Is AI energy consumption a stable 24x7 kind of thing? Inference load obviously changes with consumer traffic, so it will have a daily rhythm. But do the large providers use the rest of the capacity for training? Or are those separate clusters?
If it's a stable 24x7 load it would be ideal for nuclear energy, low carbon, but slow to adapt to changes in demand.
Might make sense to scale the load by following electricity supply/prices though?
Staying that as a genuine question since I'm not sure how the math works out at that scale, you have to weigh that against hardware depreciation of course.
Power purchase agreements are priced differently and usually written to guarantee power at a predictable price, think of it like reserved instances and spot on the cloud. Bulk of workloads don’t care or benefit from spot pricing.
Also Modern neoclouds have captive non grid sources like gas or diesel plants for which grid demand has no impact to cost. These sources are not cheap but DC operators have not much choice as getting grid capacity takes years . Even gas turbines are difficult to procure these days so we hear of funky sources like jet engines.
It’s not like when you ask GPT a question, the energy grid takes a dip. No, data centers have massive power draw. They also have battery backup systems that are the primary drivers of stable power along with power inverters and all sorts of power equipment on site. The fact that we are building out more data centers means we need more power. The energy marketplace has only so much extra capacity (various forms) before it too is depleted. So, you bring on more power plants, more reactors, more solar farms, moar powah!
No, what is sad is that we have the ability to turn every roof, every window, every side wall into a power source and yet we choose not to.
(I wrote a demand response energy grid “manipulation” platform)
"You're totally right in catching that mistake! The temperature is indeed one order of magnitude off. Let's walk through step by step what it should be:"
1) current admins working hand in glove with current AI hypemen safely bring old nuke plants online.
2) they bring it online but cause a meltdown or two in the process. Investor pressure lead to cut corners lead to meltdowns lead to public opinion turning against nuke power...again. Second nuke power winter begins.
I mean, fission is already a miracle. If you described the tech to the people in the 1800s and told them we just keep using fossile fuels they would laugh at you
To me one of the great tragedies of our time is that we could probably solve fusion if we just invested like 50 billion or something in it. Instead wasting so much effort on things like quantum computers seems insane.
Solving fusion could usher in the golden age that atomic power failed to produce
The thing that worries me is that it's still not obvious that fusion wouldn't also be the extremely expensive, slow to build boondoggle that fission is.
Because it has the word nuclear nearby and we'll be surprised at how ignorant our regulators can be, or because it'll turn out to be less safe than we think and it'll get red taped to death like fission did, or some non regulatory reason?
There are fundamental reasons why fusion, at least the DT variety, will be more expensive than fission. It has to do with inherently low volumetric power density of DT fusion reactors. ITER is 400x worse than a PWR; ARC is 40x worse. So, the reactor itself becomes much larger and, because it's also more complex, much much more expensive. The other putative advantages of fusion cannot make up for this. And fission itself is too expensive, so DT fusion loses to a loser. It's a double loser.
If you want to see where energy will come from in a deregulated environment, look at Texas. New grid capacity there is solar and batteries. Even gas isn't being installed much; the Texas state government put down $7.2B to fund more gas capacity yet this money has been mostly spurned, I think < $400M has been taken. New nuclear is completely out of the picture there.
Because most plans for it still involve attaching a giant steam boiler to turn the heat it produces into electricity and that bit alone will cost more than renewable alternatives.
Call me ignorant, but I’d rather we focus on stuff like increasing photovoltaic cell efficiency (and possibly cost-efficiency) by the 40%-60% we’re leaving on the table keeping them fully loaded and cooking.
Simple physics upgrades, like rotating cones, or lines of panels to swap with each other in Arizona-parking-lot conditions, can take us further, faster, and cheaper.
Nuclear is only safe after and during spending a bunch of money to keep it that way.
That makes me uncomfortable, because we’ve never had more instability in my lifetime, as far as “wildly important things not being addressed”.
Fusion will be slow to commercialize. Proof of concept is going to be much harder than for fission reactors. But if and when POC is attained, building commercial fusion reactors will not have as nearly much project risk, much less waste management risk, no proliferation risk, and much less financial risk in decommissioning. If you screw it up you can expensively damage your reactor, but you don't spread fallout, and you don't have to guard your waste like it's plutonium.
