Microsoft Needs So Much Power to Train AI That It's Considering Small Nuclear Reactors
Microsoft is betting on small modular nuclear reactors to grow its energy infrastructure for training AI, a highly ambitious plan.
The thought of a nuclear reactor running on Windows is terrifying.
Even Microsoft does not trust Windows on Azure 🤣
Modern nuclear reactors are designed to fail safely, so Windows couldn’t actually create a Chernobyl. Everything wrong with nuclear in our world is with old-gen plants. It’s a technology that got ahead of itself by 50 years.
Lol, even Microsoft wouldn't use Windows to train AI.
Like Microsoft uses Windows for anything that matters since they got rid of Balmer.
But... Developers!... /s
Could be worse, could be running MacOS. Surely nothing bad can happen while the entire system freezes for no reason for 15 minutes or more without any possible input from the user. It will always fix it self... (hopefully before the reactor achieves a run away meltdown chain.)
What are you running? I’ve never had an issue like that at all
Reminds me of that time the technodork ran his minecraft reactor with opencomputers and lost his base because the computer blue screened. Almost as funny as that time the entire city lit up because they were using raw radio signals to control their reactor and a nearby thunderstrike instructed the reactor to drop all the fuel and go supercritical. This is why you add realism to video games, it leads to hilarious stuff like this.
EDIT: That was actually the same server where they sabotaged the entire electrical grid to blow up everyone's base as a send-off and mine was the only one standing at the end because I was the only one who bothered to set up a surge protector under OHSA (Omega Haxors? Safety!? AHAHAHAH!) it just so happened that the system designed to save the grid from my many exploits just so happened to work in reverse.
Better than coal or oil, it might even result in more R&D into reactor designs.
Yeah, I don't understand why building a relatively clean energy source is a bad thing. Reactors are now like 3+ generations past the versions that were super dangerous. Hell, they even have reactors that can use spent fuel from other reactors.
There’s no shortage of modern reactor designs. We have amazing stuff designed and even prototyped and proven - low waste, safely-failing reactors that basically can’t melt down. All we really lack is funding and regulatory clearance to build more.
Cortana, can you design a nuclear reactor to train you better?
Yes daddy
Searching “Design a nuclear reactor to train you better” on Bing…
I thought this crazy energy consumption shit would cool off a bit after assholes stopped bitcoin mining.
Glad AI stepped up so we can generate bad art and prose while buttfucking the planet
Ok, hear me out: crypto, based on "proof of training an AI"
If it takes so much power, it must be secure, and this way it wouldn't be "totally wasted"...
The planet will be alright. It will be lush green in a few million years when humans no longer exist.
The current ecosystem, though.... yeah. Buttfucked.
That doesn't sound crass enough to grasp the full extend of the development. AI will take all of the energy. E v e r y t h I n g.
There won't be a planet once there is a Dyson sphere. https://en.m.wikipedia.org/wiki/Dyson_sphere
Cryptocoins, blockchain, NFTs, AI craze. It's all the same people who think that the solution to the problems that capitalism has created is technology.
Yeah, it's treating the symptoms... not the cause.
The GPU manufacturers are having the time of their lives.
So we finally get thorium power, but its only used to make celebrity porn for incels.
Hey, whatever keeps them out of Walmart parking lots at 1am.
Like there's somewhere better to buy celebrity porn at 1am. Psh. I'll believe it when I see it.
Nice!
Honestly getting Thorium power AND never having Incels leave their home or interact with society again sounds like a win-win.
requires an intensive carbon footprint
Maybe we should focus on the collapsing ecosystem then instead of training AI datasets.
Nuclear power means they can do both.
Nuclear power still requires huge front costs (goal of SMR is to reduce that, but first generations will not solve it), so it could be better to use them for every day life needs rather than a prospective commercial venture.
AI might be a fantastic tool to help fix the environment, though.
