Google, Amazon buy nonexistent mini nuclear reactors for AI data centers

BRING IT ON NITPICKY NUKE NERDS

Well acthtually we prefer to be called fission/fusion nerds

"greenwashing is cheaper than action" indeed. (edit2) On that note, storytime about the clownshow that is Dutch politics. So our radical right wing government is pro nuclear power, of course, and they want to build more powerplants. So what are they planning on doing? They are going to start a study on which locations are best. Which is maddening, as these studies have already been done before (so it prob is just an attempt to hopefully have the study finish when it isn't them in power anymore so they are not at risk of starting an too expensive megaproject). But it gets worse, the absolute clowns of our farmers party just went 'fuck the studies' and they just pointed at a province where there are a lot of farmers and went 'we will put a powerplant there'. And this is how they discovered nuclear powerplants need running water and they picked one of the areas without a major river. ('im ignoring the clownshow re 'the immigration crisis' (not a crisis) as this post is already too long, and there is a big risk of honk overdose if I go into that).

Our local Swedish right-wingers in gov have a chubby for nukes too[1], because their main motivation besides hating on brown people is pissing off Greens. But in the Swedish way they handed this off to a researcher ("utredning") who found out that to get the industry on board you need a) rock-solid political promises (so need to get the Social Democrats at least on board) and b) have a price guarantee for power for at least a decade, along with massive government loan guarantees.

It's gonna be hard to get voters interested in 10 new reactor sites (NIMBY gets supercharged when it comes to nukes) if it slightly pushes up lending rates and power bills.

[1] the right-wing part of the opposition social democrats like them too to be fair
wouldn't it make more sense to put NPP on seashore and just dump waste heat to ocean like everyone else does
- Iirc at the sea in friesland was one of the options yes. But the farmer option was also landlocked, de achterhoek if you want to look it up.

I'm looking forward to seeing the tech attitude of "move fast and break things" being brought to nuclear reactors.

What could possibly go wrong? Better ask for forgiveness then plant permissions!

how come all these botfans talking points are just repackaged cryptobros talking points

it keeps happening

you'd almost think they were literally the same fucking guys
- coincidence?? yeah probably
just one more SMR bro

If these nuclear plants manage to come to fruition, it'll be the sole miniscule silver lining of the bubble. Considering its AI, though, I expect they'll probably suffer some kind of horrific Chernobyl-grade accident which kills nuclear power for good, because we can't have nice things when there's AI involved.

even if you're ardently pro-nuclear, SMRs are just a failure purely on the economics and always have been. And that's before wind/solar/battery made them just obsolete. So SMRs are the perfect tech when you don't want to do anything useful.
- See, I feel like AI might have the actual solution to this problem. We can overcome the economic issues with setting up SMR infrastructure the same way AI has powered through all their economic problems: setting VC money in fire and trust that the smokescreen will hold out for another funding round.
  
  Once the reactors exist, I'm assuming that their operation can be relatively cheap for whoever ends up owning the actual plants once the AI bubble pops and the datacenters around them are shut down or repurposed.

BRB making a video for a cold fusion based kickstarter. I smell money

Remember to say it's optimized for net zero datacenter operations
- few months late, it's called helion energy

I don't claim to be an expert on nuclear power, so take what I say with a grain of salt, but from what I've seen, smaller reactors don't seem to make much sense. The trend seems to be towards bigger reactors with bigger power output. Some of it thanks to the bureaucracy of getting permits per reactor, but also the physics, engineering, real estate and economics involved. Conventional (i.e. existent) reactors are typically a fairly small part of a nuclear power plant's footprint, so no matter how much you miniaturize them you will have the overhead of security, operations, cooling and electrical infrastucture.

If someone can fill me in on the benefits of smaller, more modular nuclear reactors and how they might outweight those of large installations, I'm interested.

square-cube law is in full force there

one argument in favour of SMRs i've seen is that while less efficient than regular sized reactors, these are cheaper per unit (but not for MW) so some of them can be built earlier than bigger reactors. which doesn't matter because these things don't exist
The hypothetical benefit is that prefabricated parts are a lot less dependent on the site. This will make the reactor cheaper to build.

There's also a perception sleight of hand - "modular" doesn't mean the reactor is a module you ship in on a big truck, put some uranium in and away you go. You're building a power station in a fixed location.

Also you still need a shitload of water.

Google has signed a deal with California startup Kairos Power for six or seven small modular reactors. The first is due in 2030

So, well after the bubble is going to pop.

Oh no, they created the means to generate non-fossil-fuel energy for nothing /s
- Who knew that the only thing stopping nuclear power, the most morally and environmentally correct power source (uranium is only produced by popes shitting in the woods), was that Google and Amazon hadn’t thrown money in the direction of Chernobyl first. It was so simple this whole time. Now it’s solved and I can go back to gaming.