Neither fission nor fusion are going to put any juice on the grid before the AI bubble resolves, and then the financial calculations will be totally different.
Fusion likely wouldn’t solve much. Fuel and disposal costs are a small part of nuclear costs. It’s amortized capex for the extremely expensive plant, then maintenance. Fusion would make both of those costs worse
Exactly right. Fusion's only hope (and IMO it's not a great one) is a system where entire parts of a fission plant can be deleted. Helion does this by not needing turbine + generator, but doing direct conversion of plasma energy to electrical energy. But even so, they have to struggle with capex and reliability. Their reactor is coupled with a huge bank of capacitors (Zap has a similar problem; it's startling how large the capacitor bank is compared to their small fusion cell.)
Maybe it'll be a model running on a quantum computer that points us towards high temperature superconductivity, which would simplify the plasma confinement problem and unlock fusion for us.
The great tragedy is that we already have a practically unlimited and environmentally safe source of energy, which is nuclear fission. And we simply don't use it at a significant scale because of irrational fears about meltdowns.
It is not rational aversion. Nuclear is currently 10,000 less dangerous per unit of energy produced than the largest sources of energy: coal, oil and natural gas. We could afford to let nuclear get 10x less safe, so that it becomes vastly less costly to deploy, and a very possible result would be that it would replace the largest sources of energy, and would still be three orders of magnitude less dangerous than the sources of energy it replaced.
Regulation is inescapable, because the maximum damage from a nuclear accident would exceed the value of the company operating the reactor. A rational business treats any liabilities larger that what it could pay as equivalent, regardless of how large they could become, and hence will underinvest in safety measures.
And I'm sure you will agree there is a great and sorry history of nuclear efforts failing to achieve their cost targets. At this point, it is clear that such targets are sales numbers, not something one should actually believe. One cannot make this history go away just by wishing, as nuclear advocates like yourself seem wont to do.
I agree fossil fuels should go, but that's not an argument they should be replaced by nuclear. It's the argument nuclear advocates used to be able to lie back and comfort themselves with, but then you all got blindsided by renewables and storage zooming past you. You have to address those now, not the old competition you wished you were still running against.
That’s not true. They are physically massive, incredibly complicated machines with all kinds of large scale pressure welding, forging, containment systems, 100s of miles of plumbing, and other serious large scale engineering. They will never be anywhere close to as cheap as something as dead simple & mass manufacturable as solar.
Coal would harm the nature for months until the bubble bursts, nuclear waste will last forever. And hopefully the old nuclear plants won’t fail uncontrolledly.
Radiation from coal goes into the air and tailings, which aren’t well controlled, and stays dangerous for centuries.
Radiation from nuclear waste is constrained to steel casks in cooling ponds, and the waste can be reprocessed for use in breeder reactors instead of letting it sit.
The costs of protecting nuclear waste for 100.000 years from terrorists and during wars will be impossible. It's a super easy target. Just attack the power plants and nuclear waste facilities of the opponent and you have won the war.
Well we've had them since the 1950s, where are all these terrorist attacks? The only nuclear disasters have been accidental, and those were made worse by early reactor designs that didn't account for safety to the degree necessary.
if you operate a coal plant for 3 years until the bubble bursts, you will emit 3 years of toxins and carbon dioxide. If you operate a nuclear plant for just one minute, your highly radioactive waste will be a burden for thousands of generations.
Nuclear waste is not a burden. It's very manageable. The problem would be in any uncontained accident. The probability of that is low but not zero, especially in old reactor designs such as the ones they are talking about reactivating.
Some points:
- The generation of electricity from a typical 1,000-megawatt nuclear power station, which would supply the needs of more than a million people, produces only three cubic metres of vitrified high-level waste per year, if the used fuel is recycled. In comparison, a 1,000-megawatt coal-fired power station produces approximately 300,000 tonnes of ash and more than 6 million tonnes of carbon dioxide, every year.
Important condition being "if the used fuel is recycled".
To quote from the article on recycling:
"Although some countries, most notably the USA, treat used nuclear fuel as waste, most of the material in used fuel can be recycled. Approximately 97% – the vast majority (~94%) being uranium – of it could be used as fuel in certain types of reactor. Recycling has, to date, mostly been focused on the extraction of plutonium and uranium, as these elements can be reused in conventional reactors. This separated plutonium and uranium can subsequently be mixed with fresh uranium and made into new fuel rods.