We already use AI in climate change models. This is a large language model that honestly, we don't need.
bold of you to assume microsoft is interested in that
We already know how well Microsoft optimizes code, so this comes as no surprise.
Building and maintaining one isn't really the concern I have with this one, nuclear reactors are incredibly safe these days. What are they going to do with the nuclear waste? That's the real issue here. Governments can barely figure that out, how's a megacorp going to do that in an ethical way? I already see them dumping it in a cave in some poor country in africa.
If they're actually using a new type nuclear reactor, the small portable ones, then the waste is both incredibly small and recyclable. Nuclear technology has come a long way since the decades old reactors, we just haven't built very many new ones to showcase that.
As noted elsewhere, these don't create the same kind of spent fuel as a PWR. So that helps.
But also, the people who designed the PWRs didn't just say "and then we'll make shitloads of unmanageable waste lol!" Up until the Carter Administration, we ran a system called "reprocessing" that essentially shredded and dissolved the old fuel rods, isolated the metals chemically, and packed out separately.
France does this. Finland does this. Japan does this. Their waste concerns are negligible compared to ours.
Meanwhile Carter, bless his heart, determined that reprocessing was a proliferation risk, and shut down the US industry, saying "y'all will figure out a way to dispose of these things".
So now we are using circular saws to hack these things apart, cramming them into barrels stuffed with kitty litter (you read that right), and hoping that nothing will happen to the barrels for 50 million years?
Long-term waste disposal became an impossible problem to solve in the US because our one and only allegedly nuclear-savvy president made the solution to the problem illegal. It became one immediately, and has never stopped being one.
I'm generally against nuclear--or more accurately, think the economics of it no longer make sense--but there's one thing I think we should do: subsidize reactors that process waste. It's better and more useful than tossing it in a cave and hoping for the best. Or the current plan of letting it sit around.
Nuclear waste is a technically solved issue with long term geological storage, long term dangerous waste which requires more tech is a very small mass. The problems are political, uneducated people are irrationally scared of those waste that they associate with Chernobyl so they oppose any kind of geological storage, and politicians don't have the balls to openly contradict them.
Weird thing is, I'd trust them to not abandon the reactor during a budget shutdown...
I mean you say that as if just burying it isn't actually the proven safest option.
Startups are already beginning to explore using old oil drilling equipment to sink nuclear waste below where it'll pose a threat, after it's been suffused into a shitton of concrete of course.
Very rarely is nuclear waste of the corium toothpaste variety, more often it's the old hazmat suits that are getting replaced and need to be disposed of with special care, or expired rods you can still have limited contact with without many issues.
They'll ship it to India/Thailand
Governments can barely figure that out,
Governments aren't exactly known for efficiency. A corporation is less likely to bogged down by just the mere fallacy that "other entities can't figure it out, why should they do it?"
The human body produces a lot of electrical impulses. What if they just took all their workers and put them in some type of "work pod" and harnessed the energy to run the large scale AI?
It would cost more energy to feed them then they would produce. /s
That's just energy conversion. Loss is expected.
This is what happens when you don't teach your kids the Laws of Thermodynamics in school...
The reason is ultimately irrelevant, but I welcome more nuclear energy.
They could just invest in a solar farm or something, they are just a lot more economical.
Nuclear is okay, but the costs compared to renewables are very high, and you have to put a lot of effort and security into building a reactor, compared to a solar panel that you can basically just put up and replace if it snaps.
You probably know this discussion already through.
Edit: Glad to see a nice instance of the discussion going here.
In their specific use case that won't really work.
They want to use all of their available property for server racks. Covering the roof with solar won't give enough power/area for them. A small reactor would use a tiny fraction of the space, and generate several times the power. That's why it'd be worth the extra cost.
For those who haven't seen this discussion before, I feel like doing the next step in the dance. Cheers Plex.
It's important to note that nuclear is capable of satisfying baseload demand, which is particularly important for things like a commercial AI model training facility, which will be scheduled to run at full blast for multiple nines.
