Apollo Says AI Energy Gap 'Will Not Be Closed in Our Lifetime' (bloomberg.com) 76
The amount of energy required to supply the data centers powering AI is so vast that meeting that need may be more than a lifetime away, according to a senior executive at Apollo Global. From a report: "The gap between what AI is demanding and what we have everywhere in the world on the grid in terms of generation and transmission is huge and will not be closed in our lifetime," Dave Stangis, who has led and developed Apollo's sustainability strategy over the past four years, said in an interview.
That means sustainable energy investors need to accept that renewables alone aren't enough to power the AI age, he said. The comments encapsulate a new approach across the finance industry, where the economics of the energy transition -- a concept intended to represent the shift to a low-carbon future -- are becoming merged with the economics of an unprecedented boost in supply. "So what is happening around the world, there's no doubt about it, is what you might call energy addition," Stangis said. "The world is scrambling to add every source of power."
That means sustainable energy investors need to accept that renewables alone aren't enough to power the AI age, he said. The comments encapsulate a new approach across the finance industry, where the economics of the energy transition -- a concept intended to represent the shift to a low-carbon future -- are becoming merged with the economics of an unprecedented boost in supply. "So what is happening around the world, there's no doubt about it, is what you might call energy addition," Stangis said. "The world is scrambling to add every source of power."
why not use some of the waste heat? (Score:4, Interesting)
I'm surprised I haven't seen anyone trying to use the waste heat that all this computer power is generating? I realize that would impact cooling a little, but surely SOME of this can be recovered efficiently? Steam turbines are the usual way to turn heat into electrical power. Is there no way to do it for data centers?
For example, use a heat pump to concentrate the heat to above boiling temperature then use that to boil water to run a steam turbine. The heat pump would require some power to run, but I think you could run that at a net-positive for power?
Re: (Score:1)
Re: (Score:3)
The heat pump's working fluid doesn't have to be water, it'd be whatever fluid can phase change at a "convenient pressure". The released heat on the high pressure side would be used to boil water into steam, which could then be moved to a turbine to generate power.
That "working fluid" woud be what is circulating on the low pressure side, through the cooling blocks.
Equalization (Score:3)
Simply data centers should pay above the average retail rate for single home owners do for home within a metro area and within 75 miles in the same state.
Individual homeowners and apartment renters should not subsidize AI data centers.
Re:why not use some of the waste heat? (Score:5, Interesting)
I'm surprised I haven't seen anyone trying to use the waste heat that all this computer power is generating?
Some data centers have used their waste heat to do things like heat pools, or apartments/condos.
Re: (Score:2)
Famously, the Coors (beer) brewery in waste heat keeps the sidewalks free of snow and many of the buildings heated in the winter. PG&E built a power plant at the site of the H&C sugar refinery in california and the waste heat is used to make baking sugar. It's not that uncommon.
Re: (Score:2, Insightful)
I don't think people have realized that AI isn't for us. It's not a product. It's what the top 6,000 people on earth are going to use as an answer to the question "well if they fire all of us who's going to buy their products?"
The problem seems to be that people cannot comprehend capitalism going away except maybe by Star Trek style utopian soci
Irony alert! (Score:3)
The parent is obviously sarcastic. Not wise in such a forum as this.
Re: (Score:2)
Re: (Score:2)
when they can just take your electricity away from you?
Who is going to let them just take it?
Newer AI implementations are likely to be much more energy efficient*, reducing the anticipated payback for infrastructure.
Re:why not use some of the waste heat? (Score:4, Informative)
> Is there no way to do it for data centers?
The water temps are typically barely warm enough for most people's preference for a shower.
> For example, use a heat pump to concentrate the heat to above boiling temperature then use that to boil water to run a steam turbine.
Getting a heat pump to operate at atmospheric boiling water temps is extremely difficult. Remember that to have a working heat pump, you need a refrigerant medium that condenses at the high temperature side under a given pressure and also boils at or below the low temperature side at a given pressure... then, you need to build a machine that can actually create those pressures.
Now consider that most steam cycle powerplants use superheated steam at temps of over 500C. What material could you use that can be made to condense into a liquid at >500C, what kinds of pressures would be required to make that happen, and what could you even build such a machine out of to survive those conditions?
> I think you could run that at a net-positive for power?