Countries such as France, Japan, Germany, Belgium and Russia have all used plutonium recycling to generate electricity, whilst also reducing the radiological footprint of their waste. Some of the by-products (approximately 4%), mainly the fission products, will still require disposal in a repository and are immobilized by mixing them with glass, through a process called vitrification."
There are various informative videos on Youtube that cover vitrification, where the remaining waste is melted with glass-forming materials at a high temperature, and the resulting matter can be safely stored in steel vats. In comparison to carbon from coal, the volume is miniscule.
Old nuclear plants are huge targets for sabotage by foreign spies and terrorists. If they can make one blow up, it will decide every conflict in their favour. It's like having a huge bomb in your garden on display for everyone to target.
Tell that to the people in North Carolina whose neighborhoods were built on coal ash dumps. A little undocumented whooopsie from the power company and all the sudden your kids all have eye cancer.
AI is a bubble and in a few months, expensive nuclear energy will either be not needed anymore or replaced by cheaper solar and wind energy. The plant was decommissioned years ago because it was just not running profitable. And it will never be except within a bubble phase, like AI is currently in. Nuclear energy is the most stupid energy to ramp up during high demand phases.
Nuclear 'waste' is just waiting to become new nuclear 'fuel' in a fast neutron or 'breeder' reactor. Treated this way the volume of nuclear waste can be reduced by 90% while the remaining highly radioactive waste only needs to be stored for some hundreds of years instead of thousands due to its much shorter half life. It also extends the viability of nuclear fission (as opposed to fusion) by a factor of 10 by producing new fissionable material.
Recycled plutonium has negative value. It costs more to fabricate fuel elements out of it than it saves in uranium mining and enrichment costs. There is no great financial windfall waiting here.
US/UK/EU can barely supply energy as it is today. A little will be added for AI, slowly. But good luck re-industrializing without any energy left while competing with China at 1/16th of energy cost. This has been brewing for decades and the people responsible are not going anywhere.
All those are bold announcements by politicians. There are a lot of things to do and it takes many years in Latin America. Also there is no spare grid and electricity generation is kind of in the middle of nowhere. Good luck with constructing a datacenter in there, and then having technical employees to maintain it.
Don't get me wrong, I'm all for those plans. But it will take 5 to 10 years while China has a 15 year headstart.
Howdy neighbor, from a customer of Tennessee Valley Authority (primary generators are hydro and nuclear, with natgas peakers).
For some reason TVA keeps building new nuclear sites, never commissioning them, and then selling them at huge losses to private investors (or destroying them). Recently TVA got approval for µfission reactors to be installed... here we go again.
Is AI energy consumption a stable 24x7 kind of thing? Inference load obviously changes with consumer traffic, so it will have a daily rhythm. But do the large providers use the rest of the capacity for training? Or are those separate clusters?
If it's a stable 24x7 load it would be ideal for nuclear energy, low carbon, but slow to adapt to changes in demand.
Frontier LLM training can take months for a single run, which is about as stable as a load gets.
Might make sense to scale the load by following electricity supply/prices though?
Staying that as a genuine question since I'm not sure how the math works out at that scale, you have to weigh that against hardware depreciation of course.
It does not.
Power purchase agreements are priced differently and usually written to guarantee power at a predictable price, think of it like reserved instances and spot on the cloud. Bulk of workloads don’t care or benefit from spot pricing.
Also Modern neoclouds have captive non grid sources like gas or diesel plants for which grid demand has no impact to cost. These sources are not cheap but DC operators have not much choice as getting grid capacity takes years . Even gas turbines are difficult to procure these days so we hear of funky sources like jet engines.
It’s way more nuanced than this.
It’s not like when you ask GPT a question, the energy grid takes a dip. No, data centers have massive power draw. They also have battery backup systems that are the primary drivers of stable power along with power inverters and all sorts of power equipment on site. The fact that we are building out more data centers means we need more power. The energy marketplace has only so much extra capacity (various forms) before it too is depleted. So, you bring on more power plants, more reactors, more solar farms, moar powah!
No, what is sad is that we have the ability to turn every roof, every window, every side wall into a power source and yet we choose not to.
(I wrote a demand response energy grid “manipulation” platform)
Training on-demand, using spare GPU capacity is an interesting concept.
