Modern AI data centers consume enormous amounts of power, and it looks like they will get even more power-hungry in the coming years as companies like Google, Microsoft, Meta, and OpenAI strive towards artificial general intelligence (AGI). Oracle has already outlined plans to use nuclear power plants for its 1-gigawatt datacenters. It looks like Microsoft plans to do the same as it just inked a deal to restart a nuclear power plant to feed its data centers, reports Bloomberg.
Are we eventually gonna get more fusion because billionaires are demanding more energy for their stupid projects?
Sure, knock yourselves out.
Are we eventually gonna get more fusion […]
Either you mean fission, or the “more” could be omitted.
Yeah, too bad there’s no long-term storage for the waste so it will mean more and more leaks polluting land for centuries since the power companies will just go bankrupt when it’s time to do anything about it like with most forms of pollution.
Most likely more fossil fuels because they’re faster and cheaper to roll out.
This is how I’m able to sleep without worrying about death, one of these billionaires has got to be funding research so they can live forever. No guarantee they’ll share but that’s at least a less dread inducing issue.
Lol. I just love it how so many people complain that Nuclear doesnt make financial sense, and then the most financially motivated companies just actually figure out that using a nuclear reactor completely privately is best.
Fuck sake, world.
It has been operated privately for a long time, unit 1 (this one) being operated by constellation energy. It stopped in 2019 because Methane had undercut it, and MS has now made an agreement to buy 100% of unit 1s output, but they aren’t buying the facility. Most power generation in the US is private, for better or worse (usually worse).
Microsoft jumped fully on the AI hype bandwagon with their partnership in OpenAI and their strategy of forcing GenAI down our throats. Instead of realizing that GenAI is not much more than a novel parlor trick that can’t really solve problems, they are now fully committing.
Microsoft invested $1 billion in OpenAI, and reactivating 3 Mile Island is estimated at $1.6 billion. And any return on these investments are not guaranteed. Generally, GenAI is failing to live up to its promises and there is hardly any GenAI use case that actually makes money.
This actually has the potential of greatly damaging Microsoft, so I wouldn’t say all their decisions are financially rational and sound.
On the other hand, if they ever admit the whole genAI thing doesn’t work, they could just sell the electricity produced by the plant.
if they ever admit the whole genAI thing doesn’t work
. . . The entire multi-billion-dollar hype train goes off the cliff. All the executives that backed it look like clowns, the layoffs come back to bite them - hard - and Microsoft wont recover for a decade.
I mean . . . a boy can dream
The workers will take the blunt of the executive’s mismanagement.
Unless they unionize
I mean, even with a union, if a company crumbles the laborers are out of a job.
My org’s Microsoft reps gave a demo of their upcoming copilot 365 stuff. It can summarize an email chain, use the transcript of a teams meeting to write a report, generate a PowerPoint of the key parts of that report, and write python code that generates charts and whatnot in excel. Assuming it works as advertised, this is going to be really big in offices. All of that would save a ton of time.
Keep in mind that that was a demo to sell Copilot.
The issue that I’ve got with GenAI is that it has no expert knowledge in your field, knows nothing of your organization, your processes, your products or your problems. It might miss something important and it’s your responsibility to review the output. It also makes stuff up instead of admitting not knowing, gives you different answers for the same prompt, and forgets everything when you exhaust the context window.
So if I’ve got emails full of fluff it might work, but if you’ve got requirements from your client or some regulation you need to implement you’ll have to review the output. And then what’s the point?
Keep in mind that that was a demo to sell Copilot.
And whether it works as well as they described remains to be seen. However, they did prove that there’s a legitimate use case for generative AI in the office, in most offices. It’s not just a toy.
Honestly it seems crazy that companies that are so focused on short-term profits in 2024 would be able to make nuclear work.
Every once in a while they get faced with a line on a chart somewhere so unbelievably vertical that they have no choice but to look beyond next quarter. Power consumption going 10x in 2 years is one of those times.
Nuclear safety and penny-pinchers don’t make good bedfellows.
Nuclear safety and
penny-pincherscapitalism don’t make good bedfellows.ftfy. Possibly ironically, nuclear safety and communism (or totalitarianism) don’t work either. It’s odd, innit.
Pretty sure it has to do with how the plant is designed and operated as opposed to what economic or governmental system it happens to exist under.
Doesn’t that design and operation get created by the economic or governmental system it’s under?
I think with the USSR at least, that their reactor designs were supposed to be less safe than western reactor designs.
Was it because they were a shitty oligarchy claiming to be communist? Maybe, they did make a lot of garbage decisions.