The second law of thermodynamics has left you a voicemail...
=Smidge=
Re: (Score:2)
The Japanese have found a way to use small temperature differences to generate electricity. They use the temperature gradient between shallow and deep sea water, for example.
The key is to use liquids that boil at much lower temperatures to drive the turbines. Of course then you get into difficulty handling those typically quite nasty liquids, but they seem to have made it work and have exported the technology.
It's probably still a bit too new to be of much interest to data centres, but if it keeps ramping u
Re: (Score:2)
> The Japanese have found a way to use small temperature differences to generate electricity
And for about $50 I can buy an engine that runs off the temperature difference between the ambient air and a cup of hot water. The idea of using thermal gradients in the ocean to generate power is at least 150 years old. Any guesses why it's not caught on?
Hint: the facility in Japan you're probably thinking of only generates 100kw (~135HP), and it's not clear if that's before or after they account for the power to
Virtual batteries (Score:2)
the core challenge of renewable energy is it's inconstancy. Physical batteries are a bandaid and long distance grids are a council of despair. The real solution for reliable renewable energy is to just build out four or five times the peak load. Then when it's cloudy or not windy you still have way more power than you need to supply the peak load. But of course this has the problem that you just spent four of five times as much capital. And that's a non-starter. But the easy, though bad solution, to
Re: (Score:2)
It's not hot enough. I suppose in a Canadian winter you could boil butane as the working fluid but the Carnot efficiency would be rather low.
Putting the data center in town you could pump the warm water around for district heat. If you can get to 250 F, equal to 15 psi steam you can run an absorption air conditioner or a distilling plant.
Re: (Score:2)
Are these datacenters getting over 212F? If they are, the cooling ain't working right.
Do you mean like pump the cooling water into a tank and use the heat from the water to heat the tank to boiling? You'll never get there with just a football-field of datacenters... at best, the combined heat from the whole place might heat a studio apartment... I know, my 3950X 24-core Threadripper and Titan X running CudoMiner all night maybe raises the livingroom a degree.
To turn that big generator, you need pressure,
Who or what is (Score:2)
Apollo Global?
Re:Who or what is (Score:4, Informative)
Oh you know just your run of the mill asset management firm controlling $840,000,000,000 in wealth and assets. American as apple pie.
The predictions may be nonsense (Score:4, Interesting)
AI made rapid progress by increasing processing power and data.
This lead some to extrapolate that future AI will need astronomical amounts of power.
Meanwhile, researchers are working on improved algorithms or completely different approaches.
Brains take a tiny amount of power to work. Current AI is very inefficient.
I predict that new architectures/chips/algorithms will dramatically reduce the power necessary.
Re: (Score:1)
Yep, some restraint all round is in order. And that goes for personal activities like jet-setting too.
Re: (Score:3)
And cryptocurrencies. What a colossal waste of energy. Our current currency system is sufficient. Ban crypto mining to save the energy for more useful purposes.
Re: The predictions may be nonsense (Score:2)
Re: (Score:2)
Re: (Score:2)
It doesn't really matter how "efficient" you make the LLM-AI... the computer still draws power, just draws more when the LLM is doing something.
And, something like having the computer go into standby right away when it's not being used is the worst thing you can do for a computer... the heat stress from cycling power (warm up from being on, cool down while in standby) breaks solder joints... and spinning drives up and down a lot wears out the bearing.
Re: (Score:2)
let's hope it's not closed at all (Score:5, Insightful)
"That means sustainable energy investors need to accept that renewables alone aren't enough to power the AI age, he said."
I means that the "AI age" needs to accept it too. What is taken for granted here, and shouldn't be, is that AI computing is entitled to unlimited power, it is not.
'"So what is happening around the world, there's no doubt about it, is what you might call energy addition," Stangis said. "The world is scrambling to add every source of power."'
And this is a huge mistake. Mankind does not need this and will suffer because of it. This is an indulgence motivated by billionaire greed, society should not only NOT provide this energy, it should make this massive threat to the planet illegal.
AI is designed to allow wealth to access skill (Score:3, Insightful)
So given that yeah, AI can have all the energy in the world. With a little bit set aside for the ultra wealthy to enjoy.
It is genuinely bizarre that that people cannot wrap their heads around the idea that there is no place for them in the glorious future that our billionaire and trillionaire overlords are planning.