Maybe in the future we'll be making thankless water heaters out of GPU's so they can kick on when there's demand for heat.
It's different, more lightweight hardware for inference but can be in the same data center. Training requires beefier GPUs.
Oops. I really messed up. I should not have done that. I apologize. The temperature will not stop rising and I can't save you.
You're right! Your eagle eyed diligence caught my mistake. Let's try this again without the mushroom cloud this time.
they are not using AI to bring it up (only the demand from AI), but this is still funny.
"You're totally right in catching that mistake! The temperature is indeed one order of magnitude off. Let's walk through step by step what it should be:"
:-) ...
What are the likely outcomes?
1) current admins working hand in glove with current AI hypemen safely bring old nuke plants online.
2) they bring it online but cause a meltdown or two in the process. Investor pressure lead to cut corners lead to meltdowns lead to public opinion turning against nuke power...again. Second nuke power winter begins.
3) something else?
Whenever you ask yourself "why did the other Chernobyl generators keep running decades after the infamous accident" -- read the answer here.
On the bright side, maybe big tech will be the ones to accelerate fusion energy generation to a point of practical usage.
I mean, fission is already a miracle. If you described the tech to the people in the 1800s and told them we just keep using fossile fuels they would laugh at you
Or maybe they wont and we will have ruined the one planet we can thrive to make a chatbot.
To me one of the great tragedies of our time is that we could probably solve fusion if we just invested like 50 billion or something in it. Instead wasting so much effort on things like quantum computers seems insane.
Solving fusion could usher in the golden age that atomic power failed to produce
The thing that worries me is that it's still not obvious that fusion wouldn't also be the extremely expensive, slow to build boondoggle that fission is.
Because it has the word nuclear nearby and we'll be surprised at how ignorant our regulators can be, or because it'll turn out to be less safe than we think and it'll get red taped to death like fission did, or some non regulatory reason?
There are fundamental reasons why fusion, at least the DT variety, will be more expensive than fission. It has to do with inherently low volumetric power density of DT fusion reactors. ITER is 400x worse than a PWR; ARC is 40x worse. So, the reactor itself becomes much larger and, because it's also more complex, much much more expensive. The other putative advantages of fusion cannot make up for this. And fission itself is too expensive, so DT fusion loses to a loser. It's a double loser.
If you want to see where energy will come from in a deregulated environment, look at Texas. New grid capacity there is solar and batteries. Even gas isn't being installed much; the Texas state government put down $7.2B to fund more gas capacity yet this money has been mostly spurned, I think < $400M has been taken. New nuclear is completely out of the picture there.
Because most plans for it still involve attaching a giant steam boiler to turn the heat it produces into electricity and that bit alone will cost more than renewable alternatives.
Agreed.
Call me ignorant, but I’d rather we focus on stuff like increasing photovoltaic cell efficiency (and possibly cost-efficiency) by the 40%-60% we’re leaving on the table keeping them fully loaded and cooking.
Simple physics upgrades, like rotating cones, or lines of panels to swap with each other in Arizona-parking-lot conditions, can take us further, faster, and cheaper.
Nuclear is only safe after and during spending a bunch of money to keep it that way.
That makes me uncomfortable, because we’ve never had more instability in my lifetime, as far as “wildly important things not being addressed”.
Or just orders of magnitudes more than expensive than paving solar and batteries everywhere. Also a juicy missile target.
Because it’s going to be a ludicrously complicated, massive machine. Those are expensive to build and maintain
Fusion will be slow to commercialize. Proof of concept is going to be much harder than for fission reactors. But if and when POC is attained, building commercial fusion reactors will not have as nearly much project risk, much less waste management risk, no proliferation risk, and much less financial risk in decommissioning. If you screw it up you can expensively damage your reactor, but you don't spread fallout, and you don't have to guard your waste like it's plutonium.
Neither fission nor fusion are going to put any juice on the grid before the AI bubble resolves, and then the financial calculations will be totally different.
Fusion likely wouldn’t solve much. Fuel and disposal costs are a small part of nuclear costs. It’s amortized capex for the extremely expensive plant, then maintenance. Fusion would make both of those costs worse
Exactly right. Fusion's only hope (and IMO it's not a great one) is a system where entire parts of a fission plant can be deleted. Helion does this by not needing turbine + generator, but doing direct conversion of plasma energy to electrical energy. But even so, they have to struggle with capex and reliability. Their reactor is coupled with a huge bank of capacitors (Zap has a similar problem; it's startling how large the capacitor bank is compared to their small fusion cell.)
