Power shortages will constrain AI more than chips

Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

1. 40:42

   I think the way to think about it is when you look at load growth, just take the United States, because I think that's a more tractable example. When you look at the load growth and you look at the sources of load growth, they're relatively pedestrian. It's more electric cars, more electric fridges, more electric televisions, PCs, laptops, and data centers. But it's just broadly speaking that we are more of an electrified society.

2. 41:08

   So there's just like this, you know, this demand of a need for electrons, basically. And then when you look on the supply side, there's huge issues. even if you assume that every project that's planned actually gets not only permitted and greenlit, but also built on time and on budget. And when you net those two things out by 2050, we're still 1.4 terawatt hours short, which is effectively six California's worth of electricity.

3. 41:34

   So there is no world in which we catch up. And then people's reflexive reaction to that is, but nuclear. And I'm like, well, I just think the problem with SMRs as an example is the S in the MR. It's like, you know, what people don't realize is it's the same unit effort to get an SMR approved as an MR approved or as an R approved, as an LR approved, a large reactor. Because, you know, you're going through state, local ordinances and people and conflicts and pushback and all of this stuff that does not change whether the form factor is it goes under your daughter's bed or it sits, you know, over there and it's gigantic. So we're just massively short and that's where my thesis started which was just that observation then you know where i spent a little bit of well not a little bit a lot of my time

4. 42:26

   was what is the most tractable all of the above solution Because, you know, once you realize that we're so short, you immediately abandon all of the rhetoric around climate change because you're like, just give me anything. Oil, not gas, coal, wind, solar, nuclear. And if you look at them all on just a total, you know, cost basis, what jumps off the page is solar is slightly a little bit more expensive than wind, but they're the cheapest.

5. 43:00

   but it's just much faster. It's two years. It's do no harm. It's plunk it on the ground and off you go. So I spent just a bunch of time accumulating solar assets. And then that accidentally led me into the data centers because it was like, you know, through NVIDIA and Grok, what I realized was, my gosh, the capacity buildout was so upside down because of all these constraints. Because the data center is like this incredibly temporal moment where all energy and energy, meaning like people's energy, ire, anger, they all meet right there.

6. 43:32

   And so, you know, you have the permitting issues, you have the pushback issues, you have the political issues, you have pushback against the oligarchy, you have capitalism, you have wealth inequality, all concentrated in that thing. And so, you know, about two years ago, I started to kind of parcel together assets that I thought would be valuable in a world where that looks like it is it is the scarcest asset not that's how i underwrote it that's that's impressive that yeah it's not it's not it's not it wasn't cheap it wasn't cheap but uh but and by the way just to be clear there's huge issues with that meaning even when you have these assets um when you actually try to get them turned on and get them built my gosh it's incredibly complicated the funding cycle is very complicated the supply chain is very complicated the labor

7. 44:21

   shortages on critical people, because these are licensed, you know, skilled workers. You can't just, you know, grab some bumpkin off the street or go to like, you know, I don't know, Craigslist. I mean, you have to go and find like a licensed master electrician. They're very, very hard to find. They cost a lot of money. When you start to put all of this stuff together, we are probably going to see a massive, massive, massive compute shortage in the next three or four years.

8. 44:45

   Massive. Absolutely. Massive. so i love the fact that you said that because so many that's the biggest look so many people come back and say what what are we going to do what's my purpose where where's my mission everything you just outlined it's not even going to be an issue for the next five years Yeah, exactly. All this whole, look, I do think that we have achieved AGI. It cannot diffuse. And the reason it cannot diffuse has nothing to do with the technology.

9. 45:06

   It has everything to do with the buildup. And until you actually build it out so everybody can feel it, or a plurality of human beings can feel it, it's just something that exists in a lab. Now, that has value. Don't get me wrong. But I would not underestimate the difficulty. And if you look at, for example, the Anthropic S1 that just dropped today, it's shocking to me that the claim that they have to put as a risk factor, which I would never have thought would have happened for a company growing this fast and so powerful with all this money, the hyperscalers will not give them compute.

10. 45:42

    I've never seen that before. Wow. Yeah, it's wild. No, it's a really good indicator. Also, Elon was saying there's 15 gigawatts of chips just waiting for liquid-cooled data centers to fall into. that's a massive asset depreciating idly it's it's worse than that there's there's that then there's also the the not the non vera ruben stack chips the older generation that are also still sitting there's a bunch of these guys that have these things sitting in inventory those aren't being used then even when you try to do liquid cooling if you saw what happened with nscale and microsoft you had two massive leaks of coolant two times in a row where microsoft finally just triggered the liquidated damages clause and exited the deal yeah so everybody thinks this stuff is easy and it turns out well again you know back to ilan as he's proven physics is really hard

11. 46:31

    and building things and mechanically building things is really hard and it's probably particularly hard because we've spent you know 20 years teaching the younger generation of kids who should have been learning how to do these trades just to you know be on tiktok and social media and video games. And so, you know, the, the, the, the ability to actually build these things is gone. Yeah. And it's not just trades too. There's regulatory components. If you're a good negotiator, you can get involved. There's planning. There's, it really involves every human skill. And it just drives me nuts that so many people are saying, well, look, math is cooked. Teaching is soon to be cooked. I guess I'll just sit around and watch, you know, late night TV and maybe play some video games. No, no, no. There's a, there's a bigger opportunity that just opened up than anything

12. 47:14

    you've ever seen before. You just got to get into it. But it's also not that fatalistic. I just think people grossly misunderstand. They're wildly, wildly wrong in the short term. They'll be right in the long term, but they're wildly wrong in the short term because of these constraints. I'll give you another example just to give you an example. give you an example. In America, the decrepit infrastructure we have has forced people to organize into some of these co-ops. What is a power co-op? It's basically across a multitude of states. And so this undergirds our growth. And so what happens in a co-op? Well, you think a co-op is great. Everybody just like kumbayas and agrees on stuff. But really what happens is you have small players and big players in the co-op. And there's an example of one that goes

13. 47:55

    right through the spine of America. And there's an incredible amount of data center capacity that one could build in the northern states of this one path. And it would be great because it would alleviate tremendous constraints that we have in diffusing AI. And so I went in, my partner and I, we bought some stuff. And it turns out that they're part of a co-op. And when you dig in a few layers to your point it's all personal relationships and shit flows downhill and so you know the big boys who are actually in the south actually dictate how the co-op works it's like an organized mafia and so you know when you're up in the northern states you can't do anything you can't build you can't get the right off take agreements and you wonder why and it's because the guys down south are like you know oklahoma gas and electric calls the shots so