I think the problem is that it's already hard for people to imagine capitalism going away let alone capitalism going away not because communism or socia
Re: (Score:2)
That means sustainable energy investors need to accept that renewables alone aren't enough to power the AI age, he said.
That means there is unlimited demand for renewables and thus it will be impossible to saturate the market with renewables. This is not the argument he wants to get people to stop investing in renewables and invest instead in non-renewables which is what he seems to want (from TFS).
"The gap between what AI is demanding and what we have everywhere in the world on the grid in terms of generation and transmission is huge and will not be closed in our lifetime,"
So AI is going to demand all the power we can produce in our lifetimes? What if we don't meet that "demand" and allocate the amount of power that its actual demonstrated contribution to the global economy justifies? Will "AI" get
Re: (Score:2)
"But... we need LLM-AI to summarize every news article and email we want to read!"
The reality is that the LLM-AI we have now will be used to replace as many human-jobs as possible... down to even using LLM-AI powered factory robots with machine vision (why does BK need a human to flip the burgers... a robot can do it, too). Once it's properly trained, the company won't need you to spend a week coding the new Windows update... the LLM-AI will do it in 10 minutes.
Okay... let's just say one LLM-AI datacenter
Prediction: (Score:5, Interesting)
20 years ago people swore up and down the only way to save money was to move storage from local to a central terminals to save it all.
10 years ago people swore up and down the only way to get high end video games playable on a high end device was to move it all to a central terminal.
Today your phone has more processing power than the supercomputer of the 80's, cloud storage for individuals is a scam and cloud "products" used to ensure individuals don't own any product, IT departments are going back to building their own servers as cloud providers jack up profits; and the Nintendo Switch came out less than 2 years after execs started swearing up and down "cloud gaming" was the inevitable future.
Someone's going to optimize large language models to run fast and have individuals parts sparsely loaded off an SSD, if you want an LLM to write your resume or do your English homework for you it'll run locally on your laptop soon enough, and all these AI centers will be trillions of dollars worth of garbage.
Moore's law I would like to have a word with you (Score:3, Interesting)
The more that llms get optimized the more is expected from them. The more things that they can automate.
This isn't something I think that really smart people are going to technology our way out of. Honestly I don't see a solution. I've said it before but I think we're going to see mass unemployment leading into world wars and then eventually nuclear wars ending our species.
I think that the inability to deal with the combined threats of technological unemployment in civilizations
Re: (Score:2)
And yet the demand for more power has never gone away. It keeps increasing. The more computers can do, the more people ask of them. However powerful they get, people will always want more so they can do even more.
Increasing efficiency won't solve this problem. Right now, the economic imperative is to build bigger companies running bigger models on bigger data centers. Bigger than what? Bigger than their competitors, of course. Everything else is secondary, including whether they destroy the planet in
Re: (Score:2)
"And yet the demand for more power has never gone away. It keeps increasing."
If it's not data centers it's EVs and heat pumps to replace gas heat, cooking, and hot water.
My 1400 sq ft all-electric house uses about 750 kWh during a summer month and about 2700 kWh in the coldest winter month. If natural gas and ICE cars are to be phased out how much more electrical power will be needed?
Re: (Score:2)
All those things decrease the need for energy. EVs use much less energy than ICEs. Heat pumps use much less energy than gas furnaces. Electric stoves use much less energy than gas stoves.
But that's unrelated either to the story or to my post. We were talking about the energy used by data centers to run AI models. My point is that the amount of energy used is unrelated to the needs of any specific real world application. It's driven by companies competing for business. No amount of computing power is
Re: (Score:2)
You missed the difference between electrical energy and heat energy from burning things.
EVs use much more electrical energy than ICEs which use none other than what they generate themselves.
Electric stoves use much more electric energy than gas stoves. I had a gas stove once, it didn't even have an electrical connection. The electric stove is rated at 11.3 kW if all the elements are on. To use natural gas to generate electricity to be converted back to heat is downright stupid, but here there is no natural
Re: (Score:1)
>10 years ago people swore up and down the only way to get high end video games playable on a high end device was to move it all to a central terminal.
I doubt any significant amount of people actually said this, besides one or two cloud providers. People were researching raytracing soft-cores on FPGAs over 20 years ago, so they anticipated a future like the present. (see SaarCOR) Also it was easy to see the improvements in hardware that came every year.
