Nuclear Renaissance, attempt 2: Electric etc
In somewhat surprising news in the prior weeks, Amazon has just confirmed a funding agreement with both Talen Energy and X-Energy to power some data centres, following in the footsteps of Google (purchasing a power agreement from Kairos power), and Microsoft's deal to restart Three Mile Island (or at least, the half that remains solid). They're a trio, they've chosen entertaingly different approaches - X-Energy experimenting with gas-cooled, Kairos with molten salt, and Microsoft reusing existing infrastructure. It all sounds a bit Top Gear if I'm honest. For what it's worth, the UK govt is advancing in its tender process, with Rolls-Royce doing well in Czechia.
There are a few things surprising. For one, companies like MS have previously been buying power agreements from wind farms. The pros and cons of low-carbon tech should be well-worn - both solar and wind are much cheaper than nuclear, uncontroversially, as well as being less controversial. The downside, one I will harp on about ad infinitum, is that they deliver power on their schedule not yours, and the cost of storing enough kWh to avoid some form of load shedding is simply laughable. Quite possibly always will be.
(The other issue is that MS et. al have been struggling to actually connect these farms directly, especially to data centres already built near thermal plants. That surprises me, I'd usually put transmission as one of the shibboleths of real engineers - this stuff is solved, stop messing about with local generation, you don't have spend your brain cycles worrying about it. Well, apparently now you do.)
So, the only way to make renewables work is by "demand shifting" and "smart meters", both euphemisms for "well we just won't deliver power, but fancy tech will make that sound like it's your choice somehow". The race has been on to find workloads which can shift a lot, to accommodate the vagaries of the weather. That's interesting, because all this datacentre power was supposedly for AI, and AI training seems very shiftable. As do quite a lot of IT workloads, really. Delay syncing a backup, or some render or scientific compute? Fine. Delay part of a pharmaceutical or chemical process a few hours, you have at best turned an entire batch into hazardous waste. You are, to use a technical term, buggered. Go to a chemical conference in Europe these days and it's like attending a wake.
So I don't know why the tech boys are the ones funding this, sorry. My best guess is that FANGs just have that much money, that throwing around billions just for a reliable PSU is worth it. And a nuclear powered city-block artificial brain does, admittedly, sound extremely cyberpunk, possibly a bit Adams-esque. And maybe I'm miscalculating the ratio of demand from training vs less predictable service (inference) type workloads.
On the supply side, we finally have some money and some small companies - even nuclear proponents can't really wholly blame regulation for the decades of stagnation. Dr. Keefer seems to have summed it up in Decouple - big companies, Westinghouse and EDF, have just flat out seemed uninterested in selling their atomic wares. Westinghouse apparently didn't have a sales team, EDF practically told the May-era goverment to sod off and that they were doing them a favour. Having companies now appear interested is a big positive. Banks are also interested, and that's really good.
However, let's be realistic. New approvals are still hard, and if I'm honest, molten radioactive fluoride doesn't sound all that fun to me. (Yes I know the water in a boring old PWR would still kill me very quickly if it escaped, and it needs some chunky steel to contain, but still... it feels like it turns a lot of "one in a billion" Really Bad Things into "one in a million"). And startups generally don't do research. Molten salt needs research. If, frankly, it's ever practical. Still, I wish them luck. Could be worse, could be a gas. Oh.
And there will be a financial winter for these projects as any, both from the AI trough of dillusionment when it comes, and, I suspect, enough AI efficiency improvements. Or a new President who's more oil-friendly. When? Who knows, but it'll be there. They have some firm funding now and that's great, but those customers may well wander, so move fast.
How hard can it be?