Solar+storage is considerably more unreliable than a local power plant (be it coal, gas, hydro, or nuclear). I have solar panels in an area that gets wildfire smoke (i.e. soon to be the entire planet), and visible smoke in the air effectively nullifies solar.
Solar is fantastic for covering the amount of load that is correlated with insolation: for example colocated with facilities that use air-conditioning (which do include data centers, but the processing is driving the power there).
The more people who invest the better the tech becomes the more the price comes down. Nuclear is excellent base energy
This is false. Nuclear has a very competitive levelized cost of energy (LCOE). Nuclear has high upfront costs but fuel is cheap and the reactor can last much longer than solar panels. The big picture matters not just upfront costs.
Source: https://www.energy.gov/sites/prod/files/2015/08/f25/LCOE.pdf
Right, let's welcome throwing millions or billions of dollars at wasting enormous quantities of concrete and water and at generating highly toxic waste that will irradiate its environment for millennia, and at ripping apart landscapes to extract uranium, I mean that's such a nice thing, we need much more of it! It's not like we already have perfectly renewable solutions to providing power...
This seems kind of ideal though, computers provide a near constant load (relatively speaking) that combines very well with nuclear energy.
Perhaps we should be asking why we haven’t already been doing this for the past decade?
Because it costs less money to push the cost to for taxpayers to subsidize it than owning it
Correction
Tell me more about how capitalists efficiently allocate resources.
This may actually be one of those things where it turns out to be worth it (for them anyway), if they can get some major technological advancements out of it.
There are so many other things in the world that are way more wasteful and way more pointless.
Allocative efficiency in economics just means that you can't make someone better off without making someone else worse off.
An efficient allocation isn't necessarily equitable.
And the first welfare theorem of economics only claims that the market will produce an allocatively efficient result if its complete, in perfect competition, and everyone has complete information. Which has the obvious problems of those preconditions not matching reality.
I predict that within 10 years, computers will be twice as powerful, ten thousand times larger, and so expensive that only the 5 richest kings of Europe will own them
Hi bing. How do I stop a nuclear reactor from going critical?
For those correcting my error It was just a joke. The only things I know about nuclear power I learned from the simpsons and Kyle hill
you turn it off.
"critical" is the normal operating state of reactor when it's working. what you want to avoid is supercriticality, which means that power is rising. if it's delayed supercritical but prompt subcritical, power rises and may or may not stop on its own at some point. when it's prompt supercritical, you don't even have time to ask https://www.nuclear-power.com/nuclear-power/reactor-physics/nuclear-fission-chain-reaction/reactor-criticality/
Modern reactor design also pretty much makes runaway reactions nearly impossible, as in, you have to actually try to fuck it up.
Even Fukushima didn't have a runaway reaction, it just lost coolant.
Don't turn it on? Critical means a reaction that is exactly self sustaining, i.e. a constant power level.
Buy MS office subscription.
LLM seemed really impressive at first, but it made it to “this year’s NFTs” in record time.
with the hopes of buying electricity from it as soon as 2028.
Fusion won’t be ready by then
Energy should be public
energy, water, food and housing should not be used for profit. Everything else is fair game.
Max Planck seems to have learned from the Fraunhofer mp3 thing and is getting their hands dirty with commercialising Stellerators. "In the 2030s", right now they're drafting, maybe simulating. Awfully short on details right now if you understand anything about fusion.
Just fill the Country with Solar, Wind and Water... won't take 10 years and will be cheaper too.
Hydropower is about as bad for most ecosystems as burning fossil fuels. And its definitely not something that can be done quick or cheaply.
Power density matters. And nuclear is pretty fucking dense haha
.. for some applications. Not most tho. Really like 5. Everything else should be solar/wind/hydro
... And cause a lot of pollution and ecological stress, unless you funnel a LARGE amount of money and time into it.
Do you want to argue, that the construction of a nuclear power plant causes significantly less ecological stress and pollution than solar panels and windturbines?