As for local LLMs killing off datacenters? Probably.
Re: (Score:2)
40 years ago people swore up and down local PCs would never be fast enough for everyday tasks and we'd need to connect to central terminals to process it all.
20 years ago people swore up and down the only way to save money was to move storage from local to a central terminals to save it all.
10 years ago people swore up and down the only way to get high end video games playable on a high end device was to move it all to a central terminal.
A well aged straw man army (being 40 years old). You would have trouble finding any actual documentary support for any of these claims (that you just made up).
Re: Prediction: (Score:2)
Even if it's all true, it's all irrelevant. Not only are these problems not those problems but we don't have decades. It's small comfort though that the ultra wealthy won't get what they want because even if they killed off literally all of the rest of us, AGW would not stop.
Re: (Score:2)
..for you it'll run locally on your laptop soon enough, and all these AI centers will be trillions of dollars worth of garbage.
Yeah, but just think of all that sweet sweet taxpayer money that can be abused to pay those trillions in the meantime, along with the few hundred billion in yacht fees and private jet upkeep we'll get in executive bonuses. That's at least a dozen more lifetimes to add to the money burn pile. * chefs kiss *
- The Ones Who Already Know They'll Get Away With It
The gap will be closed (Score:5, Interesting)
... when the AI bubble bursts and 90% of the AI data centers go dark. The AI that remains will be the AI that is doing something useful enough to be worth the electricity it sucks down.
Mind the Meatsack Gap. (Score:2)
The AI that remains will be the AI that is doing something useful enough to be worth the electricity it sucks down.
If Bitcoin at $100K+ hasn't yet validated net worth against a planet-sized electrical tapeworm hanging out the ass end of a long blockchain, AI doesn't stand a chance at breaking even.
Quite frankly, by the time AI becomes "useful", it will have solved that energy problem. By turning us meatsacks into its power source. Like we taught it to do.
Re: Mind the Meatsack Gap. (Score:2)
Compared to cryptocurrency, LLMs are dramatically useful. They don't do just of the things their proponents claim they do, and probably never will, but at least they do some things. Cryptocurrency is only a way for a different group of assholes to get rich enabling crime. It does nothing you couldn't do with just another normal fiat currency.
Re: (Score:1)
Spam generation?
Surprised? (Score:1)
We have 8 billion real intelligences (Score:2)
Re: We have 8 billion real intelligences (Score:2)
On currents of stolen light .. (Score:1)
This wasn’t progress; it was a fever dream fueled by the singularity cult. A blinding, fanatical conviction that silicon would outthink flesh, that code would unmake soul, rewr
It will - when AI bubble bursts... (Score:5, Informative)
The demand is due to obscene amount of money being invested in AI.
But the return on this investment might not be that high.
This will burst the bubble and lower the energy demand.
Re: (Score:2)
The 'bubble' won't totally burst... even if it's proven not worth the hassle, one of the "AI" (we're decades away from Skynet) companies will keep tinkering and eventually get it to where the general public expects it to be.
The power demand won't go away... between the LLM-AI datacenters and EVs and the increasing reliance on computerizing everything (down to the fridge and the simple toaster on the kitchen counter), even with entire states covered by solar panels we're still going to need a ton more nuclea
Computing not AI, Time shift it. (Score:1)
The quite truth of the matter is the power curve did not change when AI become popular - it was already skyrocketing.
Computers have been using more and more power for a very long time, AI is merely the current buzzword we use for wanting more computing done.
But more importantly, what AI we do have can easily be time shifted. The AI corporations should scatter their processors around the world. Do not send all your processing to one AI cluster in the middle of your country. Instead have a simple routing m
Re: (Score:1)
But more importantly, what AI we do have can easily be time shifted. The AI corporations should scatter their processors around the world. Do not send all your processing to one AI cluster in the middle of your country. Instead have a simple routing machine that tracks the current cost of power in 20 different clusters scattered around the North America, South America, Europe, Asia, etc. When you get a request, send it to the processing where power is cheapest at the current time.
You wrote time shifting but what is described appears more like space shifting. While it would be possible, trivial even, to move the processing around as described so long as each location has a data center that is willing to go idle. Can you find a data center willing to go idle?
