Morgan Stanley raises price target on Vistra, here's why

Just first of all, maybe this is an ignorant question, but why is AI using up so much power? We knew it was going to use a lot of compute power. Is it just a byproduct of that, that it needs energy as well, or or is an an added factor? Well, that’s exactly right. And what’s interesting is every new generation of GPU that NVIDIA comes out with is more energy efficient. And there’s a lot of attention on that and I I respect that. It’s simply that the absolute magnitude of demand for compute power is growing at an astronomical rate. So we’re seeing a lot of the new chips. That’s great. But every new chip does use more electricity. It’s just that every new chip has massively more compute capability. So essentially compute growth is astronomical power demand growth very high, about 100% a year for Gen. of AI even with that improvement in efficiency of the chip. So this is really a byproduct of just compute growth and and as we stand how close are we to putting too much pressure on the grid? Are we, are we close to capacity already or or not? We are close to capacity if not at capacity in several markets. You know in our note, we analyzed all the power markets around the world and we’re seeing major pressures in places like Dublin, Ireland, Santa Clara, CA, parts of Southeast Asia. So we’re there in certain parts, but other markets, other states etcetera really want to welcome new data centres. So it is a bit of a mix. It’s just that broadly speaking, planning to get new data centres onto the grid is very arduous and the delays are increasing to get that done. And and in terms of then what’s next and how you should play this? Are we, are we talking before we get into the names about people that are going to help expand their grid, people that are just building data centres where there is power, What, what, what’s the actual themes behind this? Yeah, great question. It’s really, it’s not so much the grid providers, it’s the power providers that can provide the fastest time to market for data centers. For example, instead of owning a grid company that’s going to slowly increase their spending on on the grid, I say slow less than 10%. What we’re interested in is a power supplier like a nuclear power plant owner that can provide an alternative to the grid, provide power much more quickly and achieve much higher margin growth as a result. So let’s go through some of these names. Vistro, you’ve just upped your price target and significantly, yes, Bistro big Texas based power generation company, lots of different types of power plants and they have a nuclear power plant in Texas that we think is ideally situated for a lot of additional data centers to be built right there. The advantage of your data center company is you basically cite the data center behind the fence line at a nuke. You avoid having to go through the fairly arduous grid upgrade process altogether and that saves you a lot of time. And what we found in our analysis is that time value is immensely valuable for hyperscalers. Really, really interesting Constellation, similar constellation. Think of that as more shots on Gold and Bistro, bigger company. Constellation is the biggest US nuclear power plant owner with nuclear power plants in many markets. I think Constellation is likely to achieve multiple large contracts with the hyperscalers, with the data center developers in places like Pennsylvania that’s that’s quite well suited. So you’ve got NRG Energy and Public Service Enterprise Group also on your list. I we’ve only got a minute left. My other question was going to be surely AI should make the grid more efficient. I mean, is there not going to be a kicker to offset all of this in five years time, maybe quicker when when some good aspects are harnessed? It’s an excellent point. AI is just starting to be leveraged and I do see some benefits in terms of optimizing the grid. The challenge is what we’re talking about here, data centers is massive amounts of power in very concentrated locations and you can’t optimize your way around that issue. The process to upgrade the grid is a established process that’s been around for a long time that just doesn’t move on a dime. So essentially the problem we have is on one end very, very fast growth in Gen. AI. The other We have an established grid planning process that just is not going to suddenly become incredibly efficient because of AI.