I think the US has the record for most nuclear disasters by a lot but two of the worst were in the USSR.
They were actually designed to be very safe. It was thought that they literally couldn’t fail dangerously. Chernobyl was a huge fluke (that had preventions put in place to ensure it never happened again) that was just a lot of weird things combining at once. The other reactors at Chernobyl continued operating for decades safely, similarly to three mile island which only stopped on 2019 because it wasn’t profitable, but now it appears it is again. Both of these nations (and child nations/successors) continued to operate many more nuclear plants without issues. Nuclear is, by far, the safest energy source, including green energy like solar.
How exactly is nuclear energy safer than solar energy?
Yeah for sure it is cheaper, if they only have to pay the operational costs. Not the ones of building and decomissioning the plant. Lol.
It doesn’t make financial sense to build new nuclear power plants. They’re hugely expensive and such projects routinely run well over time as well as budget. If it did make sense, Microsoft would be building them, instead of reviving the site of one of the worst nuclear disasters in the US. Thing is, they want lots of power, and they want it yesterday. By the time you can build a new nuclear plant to satisfy these needs, AI will have run its course and big tech will be on to the next scam.
But hey, why pay attention to such nuances?
The fact that they want to buy an old nuclear reactor instead of building a new one should be all you need to know to realise that it’s not financially viable.
It’s not quite equivalent right? Using an existing plant is cheaper and faster than building a new one?
Its like saying a datacenter is not financially viable only because top brass decided to use a perfectly good existing one.
deleted by creator
No, that’s only because the US has constructed barriers to make it cost more and take longer, to protect conventional dirty energy. Those barriers do not need to be as large. A new reactor being built would take several years, and they don’t want to wait for that. That doesn’t mean it wouldn’t be profitable, although again the barriers may make it unprofitable or at least a riskier investment.
Edit: also, they aren’t buying this reactor. They are not in the energy business. They’re buying 100% of the output of unit 1. That’s all. The previous owners are still running it. It stopped temporarily in 2019 because Methane undercut it, because Methane does not have to pay for its pollution like nuclear does.
Three Mile Island is the epitome of
conventional dirty energy
I see this as a good thing because they’ll invest more on making energy efficient. That’s something bound to trickle down and help poorer regions unless they die off first.
I’m firmly in the “building new nuclear doesn’t make financial sense” camp, but I do think that extending the life of any existing nuclear plant does. Restarting a previously operational nuclear plant lies somewhere in between.
I think when you start looking at how expensive other forms of green energy are (like wind) long term, nuclear looks really good. Short term, yeah it’s expensive, but we need long term solutions.
I don’t think that math works out, even when looking over the entire 70+ year life cycle of a nuclear reactor. When it costs $35 billion to build two 1MW reactors, even if it will last 70 years, the construction cost being amortized over every year or every megawatt hour generated is still really expensive, especially when accounting for interest.
And it bakes in that huge cost irreversibly up front, so any future improvements will only make the existing plant less competitive. Wind and solar and geothermal and maybe even fusion will get cheaper over time, but a nuclear plant with most of its costs up front can’t. 70 years is a long time to commit to something.
Can you explain how wind and solar get cheaper over time? Especially wind, those blades have to be replaced fairly often and they are expensive.
Wind and solar also have to be paired with either cheap natural gas or energy storage systems that are often monstrously expensive. Unfortunately these numbers are almost always left out when one discusses prices.
People do appreciate the lights staying on, after all.
Yeah, we haven’t even gotten into the reliability. The have dead times where no output is created that nuclear doesn’t suffer from.
With nuclear, you’re talking about spending money today in year zero to get a nuclear plant built between years 5-10, and operation from years 11-85.
With solar or wind, you’re talking about spending money today to get generation online in year 1, and then another totally separate decision in year 25, then another in year 50, and then another in year 75.
So the comparison isn’t just 2025 nuclear technology versus 2025 solar technology. It’s also 2025 nuclear versus 2075 solar tech. When comparing that entire 75-year lifespan, you’re competing with technology that hasn’t been invented yet.
Let’s take Commanche Peak, a nuclear plant in Texas that went online in 1990. At that time, solar panels cost about $10 per watt in 2022 dollars. By 2022, the price was down to $0.26 per watt. But Commanche Peak is going to keep operating, and trying to compete with the latest and greatest, for the entire 70+ year lifespan of the nuclear plant. If 1990 nuclear plants aren’t competitive with 2024 solar panels, why do we believe that 2030 nuclear plants will be competitive with 2060 solar panels or wind turbines?
Have they solved the disposal questions?