Think about if you really want to claim that as a thing you actually believe in.
I'm just gonna throw some words in a pool.
concrete, steel, space, deforestation, river, 10+ Years construction time, heavy machinery, dust, natural habitats, fuel, mining, waste, noise, cost, france...
Thank you. i rest my case.
I guess the rich don’t have an energy crisis.
I can't handle this . i'm going to sleep.
that's what the operating system running the reactor would say
At least we hope that it will do that when it can’t handle the reactor - it puts reactor to sleep.
Imagine if it ends up requiring the achieving of ignition for Microsoft to launch a version of clippy that is able to reliably comprehend English grammar enough to make writing recommendations.
Burning 3 continents worth of coal to steal artwork at maximum efficiency.
Do nuclear reactors use coal?
Does it matter what they claim they’re going to do differently in the future when they’re burning indefensible amounts of coal right now?
Coal? Did you miss the nuclear part?
If they handle their nuclear reactor like they handle their cloud infrastructure security, we’re doomed.
I don't get why a train AI would need so much power, how hard is it to drive a train?
Will the nuclear reactors be on the train with the AI, or will it be some sort of wired transfer?
I assume they can power the train AI by pantograph or third rail - no reason to have nuclear powered trains, this isn't Factorio.
If your train has a bad conductor, it will require more power.
Yeah tell me about it, I only have bad conductors.
This is how we end up with Blaine the Mono
Not sure if you're making fun or actually not understanding? To clarify, they need the power for training AI models. No trains are involved, neither passenger nor cargo – though atom powered trains sounds interesting as well!
Thank you, you are very kind. It was meant as a joke.
AI's going to kill us off by doing what we do better than us. Consuming resources and producing waste. And we're already pretty good at it.
I’m not opposed to new nuclear energy in principle. However Microsoft, an unrelentingly bad organization that consistently acts in bad faith to its customers, employees and businesses parters, and is seemingly dedicated to making awful products that never meaningfully improve, is not something I would trust to do nuclear safely.
Thankfully, any actual new type using modern tech have self-limiting reactions. Thorium ones, for example, can't meltdown because the high heat in that process kills the reaction itself.
VSCode is cool, no ?
I will never forgive them for telling Microsoft Games to stop developing Freelancer 2.
Microsoft has invested in several fusion projects too.
I guess the nuclear power people are gonna become Microsoft fans... powered by nuclear power.
Yesssss
I don't give a shit about training AI but the idea of Microsoft running nuclear reactors is hilarious to me. Either they do it well and we all benefit from the knowledge, or Windows goes out with a bang
Small Modular Reactor technology is the future, and it's really promising.
Self-contained (no onsite refueling), mass produced (cheap, higher quality), and modular (add more for more power, or small enough to power a data center).
Here's some quick videos from a professor of Nuclear Energy covering topic:
Small Modular Tractors:
Micro-Modular Rectors:
Here is an alternative Piped link(s):
https://youtu.be/TYnqJ4VnRM8?si=qODxzqzOCoiNMinH
Piped is a privacy-respecting open-source alternative frontend to YouTube.
I'm open-source; check me out at GitHub.
Mega corporations should not be allowed to use nuclear power plants purely for themselves.
Also, if you need that much power to do something bthat a human brain does with under 100 watts, I really think you're doing it wrong
If you're so smart why don't you come up with a way to do it under 100 watts???
Also this is training them not using them. Using an ai consumes significantly less power than the process to train it sort of like how humans take more to learn than to put something in practice.
Yeap. They are trying to train the AI quickly, not over the course of 18 years.
Also, it's early still early for these AI/LLM tools. The first few iterations of things are generally not very efficient. After you can prove it works, THEN you can make improvements to it, make it more efficient, etc.
However, I think that the approach of feeding in more power instead of optimizing it is the wrong approach. I feel Microsoft could get further ahead of it could find ways to train models more efficiently 🤷
People tend to forgot the millions of years of horribly inefficient evolution it took to develop the human brain.