I am 100% in support of deploying nukes to Google and Amazon

Thanks for posting this good collection of links. HN has as hard-on for SMRs and as a first-order approximation that means they're wrong, but it's good to have something more than vibes backing it up.

I got a lot of them from very pro-nuclear guys too, but ones who can count

I swear they looked at Bill Gates failing to launch SMRs and thought: "he's a smart guy"

Of all the things that will never happen, this is the one that will never happen the most.

I give it slightly higher odds than AGI.

Edit: or cryptocurrency replacing fiat

At least is technically feasible (although completely impossible to do in that timeframe)

Unlike the cold fusion energy deal that Microsoft greenwashed last year that's pure science fiction (invent, create, test and build a cold fusion reactor in just 4 years: impossible unless they got a time machine or found some alien tech in a remote cave)

We buy tech of the future that might or might not work/get government approval/make actual sense to build so we can't be blamed for ruining the climate with AI using up all the energy. That's the more earth bound version of: I don't care about climate change because we will live on Mars soon^(tm).

Honestly thank god they're vaporware. Somehow I don't think we should have startups building actual nuclear reactors.

500 megawatts? Pshhj that's not even though to power the flux capacitor.

I can easily get 500 megawatts with a few power towers burning rocket fuel, plus I don't have to worry about the logistics of recycling uranium and plutonium waste.

I'm unrelated news, the Satisfactory 1.0 update is pretty great. What are we talking about again? Oh sorry gotta go build another heavy modular frames factory
- this game destroyed my life for weeks. I'm doing shapez 2 now and I'll probably finish up my miserable tour of factory games with factorio afterward
  
  EDIT: also rocket fuel seems extremely overpowered to me, I don't think powering tier 9 should be as trivial as making some RF and slamming down fuel plants for ten minutes
- I'm more of a factorio guy myself. I'm going to become a hermit next monday

Alright so you can have them funding the next generation of nuclear power, which would eventually bring this new form into the mainstream by having them deal with the costs associated with ironing out any issues they have and very likely making it economically viable…

Or…

These tech companies can use fossil fuels to power their AI. Like it or not, they arent going to stop developing AI and data centers. They need the power either way. Solar and wind won’t keep up with that level of demand and tech companies know it. So choose. Nuclear, or fossil fuels?

as you will discover if you read, the use case for SMRs is to keep doing coal
- Reading the linked article? But it is 1/16th a ssc post long! Nobody got time for that.
- as you will discover if you read,
  
  Hey, man, that's a lot to ask, you know, brah
- The more I read, the more I'm convinced SMRs are to clean energy as gadgetbahns are to public transit.
Like it or not, they arent going to stop developing AI and data centers.

Well they should. I'm not giving them credit for investing in vaporware nuclear plants when the ostensible plan is to waste all the power on glue pizza recipes.

I wish they at least put that money in real and known working designs available right now so at least when the fad is dead, we can maybe use that power for something else. Or they can maybe have the tiniest decency to unfuck their search engine or whatever.
- Well they should. I’m not giving them credit for investing in vaporware nuclear plants when the ostensible plan is to waste all the power on glue pizza recipes.
  
  man, fuck this timeline... what are we even doing
You're correct that I can't stop them from making pants on head stupid decisions, but I'm not going to stop making fun of them.

very likely making it economically viable…

They're going to fund currently economically nonviable nuclear plants to power their currently economically nonviable genAI schemes? Over the time horizon of ~~25 years~~ a decade (edit: misread the article) before they scale up energy capacity at all past the rnd stage? Maybe pants on head is too generous.
- I’m not crunching the numbers over here, but they must be making money off AI if they’re doing this. I’m sure they have further plans that aren’t public yet.
  
  I’m not a fan of AI, so if Microsoft or Google ends up in a dumpster fire because of all this I will never stop laughing about it. I just don’t expect it.
Try option three. No one is going to pay for any of that because LLMs are useless machines.

Fun Fact: it took 42 years to start Watts Bar Unit 2.

hol up google wants 500MW? that's 1. one regular sized reactor that could be delivered in the same timeframe, and 2. google uses already almost 3GW on average (2023), this is compared to about 2.5GW for ms of which something in the ballpark of 700MW just for ai. they're gonna need much more, like five regular sized reactors if they want to use entire baseload (that's how NPPs work best. the french made load-following NPPs but i guess it'd be harder to make them small) or swing wildly with power consumption to conform to renewables

@dgerard somebody should found a startup and sell ASMRs to Google and Amazon

I got into the wrong business.

didn't we all

i sometimes wonder how i'd make money if i was unencumbered by ethics. i originally thought audiophiles, but then i discovered crypto, holy shit

ai is the same

This is promising. The high energy demands of AI has big tech investing in the field of small modular reactors (SMR).

how do you do, fellow kid. tell me more about the field of shitty muclear reactors (SMR)
- Criticize all you’d like. I just think it’s better to let corporations foot the research, especially if it turns out to be a fruitless attempt at progress.