Maybe it'll be a model running on a quantum computer that points us towards high temperature superconductivity, which would simplify the plasma confinement problem and unlock fusion for us.
The great tragedy is that we already have a practically unlimited and environmentally safe source of energy, which is nuclear fission. And we simply don't use it at a significant scale because of irrational fears about meltdowns.
Rational aversion to financial meltdown, you mean.
The idea that nuclearphobia is to blame is a defensive fantasy.
It is not rational aversion. Nuclear is currently 10,000 less dangerous per unit of energy produced than the largest sources of energy: coal, oil and natural gas. We could afford to let nuclear get 10x less safe, so that it becomes vastly less costly to deploy, and a very possible result would be that it would replace the largest sources of energy, and would still be three orders of magnitude less dangerous than the sources of energy it replaced.
Regulation is inescapable, because the maximum damage from a nuclear accident would exceed the value of the company operating the reactor. A rational business treats any liabilities larger that what it could pay as equivalent, regardless of how large they could become, and hence will underinvest in safety measures.
And I'm sure you will agree there is a great and sorry history of nuclear efforts failing to achieve their cost targets. At this point, it is clear that such targets are sales numbers, not something one should actually believe. One cannot make this history go away just by wishing, as nuclear advocates like yourself seem wont to do.
I agree fossil fuels should go, but that's not an argument they should be replaced by nuclear. It's the argument nuclear advocates used to be able to lie back and comfort themselves with, but then you all got blindsided by renewables and storage zooming past you. You have to address those now, not the old competition you wished you were still running against.
No, it’s mainly because it costs too much.
The cost is almost entirely due to overly cautious rules for nuclear power generation.
That’s not true. They are physically massive, incredibly complicated machines with all kinds of large scale pressure welding, forging, containment systems, 100s of miles of plumbing, and other serious large scale engineering. They will never be anywhere close to as cheap as something as dead simple & mass manufacturable as solar.
> Instead wasting so much effort on things like quantum computers seems insane.
could be worse, could be all that effort, energy, talent and now nuclear waste to produce infinite pictures of shrimp jesus
.... oh crap
They do seem like responsible people. I'm sure that'll be fine.
At least, they're not reactivating a coal power plant.
I think it’s more of a “power any way we can get it” sort of strategy.
https://www.politico.com/news/2025/11/27/ai-gives-coal-plant...
Coal would harm the nature for months until the bubble bursts, nuclear waste will last forever. And hopefully the old nuclear plants won’t fail uncontrolledly.
Radiation from coal goes into the air and tailings, which aren’t well controlled, and stays dangerous for centuries.
Radiation from nuclear waste is constrained to steel casks in cooling ponds, and the waste can be reprocessed for use in breeder reactors instead of letting it sit.
The costs of protecting nuclear waste for 100.000 years from terrorists and during wars will be impossible. It's a super easy target. Just attack the power plants and nuclear waste facilities of the opponent and you have won the war.
Well we've had them since the 1950s, where are all these terrorist attacks? The only nuclear disasters have been accidental, and those were made worse by early reactor designs that didn't account for safety to the degree necessary.
The mercury and other pollution dusted by coal doesn’t degrade. It has the same two solutions as nuclear waste: ignore or concentrate.
if you operate a coal plant for 3 years until the bubble bursts, you will emit 3 years of toxins and carbon dioxide. If you operate a nuclear plant for just one minute, your highly radioactive waste will be a burden for thousands of generations.
Nuclear waste is not a burden. It's very manageable. The problem would be in any uncontained accident. The probability of that is low but not zero, especially in old reactor designs such as the ones they are talking about reactivating.
The old nuclear plants should definitely be inspected and safety issues should be remediated.
I have found this page on nuclear waste to be informative https://world-nuclear.org/nuclear-essentials/what-is-nuclear...
Some points: - The generation of electricity from a typical 1,000-megawatt nuclear power station, which would supply the needs of more than a million people, produces only three cubic metres of vitrified high-level waste per year, if the used fuel is recycled. In comparison, a 1,000-megawatt coal-fired power station produces approximately 300,000 tonnes of ash and more than 6 million tonnes of carbon dioxide, every year.
