I want my RPV!

So, yet another different format to the newsletter, but again the best way for me to educate people concerning some of the big issues connected to AI

The Main Event

Those of you that have read my previous weekly newsletters, will be aware that I have expressed concern around the topic of AI energy supply. We’ve all read about the billions being invested in data and computing centres, and I have mentioned a number of multi-billion dollar deals that people like Meta and others have signed with power generating companies.

However, the issue is that energy production in the USA is not keeping pace with the demand from AI in particular, with projections suggesting that AI compute demand will exceed approximately 12% of all US energy production by 2030.

There are some wider issues to keep in mind, such as the simple rules of supply and demand meaning that electricity prices are likely to increase for everyone in the USA. This has massive political ramifications, given the impact on inflation and general day to day living costs – it’s not just household electricity costs, but also the electricity needed by other industry, all of which potentially feeds into the inflationary cycle.

The big issue though is how do you address this problem. There’s been lots of talk about the idea of “micro-reactors”, with the suggestion that every large data centre could end up eventually with their own mini nuclear power plant.

Even setting aside the obvious security concerns about having chunks of nuclear material spread around the country, and the almost certain objection of local communities, the big issue is the scales required. Rolls Royce and others are currently designing and building such mini-reactors, but their anticipated outputs don’t come anywhere near the levels required.

Arguably, the only way to address this is to go “old school” and build full-sized nuclear reactor plants. However, this presents two big problems. The first is that building even one takes a long time – think 15 – 20 years. The bigger problem though is the key ingredient that is required by all of them, namely an RPV ingot.

You’re probably now asking yourself the same question I did recently, which is what on earth is an RPV ingot. Well, it’s a very large, ultra-high-quality steel block that serves as the starting material for manufacturing reactor pressure vessels, the thick-walled steel containers that house the nuclear reactor core.

I won’t bore you will all of the technical detail, other than to say that they’re not easy to make, there’s a lead-time of 5 – 10 years to get one … and none of them are made in the USA. Currently they are only made in Japan, South Korea, France, Russia and China. There is a steel plan in the UK that makes them, although there are outstanding questions around certification – important given the role it plays.

All of this makes the situation very challenging for the US, given the geo-political dramas currently unfolding. I’m not judging or commenting, just considering the issue from a risk management perspective.

As I’ve said before, the issue of AI infrastructure, including access to water (for cooling) and electricity for power, is the next problem that needs to be addressed if the full potential of AI for change is going to be realised.

Stay informed, stay critical, and wherever possible - stay ahead.

Regards

Tom Carter