North Star from the beginning.”

“The three elements are the optical table, so these lenses, and then you have the heat battery and the PC, or the power conversion unit, all in this kind of container. And the idea is that we capture energy from the sun throughout the day. So think of this as like a solar panel, but that has a built-in battery, and that essentially you can scale linearly to any project size. If a datacenter customer needs 100 megawatts or a gigawatt, you just stack these next to each other.”

“We need to manufacture domestically. We need to source domestically. We need to make this using raw materials that don't come from China, that are not rare of minerals. And all of that is what we continue to iterate on. So every configuration that we design, we looked at the bill of materials or the BOM. And we asked ourselves, what can we eliminate from this? How can we make this simpler? And the same approach you see at SpaceX, right? Or at Tesla.”

“I think there's a lot of headlines about data center projects being planned or initiated. I don't think all of them are actually happening. So there's a lot of like phantom data centers that are being announced. Let's call it that way. I think data centers are definitely in a rush to get online fast. They do have massive supply chain constraints. Two years ago, it was maybe NVIDIA GPUs. Now, it's power. Now, it's actually not even hardware. It's actually labor.”

“The last thing about everyone coming out of Tesla, you're obsessed with vertical integration, you want to build everything in-house. We haven't done that at Exowatt to that extent yet. We've started working with contract manufacturers because we know we need to scale fast quickly, so they have that infrastructure. But Elon really taught us how to vertically integrate everything and build it yourself and build it better, faster, cheaper, and don't take no for an answer.”

“I don't think people contextualize how much energy that is. So if you talk about a gigawatt data center, that's almost a million US households of energy. So if a data center says, I'm building a five gigawatt data center, I'm basically building a city that's five million households. And they're building in a very tiny footprint, although not so tiny. The funny thing about data centers is, when they start talking about buildings in acres, you can start realizing how big these buildings are.”