We haven’t solved the “disposal” question of using fossil fuels, and those turned out (or were known along) to cause much bigger problems.
Mostly, yes. Use breeder reactors to turn long term radioactive waste to sort term radioactive waste, store for short time and done. The downside: it’s more expensive to move and process the stuff so nobody wants to do that.
We have, but of course not to the satisfaction of anti-nuclear activists because solving it would be counter to their actual goals.
Nuclear waste is actually quite easy to deal with unless your purposes are best served by it being very difficult to deal with, in which case you make as much trouble as you can for it.
Like most things with environmental impact, we just let later generations deal with it. Somehow.
Uh… Yeah. The reactor was in operation until 2019 when it stopped being profitable. Disposal was never a problem.
Personally? I don’t think this is a bad idea. The less they drain from the grid, the less they consume fossil fuel.
The reactor isn’t active right now, and they are a PWR design, and like the 1979 incident showed, they do fail safely.
So long as Microsoft pays for the operation of the plant? Seems reasonable to me if they’re going to consume an assload of energy with or without public support.
we could use that extra energy to offset a bunch of existing carbon emissions now. This is still waste. If it’s going to be started up again, and its energy used for something useless, it’s waste.
Microsoft would do it with or without the power plant. Make no mistake about that.
The same argument could be said if they made a 1GW solar farm, or any other form of power generation. Unless you have a way to legislatively prevent Microsoft from producing their own energy or prevent acquisition of decommissioned plants, I don’t see how you can prevent waste.
That argument presupposes that the reactor would otherwise be brought back into operation, which I don’t think is necessarily the case.
Is it going to be started up again?
If M$ doesn’t invest into this for their own purposes, is it still going to be started up? Or is your position that M$ should be investing in a nuclear power plant for the good of the world?
Because while I can agree with the idea, we all know that would never happen. So if it was never going to be started up again, we are at 0 gain or loss no matter what they do with it.
And that’s ignoring the fact that they are apparently intending on using that energy anyway.
it would be a missed opportunity in the sense of “if they can allow it to be turned it back on to waste its power on this dead-end tech, why couldn’t it have been allowed to operate again (earlier) for reasons we actually need?”
I’m not putting the blame on microsoft here, even though it might seem that way. But it’s not microsoft who need to give the go-ahead for this to happen. It’s the higher ups who decided to give the capacity to microsoft.
Yes it was still going to be used, but they could have been paying out the ass for it, which could fund other projects.
It operated for a long time profitably. It ceased operations in 2019 because it became unprofitable, largely because Methane undercut it. Methane should cost a lot more, but they don’t have to pay for negative externalities. Nuclear has to contain all of its waste, and handle it carefully.
as opposed to just spewing it out in the air? (carbon 14 is a thing, those things emit a lot more radioactivity to the environment)
I’m not saying they should not contain it. I’m saying other sources should have to. We only force one energy source has to pay for the cost of all of its waste. Why is that? It’s only to the benefit of dirty energy.
If there were plans for it to be used, then I’m with you. But if I’m being honest, I’d put money on the original plan consisting of letting it sit there for decades to come without being used.
And “paying out the ass” is what they will likely be doing, just to the private corpos that own the plant. It’s not government run, the money would never circle back to taxpayers beyond normal taxation.
that’s what I’m complaining about. If there can be plans now, why was the original plan just “let it rot”?
Greed? ¯\_(ツ)_/¯
I don’t think you’re going to get the answer you want here. But I’d be willing to bet M$ is dropping the $$$ for whatever retrofits and repairs need to be done, with the agreement being they get the power near cost for a set duration.
Obviously that’s speculation on my part, but would explain the situation quite cleanly.
I remember I had to do the 3 mile Island incident as part of my university degree. Apparently one of the biggest problems was that the control interface was hard to understand for the human operators.
So I guess if they just replaced the control system with a modern computer that would fix most of the problems. Obviously not a Windows system, otherwise we’ve just got the same issue all over again.
It was the SCADA view right? A lot of SCADA software is basically running on top of windows, though you typically would never see the desktop. Ignition at least is cross platform, but that is because the server is Java and Jython. A big part of why things are running on windows is due to OPC, which was traditionally all DOM and .NET. It is basically a standard communications protocol and is what allows your HMI/SCADA to communicate with PLCs. Otherwise, you use proprietary drivers and native PLC specific protocols.
