Last Tuesday afternoon the following text caught our eye:

Dell let slip during their earnings call that the Nvidia B100 Blackwell will have a 1000W draw, that’s a 40% increase over the H100. The current compute bottleneck will start to disappear by the end of this year and be gone by the end of 2025. After that, it’s all about power.

Read more here.

This finally nudges to work out a thought we have been kicking around for the last few months, maybe longer.

Be long energy infrastructure.

We have written before about how, in the AI gold rush, it is far better to own the shovel makers like Nvidia than the prospectors like Meta or Google or what have you.

This has worked out and, as Nvidia's recent earnings blowout spectacular revealed, the challenge of finding enough GPUs to satisfy the new demand for sophisticated AI models and data centers is still very real.

There is simply way more demand than Nvidia can satisfy. This is sending their profit margins ever higher in the short term.

Now, as the above extract makes clear, two things are becoming more probable:

1. The end of the compute demand is in sight even if it is not yet here. Sometime soon more supply, more competition and, most especially, more innovation will satisfy the market.

2. New challenges are rapidly coming into view.

What is interesting is all the speculation and analysis and just plain old attention was and is on the compute bottleneck of not enough chips to meet skyrocketing demand.

There is much less focus on some of the more mundane and practical challenges involved.

Namely that you don't just need high powered and ultra sophisticated AI chips for this revolution to take place. You also need a seemingly endless number of new data centers for them to work together in.

It is true that one of the reasons no one really talks about this is it is a pretty dull topic. The other reason very few people are paying attention to this is that no one ever thought this would be a problem.

On the surface data centers appear just slightly fancy warehouses. They might be the only thing more boring than the energy grid itself. Traditionally these data centers require land and the basics: water, power and being connected to the rest of the grid and internet etc.

When there were very few data centers none of these were problems. There was plenty of land and plenty of power and water available in most of the West. Even better, a data center was relatively easy to service.

Except now those issues are increasingly a challenge.

As the earlier quote makes clear. Newer chips are not just faster and more powerful. They also require far more energy.

This is very rapidly becoming a problem especially in certain places that already have a lot of data centers or are struggling with electricity supply during the sloppy transition away from fossil fuels power generation.

This has been a huge change.

Originally these businesses were considered great by just about everyone. They used up under utilized post-industrial or commercial land and were also the ideal tenant: quiet, safe, low maintenance, flush with cash and keen to expand.

Whether you are a landlord or a mayor or simply a neighbor, compared to dealing with a chemical plant or a manufacturing facility a data center was pretty sweet. You likely wouldn't even know it was there.

Now places like Loudon County in Virginia near Washington DC and the region around Dublin, Ireland are realizing that each approved data center doesn't just require significant energy but this energy demand leaps higher as the sophistication of semiconductor chips advances.

Those advances are happening quickly and the AI boom is leading to fa more demand for data centers. This is suddenly a big problem for places that are struggling to transition their creaking energy grids to become both cleaner and also manage the tricky transition to carbon neutral sources of generation.

That isn't all however.

It is far more than a problem for countries like Germany (or in the irony of ironies, states like California!) that have made overly ambitious - some might say foolish - assumptions about the viability of certain green renewable power technologies.

Those countries or states or regions are suddenly in a very tough spot. Electricity demand is rising and the cost of providing that electricity is falling in real (inflation adjusted terms) but you have to have the right policy mix

What to do about this?

The easy realization is to realize that owning Nvidia (builder of chips - i.e.: shovels) or Microsoft (builder of data centers, i.e.: the mountains where gold is prospected) isn't enough. You don't just want to own the shovel makers you also want to own those able to supply power and also, most crucially, those who own the already built data centers.

It is rapidly going to become true that there is still plenty of demand for AI capabilities but an increasing arms race to secure the land, the water and the power to build the basic infrastructure to make those possible.

The hard part is answering the question: what does that look like??

There are three possibilities from our perspective:

The first is obviously to own the companies that specialize in building these data centers. Yes, there are companies like Microsoft that own and run what is inside but there are also companies that actually build and lease out the facilities themselves.

Some examples are: Equinix or Digital Realty Trust.​

The second is you can purchase companies that service or sell goods to those data centers. Obviously, these companies are having a great time of it.

Some examples here are: Vertiv or Hubbell incorporated.​

Lastly, is most difficult to express at present but it brings together a long running and favorite Pebble theme: you can buy the stocks of the companies most likely to gain from greater energy demand.

There are lots of utility stocks - and you can just buy the whole sector for very cheap - or you could also buy the companies and entities that are most likely to be able to supply rapidly increasing energy demand.

That means nuclear.

There are only so many options that can help an economy make a significant quantum leap forward in power generation.

It is a bit speculative but we wonder whether the desperate need to compete in artificial intelligence and also simply the basic baseline power needs will finally be the thing that unlocks nuclear energy investment in the West again?

The tough thing about this view is it is VERY hard to see nuclear power suddenly becoming the go to solution to more power demand in the US. That could change! Things change all the time.

The positive is that neither the US nor the broader West necessarily needs to partake for the nuclear thesis to play out. Plenty of other countries will be watching this play out and will be mindful that to compete they don't just need sophisticated models and semiconductor chips but also the practical capabilities of actually running them somewhere.

Those constraints need to be overcome. Other countries and societies are less picky about how they power their economies and take a more realistic view of the challenges of climate change as well.

This could play out lots of ways. Renewables will almost certainly play a role as will more experimental forms of power such as geothermal and hydrogen but it also seems like nuclear energy will be a big part of the story,

If you want to read more about our nuclear energy arguments you can find the two most recent here and here.

The long and the short of it is that uranium and the various companies that mine it like Cameco are doing exceptionally well.

Energy is getting cheaper. Energy demand is also taking off. Something has to break and it will hopefully be the long moratorium on investing in nuclear energy around the world.

*******

Have questions? Care to find out more? Feel free to Download our App (!!) or reach out at contact@pebble.finance or join our Slack community to meet more like-minded individuals and see what we are talking about today. All are welcome. You can also get our newsletter as an RSS feed.