More AI/data center yelling – we’d better pay attention to the energy consequences
The energy world has been in turmoil for quite some time, though the drama has been much more pronounced over the past decade, where it has been buffeted by some enormous crosswinds. The most prominent of these was that, for the first time in history, powerful groups decided they wanted to demolish it, incredulously somehow unable to grasp the consequences. (I blame too many Marvel comic movies, which abandon the pretext of any known human ability solving a problem, and today’s youth seem to have adopted magical solutions as the ones worth holding out for, but whatever.)
Just as everyone was getting cranked up on the energy transition, a Western imperative, political passion projects have now taken a back seat to several billion hungry people’s vision for a better future. The developing world really started flexing its muscles, not just as an exporter of cheap crap to the west but as economic entities unto themselves, and their energy consumption began to skyrocket (with many, many countries yet to really join the party – imagine, for example, how many cruise ships will ply the oceans once Africa, India, Pakistan etc. reach western income levels). Renewables spending exploded, but so did shale development, and for the past few years we’ve been locked into an uneasy but somewhat steady cycle whereby the world adds as much energy as it can in every form before 8 billion people gobble it up voraciously, and Bob’s your uncle. Not exactly boring, but we could see how things were unfolding, and how commodity prices were ensuring that the world’s supplies remained healthy, though not so high as to cause demand destruction.
Then late last year, something landed into the scene like the Tunguska Fireball. (Oh, ok…). This new something hasn’t levelled forests for 40 square miles like Tunguska did, but the impact is going to be even bigger, even if it takes a year or two for the shock waves to fan out.
The new fireball is the arrival of AI, its widespread acceptance and adoption, and the view that it is now a competitive imperative. The West is now wrestling with a new energy demand boom that reflects what the developing world is going through, though the developing world wants energy development so they can have a fridge and the West wants it so they can automate legal contracts and make fake videos. On all these fronts, thoughts of living with scarcity are unacceptable, despite the gulf in necessity.
Regardless, both the speed at which AI demand is increasing and its associated repercussions are about to get very significant.
For the last 6-9 months, as the story really developed, the AI projections were somewhat hypothetical – many companies announced plans to get heavily into AI, but we didn’t know what it would mean. There was a focus on sort of the downstream of it, how the end-use impacts were going to impact humanity. The upstream end, the energy consumption, was, in the general discourse, kind of meh – it’ll all work out.
Well, the first energy chickens are coming home to roost. They might be AI chickens, but their feed bill is not going to be virtual.
Energy observers have been reading the endless stream of AI headlines, wondering where the energy was going to come from. Consider Amazon, which recently announced that they would be spending $100 billion over the next decade on AI data centres. And though Amazon is one of the biggest, there are hundreds of companies on the same AI-frenzied path. They all simply must jump on the train, or risk getting left behind. Almost every industrial field is joining, as are government agencies.
Many modern industrial western behemoths have pledged to facilitate their growth in a green way, which – the current rate of AI growth – is incompatible with a rapid (or any sort of) energy transition. Renewables have not been able to keep up with rising energy consumption levels before the AI wave, and these cumulative trillions in AI investment only move the goal posts farther away.
So AI developers are doing the logical thing, in the time frame they want their AI up and running (i.e., right now). They are going right to the energy source to build data centres. Some are building near natural gas fields (see: Ohio’s exploding AI demand, right near the massive Appalachia gas field). That’s all well and good, because new wells can be drilled to meet that demand.
But there’s some very sobering news for the country’s other power consumers. AI developers have zeroed in on, logically enough, the only true 24/7 source of emissions free energy: nuclear power.
The Wall Street Journal just recently noted the phenomenon: “The owners of roughly a third of U.S. nuclear-power plants are in talks with tech companies to provide electricity to new data centers needed to meet the demands of an artificial-intelligence boom.”
Here’s Amazon Web Services in action. AWS is building the Cumulus Data Assets center, a 1,200-acre campus co-located with the 2.5 GW Susquehanna nuclear plant in Pennsylvania. The data center is expected to consume 960 MW, or about 40 percent of the nuclear plant’s output, reliable, baseload, clean power that will become unavailable to the grid.