SCADA programming/design is kind of an art and is usually written by an either an overworked engineer or someone who had far too much time on their hands. You basically build screens using specialized software, hook up buttons and UI elements to PLC signals, and pass some signals from the UI to the PLC. They are all heading in the Edge/iot/cloud/web based/techno-babble direction these days…
Ignition, programming software is free!: https://inductiveautomation.com
Some other random ones I have seen or used in the past: https://www.siemens.com/global/en/products/automation/simatic-hmi/wincc-unified.html https://www.aveva.com/en/products/intouch-hmi/ https://www.rockwellautomation.com/en-us/products/software/factorytalk/operationsuite/view.html
“is usually written by an over worked engineer”
I’m in this post and I don’t like it.
But really these scada systems are rarely well defined by the time implementation happens. Often the architect has a great plan, but by the time it’s passed to a manager, a non-software engineer, to the product engineer to the automation team to the contractor the end result is “X data is pushed in With Y form and we use either a,b,or c date time stamp any nobody knows”
It continued operating for decades after the event. I’m sure they already solved that issue. It can still be improved I’m sure though.
So I guess if they just replaced the control system with a modern computer that would fix most of the problems
Introducing new Clippy For Reactors.
“It looks like you are trying to prevent a nuclear meltdown. I can help with that.”
I had to do a double take to make sure this wasn’t an onion article.
especially since just reading the headline made me cry.
I’m sure that everyone will recognize that this was a great idea in a couple of years when generative LLM AI goes the way of the NFT.
Honestly, it probably is a great idea regardless. The plant operated for a very long time profitably. I’m sure it can again with some maintenance and upgrades. People only know three mile island for the (not so disastrous) disaster, but the rest of the plant operated for decades after without any issues.
with some maintenance and upgrades.
Hopefully we can trust these tech bros to do that properly and without using their usual “move fast and break things” approach.
They are only buying 100% of the output. The old owners are still owning and operating it.
And if they do skimp on maintenance and upgrades and the plant melts down, we can be assured that no harm will come to the company because the scale of the disaster would wipe them out and they’re “too big to fail.”
It’s one of a hell of an old nuclear plant if it’s the original three mile island one.
It is, yeah. It was in operation until 2019.
LLMs have real uses, even if they’re being overhyped right now. Even if they do fail, though, more nuclear power is a great outcome
Once operational, the energy generated is cheap and will still be in demand
Nfts were a scam from the start something that has no actual purpose utility or value being given value through hype.
Generative AI is very different. In my honest opinion you have to have your head in the sand if you don’t believe that AI is only going to incrementally improve and expand in capabilities. Just like it has year over year for the last 5 to 10 years. And just like for the last decade it continues to solve more and more real-world problems in increasingly effective manners.
It isn’t just constrained to llms either.
The creators who made the LLM boom said they cannot improve it any more with the current technique due to diminishing returns.
It’s worthless in its current state.
Should be dying out faster imo.
That’s one groups opinion, we still see improving LLMs I’m sure they will continue to improve and be adapted for whatever future use we need them. I mean I personally find them great in their current state for what I use them for
What skin do you have in this game? Leading industry experts, who btw want to SELL IT TO YOU, told you it has hit a ceiling. Why do you refute it so much? Let it die, we will all be better off.
Even if it didn’t improve further there are still uses for LLMs we have today. That’s only one kind of AI as well, the kind that makes all the images and videos is completely separate. That has come on a long way too.
I made this chart for you:
------ Expectations for AI
----- LLM’s actual usefulness
----- What I think if it
----- LLM’ usefulness after accounting for costs
One of the major problems with LLMs is it’s a “boom”. People are rightfully soured on them as a concept because jackasses trying to make money lie about their capabilities and utility – never mind the ethics of obtaining the datasets used to train them.
They’re absolutely limited, flawed, and there are better solutions for most problems … but beyond the bullshit LLMs are a useful tool for some problems and they’re not going away.
I cannot think of one single application where an LLM is better or even equivalent than having a person do the job. Its real only use is to trade human workers for cheaper but inferior output, at the detriment to mankind as a whole because we have in excess labor and in shortage power.
So they fixed the leak?
Different reactor. Unit 2 partially melted down, there’s no turning it back on. Unit 1 continued running after Unit 2’s failure, and was only shut down because it became economically unviable
Oic
Ironically, the power hungriness of AI might actually do good for the environment if it normalizes nuclear energy.
Quite the twist
This sounds like the intro to a bad post-apocalypse sci-fi movie.
I think pre post-apocalypse is just the apocalypse. If you read the news these days that sounds like a pretty accurate description of the time we’re living in. We’re all just pretending it hasn’t started yet.
Turns out planetary extinction without an asteroid is slow AF.