Important condition being "if the used fuel is recycled".
To quote from the article on recycling:
"Although some countries, most notably the USA, treat used nuclear fuel as waste, most of the material in used fuel can be recycled. Approximately 97% – the vast majority (~94%) being uranium – of it could be used as fuel in certain types of reactor. Recycling has, to date, mostly been focused on the extraction of plutonium and uranium, as these elements can be reused in conventional reactors. This separated plutonium and uranium can subsequently be mixed with fresh uranium and made into new fuel rods.
Countries such as France, Japan, Germany, Belgium and Russia have all used plutonium recycling to generate electricity, whilst also reducing the radiological footprint of their waste. Some of the by-products (approximately 4%), mainly the fission products, will still require disposal in a repository and are immobilized by mixing them with glass, through a process called vitrification."
There are various informative videos on Youtube that cover vitrification, where the remaining waste is melted with glass-forming materials at a high temperature, and the resulting matter can be safely stored in steel vats. In comparison to carbon from coal, the volume is miniscule.
Old nuclear plants are huge targets for sabotage by foreign spies and terrorists. If they can make one blow up, it will decide every conflict in their favour. It's like having a huge bomb in your garden on display for everyone to target.
Tell that to the people in North Carolina whose neighborhoods were built on coal ash dumps. A little undocumented whooopsie from the power company and all the sudden your kids all have eye cancer.
AI is a bubble and in a few months, expensive nuclear energy will either be not needed anymore or replaced by cheaper solar and wind energy. The plant was decommissioned years ago because it was just not running profitable. And it will never be except within a bubble phase, like AI is currently in. Nuclear energy is the most stupid energy to ramp up during high demand phases.
> nuclear waste will last forever
Nuclear 'waste' is just waiting to become new nuclear 'fuel' in a fast neutron or 'breeder' reactor. Treated this way the volume of nuclear waste can be reduced by 90% while the remaining highly radioactive waste only needs to be stored for some hundreds of years instead of thousands due to its much shorter half life. It also extends the viability of nuclear fission (as opposed to fusion) by a factor of 10 by producing new fissionable material.
Recycled plutonium has negative value. It costs more to fabricate fuel elements out of it than it saves in uranium mining and enrichment costs. There is no great financial windfall waiting here.
[flagged]
US/UK/EU can barely supply energy as it is today. A little will be added for AI, slowly. But good luck re-industrializing without any energy left while competing with China at 1/16th of energy cost. This has been brewing for decades and the people responsible are not going anywhere.
Market forces will fix this.
AI compute will go to places where energy is cheap
> Market forces will fix this.
You can't print energy and you can't print infra. Nine women can't have a baby in a month.
sixty five datacenters under construction in Chile, right now
source: Chile National Data Centers Plan | 2024-2030 MinCiencia Ministry of Science, Technology, Knowledge, and Innovation
note - this is "promoting the sustainable growth of the data center industry" .. you know it has to be sustainable
Is the universe sustainable though
http://www.thelastquestion.net
All those are bold announcements by politicians. There are a lot of things to do and it takes many years in Latin America. Also there is no spare grid and electricity generation is kind of in the middle of nowhere. Good luck with constructing a datacenter in there, and then having technical employees to maintain it.
Don't get me wrong, I'm all for those plans. But it will take 5 to 10 years while China has a 15 year headstart.
At first I thought this meant they were vibe-coding the restart of an old nuclear plant. What could possibly go wrong?
inb4 that South Carolina plant ends up turning into a scam that costs the tax payers billions (again)
Howdy neighbor, from a customer of Tennessee Valley Authority (primary generators are hydro and nuclear, with natgas peakers).
For some reason TVA keeps building new nuclear sites, never commissioning them, and then selling them at huge losses to private investors (or destroying them). Recently TVA got approval for µfission reactors to be installed... here we go again.
[•] <https://en.wikipedia.org/wiki/Bellefonte_Nuclear_Plant> $4B, estimated loss
Suggested (hopefully not Enron-esque) reading: https://www.amazon.com/Power-Failure-Rise-Fall-American/dp/0...
"AI" or "massive government subsidies"? Power generation that after 60 years still need to be subsidized may not be all that.