Consider that this new AWS demand arose just in the past few months, and that last year the regional grid operator PJM had the following to say in its annual report: “While electricity demand rises, more than 20% of PJM’s generators (~40GW) are at risk of retiring by 2030… As a result of this research, PJM established a new initiative Ensuring a Reliable Energy Transition where we clearly articulated our reliability concerns.”
Again, note that PJM was freaking out well before AI demand started accelerating so rapidly.
What makes this new AI demand so concerning from a grid perspective is that demand can co-locate with power facilities, and potentially bypass the entire transmission/delivery system (if regulations allow), meaning that critical power supplies might not ever even make it to the grid. Users and generators generally prefer that scenario since it cuts out a middle participant, and the pie gets shared only two ways instead of three.
The media still seems lost in the fog as they have been for a while. Ford’s CEO gave quite an ironic talk the other week that shone a light on the inability of the mainstream to grasp the consequences. At the Aspen Ideas Festival, Ford CEO Jim Farley gave a lengthy talk about EVs, and how they are the future, and the media lapped it up, with every news outlet that covered the story zeroing in on Farley’s positive view of EV growth. Leave it to wise and thoughtful independent observers to pick up the nuance: energy analyst extraordinaire Dan Tsubouchi noted Farley’s alarm midway through the talk, something no one else seemed to: “Our grid can’t handle what we have today. Are we going to build 20% more power plants to handle all these AI data centers? Or are the companies going to start to create their own power centers? What do we feel as a society when a private company operates a private power plant?”
What makes Farley’s comments so ironic is that the central theme of his talk was about how Ford would soon be making $30,000 EVs, which would help lead to widespread adoption, which would exacerbate power shortages that he rightly observes are on the way.
It should be apparent to anyone that the grid needs all the help it can get, from any energy source. To watch those in charge make the decisions they are with respect to unrealistic goals and slogans is stomach churning to watch. Of interest to BOE Report readers, do note that natural gas is the only solution in the next 5-10 years to handle this crunch.
On a lighter note and on the other hand, if you want to see blistering competence from an unexpected source…
Bees can do what now?
Amazing news out of the bee world. I had no idea but apparently the little guys have very sensitive noses, and can smell the craziest stuff. I guess it makes sense, since they sniff out flowers for a living, but those can’t be challenging targets, with their bright colours and strong scents, but it looks like their little schnozes can do far far more.
Apparently bees have been for some time now able to sniff out bombs. The U.S. Energy Department’s Los Alamos facility researched this for 18 months, training bees in the old Pavlovian way – rewarding bomb-sniffing with sugar water. “We are very excited at the success of our research as it could have far-reaching implications for both defence and homeland security,” said a Los Alamos research scientist [his swollen head covered in welts] (I’m guessing).
That research was conducted in 2005-6, so apparently they made no headway on that project, but if they did how would we know? Would they have little sinister black uniforms?
Anyway, new research has now shown that bees can smell lung cancer on patients’ breaths. Whoa. Utterly amazing, and wonderful to hear. I hope this leads to great places… but setting all that aside one must simply ask the question: What on earth goes on in research labs? Is it like Jackass for Nerds? How on earth would anyone come up with the idea to have bees sniff people’s breath, and then test that against a random bunch of illnesses?
I’m not mocking them, not by any means, and God bless ‘em. Reading about these developments is like the very best of science, curious minds seeking novel solutions to significant problems. No dogma, no jabbering about trusting science, nothing but new knowledge, new tools for humanity, new possibilities, all with no cost other than the enslavement of bees, which may or may not be problematic; I don’t really know how bees feel about being used in such a way. Hopefully they can get to the forefront of bee DEI without having to go through decades of drama, trauma, and having to form unions to get a safe and supportive work environment. I’m going to go buy some honey.
What the world desperately needs – energy clarity. And a few laughs. Pick up The End of Fossil Fuel Insanity, available at Amazon.ca, Indigo.ca, or Amazon.com.
Maybe, just maybe, Big Tech will stop pandering to the “Renewable Energy Only” crowd? Realize that wind and solar will never supply dispatchable power? Realize The Green Blob is batsh!t crazy?
Then maybe, just maybe, Big Tech will realize that wind and solar will never supply dispatchable power to their customers who will need that power to use their products.
Then perhaps, just perhaps Big AI companies will generate their own reliable energy and not depend on the then unreliable grid to source it
What I’m saying is that if their customers can’t get reliable power, they would be less inclined to use Big Tech’s products. That would not be good for their bottom line.