Organic technology is hard. If you can figure out how to grow a compute system you will take human technology hundreds of years into the future. Silicon tech is the stone age of compute.
The brain has a slow clock rate to keep within its power limitations, but it is a parallel computational beast compared to current models.
It takes around ten years for new hardware to really take shape in our current age. AI hasn't really established what direction it is going in yet. The open source offline model is the likely winner, meaning the hardware design and scaling factors are still unknown. We probably won't see a good solution for years. We are patching video hardware as a solution until AI specific hardware is readily available.
I bet it'd be a whole lot easier to grow and organic computer if you didn't have to worry about pesky things like people thinking you grew genetically engineered slaves.
I am so excited for the advances that neuromorphic processors will bring, which is not exactly my field, but adjacent to it. The concept of modelling chips after the human brain instead of traditional computing doctrines sounds extremely promising, and I would love to get to work on systems like Intel's Loihi or IBM's TrueNorth! If you think about it, it's a bit ridiculous how corporations like Nvidia are currently approaching AI with graphics processors. I mean, it makes more sense than general-purpose CPUs, but it is at the very least a subideal solution.
The whole language model scene system started with "we accidently found something that kinda works" and is now in full "somebody please accidently find a way so it uses less power" mode.
Why should they not be allowed? Nuclear power plants are great options and will mean less demand on worse energy providing sources
Almost all nuclear power plants in the US are privately owned and operated. Regardless of whether you agree with it or not, this isn't new
Not a terrible argument. How do we limit compute vs. output?
I'm not one to be all doom and gloom about ai, but giving one its own small nuclear reactor, presumably one that's in close proximity to it and separate from the local power grid... that's obviously going to have substantial security measures around it... and be that much more difficult to cut off if need be....
I mean, it's starting to sound a lot like an unbelievable plot hole in a bad sci fi movie isn't it?
That's.... Not how things work. Creative, but no.
You're gonna be reductive about it as if AI hasn't ALREADY USED TASKRABBIT TO TRICK A HUMAN INTO SOLVING A CAPTCHA???
GPT has already neutralized our primary defence, it can now do anything.
That's a nice sci-fi thought, but also not really how it works irl.
Physical proximity is not really a factor here.
There's a lot more going on at all levels that makes this absurd at best. Movies are not a good representation of reality, they can't be if they are meant to be entertaining.
Why are you saying this as if the AI would have control over the reactor.
It's unlikely they'd even be in the same building, or even the same campus. We have these crazy things called "wires" that let us transmit a lot of power over distances, so your small nuclear reactor can be remote, safe and secure and your AI lab can just be on your main campus.
They should call it WOPR for no reason whatsoever.
Anyone fancy a nice game of chess?
Ironic you compare it to a bad sci fi movie and that's exactly where your knowledge is coming from.
You do konw, that it is IMPOSSIBLE for a nuclear reactor to explode in a nuclear fission explosion, aka become a nuclear bomb. Reaching critical mass isn't possible. Nuclear reactors can catch on fire, if built using graphite, that isn't done anymore, or have a steam explosion. but that's it.
They can also get hot enough to melt the metallic components (including the fuel itself) if the reaction isn't properly regulated (hence, "meltdown"), but you're correct that that's still not a fission explosion.
people need lots of energy for life as well, now what??
Its going to be all for nothing at the end of the day.
I've played this game and know where it's headed. They've decided to create Vega. Prepare for them to announce Argent Energy soon.
This is the best summary I could come up with:
Training large language models is an incredibly power-intensive process that has an immense carbon footprint.
Now, The Verge reports, Microsoft is betting so big on AI that its pushing forward with a plan to power them using nuclear reactors.
Yes, you read that right; a recent job listing suggests the company is planning to grow its energy infrastructure with the use of small modular reactors (SMR.)