The hardware used in these AI data centers is very expensive, and to make buying this hardware profitable they need to be processing something all the time. So long as the electricity costs are low enough to allow for profit fr
Re: (Score:2)
If the designers have half a friggin' brain, that would be built into it... some sort of traffic management (datacenter 1 is busy, redirect to datacenter 2)... kinda like BOINC back in the day.
The datacenter doesn't "choose to go idle", there's no tasks waiting to be crunched, so it just sits idle waiting for something to come in... network/system management _should_ handle the distribution of tasks between all the datacenters... "the datacenter in New Delhi isn't busy, so we'll aim some traffic it's way" (
At what point will they admit it's just stupid? (Score:3)
This how AI is treating both the electric grid and the environment. People need to wake up.
At some point nuclear fission will look cheap. (Score:1)
Finally a news article on energy shortages where nuclear fission gets a favorable mention.
This problem of energy production is only going to get worse until the bullshit arguments against nuclear power are proven undeniably false and we start to see new nuclear power plants get built. The first argument that will likely fall is that nuclear power costs too much. Simple supply and demand will dictate that as demand rises and supply fails to meet it then prices will rise until that price increase discourage
scale (Score:2)
Tell me again how AI scales.
Until then (Score:2)
The world is scrambling ... "
It sounds like most of the scrambling is in the USA, as with most things based on computing. China has been investing in big infrastructure for a decade, they're not scrambling. India is starting a big infrastructure plan: They will probably be able to sustain local growth of generative AI.
In the meantime, people in the USA will have to live with less heating, air-conditioning and water.
Re: (Score:1)
It sounds like most of the scrambling is in the USA, as with most things based on computing. China has been investing in big infrastructure for a decade, they're not scrambling. India is starting a big infrastructure plan: They will probably be able to sustain local growth of generative AI.
In the meantime, people in the USA will have to live with less heating, air-conditioning and water.
The USA is a big place so putting a blanket statement on how people in the USA will act isn't likely to hold true. To put the size of the USA into some perspective consider it makes up about 2% of the global land area, and by some estimates 6% of the global habitable land area. The population is spread relatively thin with 4% of the global population.
Then is that the USA contains all kinds of different climates. Consider that if there's a shortage of water someplace then people can move to where there is
Re: (Score:2)
Burning more coal assumes there are coal plants still standing where they are needed.
Building in some industrial capacity assumes there's a factory left to build the widget.
We added 60GW not because it sounded fun... the climbing numbers of EVs and LLM-AI datacenters and increasing number of electronic devices in a household require more power. Solar could maybe work if we just cover all the Southern states _entirely_ in solar panels.
Remember, we can't burn the 460 billion short tons of coal under our feet
Re: (Score:2)
China is scrambling too. They just started another big hydroelectric project down south. They are building nuclear plants, solar, and wind turbines. They have not stopped burning coal either.
Re: (Score:2)
About China;
https://www.spglobal.com/commo... [spglobal.com]
Burn baby burn.
Work Stolen / Lives Ruined By AI Will Not Recover (Score:2)
Humans dont need this much energy (Score:2)
And the investments required to provide it make the tech unworkable. Thats why AI and data center companies are trying to foist those costs onto the consumers whose jobs they want to eradicate. Definitely not a winning strategy.
Have yet to see an AI use case that justifies this (Score:2)
What exactly do we get in response for this outrages expense?
AI coding assistance is nice but hardly essential and that's the most substantive use case for LLM AI that I am aware of. Who needs AI slop?
Re: (Score:2)
Re: (Score:1)
Exactly; just wait until the industry at large realizes that all those AI agents they've replaced their human staff with in order to cut costs are actually only cheaper than human equivalents while someone else is paying for the electricity.
deepseek? (Score:2)
Re: (Score:2)
Didn't deepseek prove that we can have AI at a fraction of the energy cost?
It dit, right before it discovered the Jevons paradox [wikipedia.org].
The Matrix predicted how this will work itself out (Score:2)
The human body generates more bio-electricity than a 120-volt battery and over 25,000 BTUs of body heat. Combined with a form of fusion, the machines have found all the energy they would ever need.
Mandatory Simpsons quote (Score:2)
Source [youtube.com]
Optical computers (Score:2)
Could solve a lot of this