Holy sunk cost fallacy, batman. How fucking much does it cost to operate an ENTIRE GODDAMN NUCLEAR REACTOR just to fuel a tech project that nobody wants???
that nobody wants
lol
Investors want it, because they want to ride the wave towards profit. It doesn’t matter if it’s good, sustainable or not. That is what matters.
A lot of the cost is building a giant centralized nuclear facility. Once they are built it is not nearly as expensive to run them.
I think this is generally a good thing. Companies should be thinking of ways to supply their power needs.
Having said that, people want a good AI. The LLMs they are working on are probably not that. I am very skeptical we are anywhere close to where the hype train has taken us
Just because you don’t want it doesn’t mean others don’t.
And just because you don’t know much about the actual tech product itself doesn’t mean that it’s as narrow as you consider it to be.
There is a ton of vapid hype that everyone including myself is getting sick and tired of. I’m more than happy to recognize that. However, there are still real world problems and continued advancements being made daily.
It’s not all about LLMs either, there are many other types of science being done to develop improve and augment various other flavors of artificial intelligence. This has been a pretty constant trend for at least the last 10 years, we’ve just had a recent explosion in language capabilities with the introduction of generative AI. Thus fueling the hype.
That’s a really weird stance that I keep seeing on here which is to be proud of being ignorant. Being proud of hating something without actually understanding what it is. Being proud of not knowing how something works so that you can be more contrare.
I just hope this deal doesn’t involve using their AI to monitor the reactor …
There actually has been good work on using AI to control fusion plasmas its at the point where it can keep them stable significantly better than any human or simple automated system.
You know, that actually makes sense. Fusion is so energetic and probabilistic in nature, plus it’s effectively “charged fluid dynamics” and there are an impossible number of variables to handle. That’s literally the kind of shit AI is great at.
Fission though? Not so much
No, stick rod in / pull rod out doesn’t really need deep learning to make work well :p
Apparently, I didn’t learn that with my ex
Yes in a research lab. Here we’re talking about Microsoft.
Have you ever used something they made? Did it meet your standard of being “good work”? No. It’s a greedy, soulless cash grab disguised as software that infects the entire organization and disables common sense.
M$ actually running a nuclear plant is a guaranteed disaster. Blue Screen of Death.
Have you ever used something they made? Did it meet your standard of being “good work”?
I mean you’re ignorance of the products that they build or work on doesn’t precipitate their badness. Let’s start with the entire developer ecosystem that they have their hands deep in, it’s a pretty damn good ecosystem.
You probably need to check your bias because it’s leaking, negatively affecting your decision making.
Any company of this size is going to have shit products great products and literally thousands and tens of thousands of projects in between. You seem to be familiar with one product line, of hundreds or even thousands.
*your*
And how long have you watched Microsoft do what they do? Is it more than five years? 10? 20? 30? Really?
You can genuflect to the Ballmers all you want but their less-sucky products are still the best garbage one of the most profitable businesses in the history of earth can provide.
I’m not talking about UX, or code that doesn’t crash. I’m talking about that and the purpose of it. Good software serves the purpose of the user. Micro$oft software serves Micro$oft primarily - and often exclusively. Then there’s the monopolistic practice they have of buying innovation and crushing it beneath their heel.
Whatever “ignorance” of their products I have has been very deliberately cultivated. Glad you like VS, you’ll benefit Microsoft through it in at least one, or many, ways.
And none of what they do is different from how they’ve always been. Their self-serving greed was on full display from the day Bill Gates started demanding payment for his software that others had been freely sharing with him. Feel free to remain in thrall to them.
They’re not running it, though. They’re making a deal with the owners (an actual nuclear power company) who are going to run it themselves and sell Microsoft the output.
I am all for nuclear power, but I’d rather it be from modern reactor designs and builds, and I’d rather it not be wasted on bullshit.
I’d rather it be from modern reactor designs and builds
There’s a reason new nuclear loses out to renewables and storage, it’s just too expensive. Microsoft are paying to start up an existing reactor.
Hey now that’s not fair. AI can randomize your music playlists, summarize an email, write terrible code, steal others work, and completely invade your privacy.
What’s that? Oh, I guess you’re right, we could do all that stuff already.
There is something society could learn about itself if we spent anytime thinking honestly about how much of a dead end it is politically speaking to increase our use of nuclear power as a means of reducing our reliance on fossil fuels. Yet, when big corporate interests want it for their own reasons, it is no big thing and almost no politician will speak ill of it. Even though if some kind of disaster comes about because of it they will be left holding the bag of public opinion since that industry is so heavily regulated.