But before Microsoft can start relying on nuclear power to train its AIs, it'll have plenty of other hurdles to overcome.
Then, it'll have to figure out how to get its hands on a highly enriched uranium fuel that these small reactors typically require, as The Verge points out.
Nevertheless, the company signed a power purchase agreement with Helion, a fusion startup founded by OpenAI CEO Sam Altman earlier this year, with the hopes of buying electricity from it as soon as 2028.
The original article contains 346 words, the summary contains 147 words. Saved 58%. I'm a bot and I'm open source!
That’s pretty cool. I would have guessed it would be cheaper to use wind energy.
Cheaper but unreliable.
While I appreciate them going a greener route, if these chat AIs are still this inefficient to simply train, maybe it is best left to return them back to the research phrase.
You say "simply train," but really, the training of these models is The most intensive part. Once they are trained, they require less power (relatively) to actually run for inference.
So it sounds like they need a shitload of GPU power. You know what also costs a shitload of GPU power crypto mining? Could they not outsource the work to all those GPUs that stopped mining crypto once it plummeted?
I am surprised this hasn't become a community project already. I assume there is some limitation that I am unaware of.
There's tradeoffs. If training LLMs (and similar systems that feed on pure physics data) can improve nuclear processes, then overall it could be a net benefit. Fusion energy research takes a huge amount of power to trigger every test ignition and we do them all the time, learning little by little.
The real question is if the LLMs are even capable of revealing those kinds of insights to us. If they are, nuclear is hardly the worst path to go down.
That makes zero sense
Oh my Gwyn, this comment section is just amazing.
Yeah, every time AI or nuclear energy is mentioned the quality of the comments plummets. Since we have the two combined in this story, the results are tragically predictable.
Capitalism will make humans extinct at this rate.
We've probably got about 10yrs of normalcy left. Economically, we're already deviating. Greed is what destroyed us all.
Is this the new crypto mining is bad story?
Only if you think developing small nuclear reactors is a bad thing. I think it's a good thing.
Yeah, there's definitely some parallels between how people negatively reacted to crypto and the current wave of reactions towards AI. My personal theory is that a lot of people are becoming hostile towards big tech, and anything that's seen as the next big thing in tech will quickly get negative reactions.
It's grim to think the driver for renewable energy will be the demand for endless AI synthesized child porn.
the FBI appreciates you being so open in how you use AI. An agent will be with you shortly.
Just go to Chernobyl and build there. Deal with the aftermarch now instead of later. The order matter because then they will understand there is a big problem with nuclear power.
Yes, it have even happened in modern time. Look at what happened at Japan 2011 fukushima. It will happen again. Political/economic/misstakes will happen again. Just look at Russia firing at nuclear plants... we are doomed.
How many people did the Fukushima incident actually kill? Meanwhile people are actively being killed by air pollution and climate change caused by fossil fuel energy. Nuclear energy incidents seem worse because they happen over a short period of time, but it's just like with airplanes - plane crashes are horrific and disastrous, but statistically airplanes are massively safer than even rail and especially road transport.
It requires good governance and adherence to safety standards and upkeep to be safe, but we've shown that we can reasonably do that for the most part.
Renewables should of course be the first priority, though lithium mining is also a significant health hazard - but really when you compare everything statistically and not just by the significance of individual events, there's no reason we shouldn't be trying to eliminate fossil fuels by any feasible means, and that includes nuclear power.
adherence
You know that if we look at for example Chernobyl how many did varies a lot between who you ask. That is even today. You also need to think of those who did not die but survived with some damages. Chernobyl is not over so you can't even stop counting who died from it. It will take very long time before it is over.
Yes, I agree with you that pollution is a even bigger problem that I can't understand that no one cares about it. Seeing on TV that people going to work in a smog of smoke and they don't react. It is deadly! Personally, I agree that fossil fuel should banned for example for cars. There are alternatives, maybe not as good but you will survive. Teleports does not exist yet but we survive somehow anyway.