Dylan Patel: evaluation

Transcript context

…But beyond these numbers. If I’m remembering correctly, your blog post on how AI labs are increasing power implied that GE Vernova, Mitsubishi, and Siemens could produce 60 gigawatts a year in gas turbines. Then there are these other sources, but they’re less significant than the turbines. Only a fraction of that goes to AI, I assume. If in 2030 we have enough logic and memory to do 200 gigawatts a year, do you just think that these things are on a path to ramp up to more than 200 gigawatts a year, or what do you see? Right now we’re at 20 or 30. This is critical IT capacity, by the way, which is an important thing to mention. When I’m talking about these gigawatts, I’m talking about critical IT capacity. Server plugged in, that’s how much power it pulls. But there are losses along the chain. There is loss on transmission, conversion, cooling, et cetera. So you should gross this factor up from 20 gigawatts for this year, or 200 gigawatts by the end of the decade, to some number 20-30% higher. Then you have capacity factors. Turbines don’t run at 100 percent. If you look at PJM, which I think is the largest grid in America—covering the Midwest and some of the Northeast area—in their models they want to have roughly 20 percent excess capacity. Within that 20 percent excess capacity, they’re running all the turbines at 90% because they are derated some for reliability, maintenance, and so on. In reality, the nameplate capacity for energy is always way higher than the actual end critical IT capacity because of all these factors. But it’s not just turbines. If you were just making power from turbines, that’s simple, boring, and easy. Humans and capitalism are far more effective. The whole point of that blog was that, yes, there are only three people making combined-cycle gas turbines, but there’s so much more we can do. We can do aeroderivatives. We can take airplane engines and turn them into turbines. There are even new entrants in the market, like Boom Supersonic trying to do that and working with Crusoe. Also there’s all the other ones like that already exist in the market. There are also medium-speed reciprocating engines: engines that spin in circles, like a diesel engine. There are ten people who make engines that way. I’m from Georgia, and people used to be like, “Oh man, you got a Cummins engine in there,” regarding RAM trucks. Automobile manufacturing is going down, so these companies all have capacity and could scale and convert that for data center power. You stick all these reciprocating engines in. It’s not as clean as combined-cycle, but maybe you can convert them from diesel to gas if you want. What about ship engines? All of these engines for massive cargo ships are great. Nebius is doing that for a Microsoft data center in New Jersey. They’re running ship engines to generate power. Bloom Energy is doing fuel cells. We’ve been very positive on them for a year and a half now because they have such a capability to increase their production. Their payback period for a production increase is very fast, even if the cost is a little bit higher than combined-cycle, which is the best for cost and efficiency. Then there’s solar plus battery, which can come online as those cost curves continue to come down. There’s wind, where you might only expect 15 percent of the maximum power because things oscillate, but you add batteries. There are all these things. nline as those cost curves continue to come down. There’s wind, where you might only expect 15 percent of the maximum power because things oscillate, but you add batteries. There are all these things. The other thing is that the grid is scaled so we don’t cut off power at peak usage on the hottest day of the summer. But in reality, that’s a load spike that is 10-20% higher than the average. If you just put enough utility-scale batteries, or peaker plants that only run a small portion of the year—and those could be gas, industrial gas turbines, combined-cycle, batteries, or any of the other sources I mentioned—then all of a sudden you’ve unlocked 20% of the US grid for data centers. Most of the time that capacity is sitting idle. It’s really only there for that peak, which is just a few hours over a few days of the year. If you have enough capacity to absorb that peak load, then all of the sudden you’ve transferred it all. Today, data centers are only 3-4% of the power of the US grid, and by 2028 they’ll be 10%. But if you can unlock 20% of the US grid like this, it’s not that crazy. The US grid is terawatt-level, not hundreds-of-gigawatts-level. So we can add a lot more energy. I’m not saying it’s easy. These things are going to be hard. There’s a lot of hard engineering, risks people have to take, and new technologies people have to use. But Elon was the first to do this behind-the-meter gas, and since then we’ve seen an explosion of different things people are doing to get power. They’re not easy, but people are gonna be able to do them. The supply chains are just way simpler than chips.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.