00:00AI's power demands are putting nuclear back in the spotlight.
00:04Now, Valor Atomics says it's the first company in the US to use an advanced nuclear reactor
00:08to actually generate electricity, and then it powered an NVIDIA Blackwell chip.
00:14It was a small demonstration, but it comes as the two companies announce a partnership
00:18to explore nuclear-powered AI systems, and actually there's a bit more in there as well.
00:23Joining us now is Valor Atomics CEO Isaiah Taylor.
00:26I want to be transparent about the scale of the demonstration, but it was a number of firsts.
00:33Explain what happened.
00:35Yeah, thanks so much for having me on, Ed.
00:37We're standing here right now in front of the Ward 250 reactor, where we just powered the NVIDIA Spark yesterday,
00:43the first ever advanced nuclear reactor to power an AI chip.
00:47Ed, you're right, it was a very small demonstration.
00:49This reactor only makes 100 kilowatts.
00:51That's very intentional.
00:52It's part of our philosophy.
00:53We like to move quickly in small steps, right?
00:55So, this is a small demonstration, but it is a big first.
00:58It's also the first time an advanced reactor has been built in American soil outside of the national lab system.
01:04And yesterday, we actually became the first ever startup to make nuclear electricity.
01:08So, a bunch of awesome historic things and a small demonstration, and we're really proud to partner with NVIDIA to
01:13do it.
01:14I think what is interesting about this is the technology itself, right?
01:20Could you just bring us up to date, Isaiah, on where this sort of technology stands today?
01:24I appreciate you've outlined it was a first, but the challenges of scaling, the benefits of using that technology relative
01:31to other energy sources.
01:33A hundred percent.
01:34So, I mean, listen, nuclear has been around for a long time, and we've built these very large-scale plants
01:39that have made a lot of electricity.
01:41But the problem with them is that they haven't scaled very well.
01:44They take a long time to build.
01:45They use a lot of civil infrastructure, and these are things that we're not as good at in the United
01:49States anymore.
01:49So, the philosophy of Valor Atomics is we want to build modular plants that are manufactured.
01:55And the other really unique thing is that they are high-temperature reactors.
01:58This is standing behind us here.
01:59This is a high-temperature gas reactor.
02:01And those high temperatures are actually really good for both efficiency, but also for things like air cooling.
02:07So, a lot of the power debate and AI scaling debate right now is how do we do this without
02:12having to tax local communities of water usage?
02:15One of the best ways to do that is just to raise the temperature of the reactor.
02:18This allows you to reject heat at a higher temperature, which means that you don't need water cooling anymore.
02:22And that was part of our announcement yesterday.
02:25I want to stay with the waterless component of it.
02:28You know, NVIDIA has been speaking publicly a lot about the economics of cooling through different technology sets.
02:36It is a big focus.
02:37They've tried to dispel some misunderstanding that's out there.
02:41How does your technology work in that respect, and how unique is it in the field?
02:47Absolutely.
02:48There's two sides of this equation, right?
02:49We need to get the energy generation side to be waterless, and we need to get the data center cooling
02:54side to be waterless.
02:55And that's why this collaboration with NVIDIA is really exciting to us.
02:58We realized that we were both working on the problem from different angles.
03:02We're working on a nuclear reactor that doesn't need water cooling due to high reject temperatures.
03:06And NVIDIA has already been working on the data center architecture to do that.
03:10And so when you bring these two technologies together, this collaboration for a 30-megawatt data center with no water
03:16cooling needed to extract water from the local community, that's really what's going to allow us to scale.
03:22We think it is very important for the United States to win on AI, to win this scaling race.
03:27But we need to do that without taxing local communities in both power price and water usage.
03:32And so NVIDIA and Valor Atomics coming together in this collaboration is really solving both sides of that problem.
03:38NVIDIA has historically made equity investments in all layers of the stack with the argument that they want to support
03:50the ecosystem to get them moving faster.
03:54Have you had any discussion with Jensen or with NVIDIA about the idea of there being a financial relationship between
04:01Valor and NVIDIA?
04:03You know, I can't announce anything on fundraising today.
04:07You know, we have a lot of work to do here that we're focusing on and a lot of awesome
04:10capital providers behind us allowing us to go do that.
04:13But what I will say is that this collaboration is focused on how do we take the next step of
04:18scale, right?
04:19I think that NVIDIA has done a lot of fundamental work on scaling the compute side and the architecture and
04:24their DSX architecture.
04:25They've done a lot of this really fundamental work on scale.
04:27And we know that energy has been what is missing.
04:30You know, natural gas is going to be a bridge.
04:32It's going to be a good bridge.
04:34But it only goes so far and we continue to have the climate change question associated with that.
04:38And so if we really want to scale, and especially outside of the pre-existing natural gas infrastructure in terms
04:44of pipelines, we're going to have to do that with uranium.
04:47Isaiah, we just have 45 seconds to end.
04:50Demo one, a single spark.
04:53What are the milestones that come next in terms of scaling the demo and infrastructure at different levels?
05:00Yeah, Valor Atomics likes to take small steps quickly.
05:02We're going to go build another reactor.
05:04The next one is going to be a lot more powerful than this one.
05:06And we'll take the next step in demonstration on the compute side as well.
05:10So stay tuned in the next 6 to 12 months.
Comments