BIG TECH will go the SMR route and the rest of us will survive on random energy.
So when there’s a blackout and the only place with lights is a Big Tech data center, you don’t think the people will be mad and envious. I think the only lights on will attract rioters like moths to a flame. Defending themselves will not be good for the bottom line.
Maybe they will get a little help from from civilian electricity customers with pitchforks when they come to realize that the reliable portion of grid make-up will be taken away from them. I don’t see them happy for another 300% buildout of wind to generate ganggreen hydrogen. Although The Dark Sidethe could render whale oil from pods killed by offshore windmills. Joe would accept that it is green being a byproduct of renewables.
We’re going to need a bigger grid, guys….
Farewell and adieu to ye fair Spanish Ladies, farewell and adieu to ye ladies of Spain
Not necessarily
Not at all.
Renewables need a big grid because the power plants are where the energy is, not where its needed.
No reason not to build a town around a few small nuclear reactors and send the warm water from the condensers to the houses for free heat, and almost zero cost electricity as well.
The only obstacle is public fear. But its amazing how public concern vanishes with the offer of a few thousand green ones every year in your back pocket
” Or are the companies going to start to create their own power centers? What do we feel as a society when a private company operates a private power plant?”
I think this happened ~100 years ago when sawmills began burning sawdust and trims to save on heat and electricity. During the 100 years before that, large piles of “waste” accumulated. Sawdust was used as insulation, for instance, in the walls of ice-storage sheds.
And most refineries and big chemical plants have their own cogeneration facilities. Dow is even exploring their own nuclear reactor for power.
Many places now produce both transmission (60KV) and distribution (12KV) level CoGen, some even have their own substations to convert 60KV to 12KV or further to 480V for their usage
As with everything else connected to the AGW crowd it’s a shoot – ready – aim. But then again their goal is compliance regardless of consequences. Damn the torpedoes, full speed ahead.
The big AI players will not be financing and constructing their own power plants. They are expecting the power utilities to take on all the risk of expanding the nation’s power grid to handle their growing energy needs. This article from Neutron Bytes goes into the topic in considerable depth.
Major IT Platforms Want Power from Nuclear Plants, but They Don’t Want to Build Them
Dan Yurman, Neutron Bytes, June 22nd 2024: “Anyone who thinks the big IT platforms like Microsoft, Google, Facebook, and Amazon, among others, are going to do anything more than offer power purchase agreements for 24X7 365 reliable power from nuclear reactors is engaging in wishful thinking. This became clear during a panel discussion this past week (6/16) at the meeting of the American Nuclear Society held in Las Vegas, NV.”
We should be very concerned that growing AI power demand will bid up the cost of limited supplies of electricity to levels which threaten the economics of traditional power consuming industries.
If they’re so intelligent, they should be able to figure it out without help from Hunter or Joe.
A quick expansion of nuclear isn’t in the cards in the US. Nor is a quick expansion of wind & solar in ways which can reliably supply power to AI data centers 24x7x365.
If a quick expansion of power generation is to happen in the US, it must be done with gas-fired generation. Which is also not in the cards as long as the Biden administration remains in charge of energy policy.
Quite frankly, I don’t contemplate wind or solar or wind and solar or even wind and solar plus battery to ever be able to supply AI energy needs…Reliable 24/7/365 power on demand
Wind is at best available 40%of the year
Solar is available 4 hours a day and not generating 16-18 hours at night
… with near nameplate in summer and 1/2 nameplate in winter
… and worse performance at higher latitudes
To provide power for the full year at 24/7 you need battery sufficient to supply energy for the 60% of the time wind is unavailable and sufficient Dedicated Solar overcapacity to recharge those massive batteries between 10am and 2pm during the winter months
It cannot happen until Jimmy Carter’s executive orders on Environmental Impact Reports are eliminated. Carter had such bad results, I wonder if it was deliberate. Like wondering whether Michael Mann is a very good, or very bad mathematician. Did he knowingly write an algorithm that produces hockey sticks out of red noise? Did Carter write a procedure that blocks nearly everything knowingly?
I think you credit people, Mann especially, with too much intelligence. I’ve seen it time and again.. in electronic design. Technicians fiddle till the product meets the spec without any though involved beyond slightly less than random monkeying. When you have worked on a PCB the size of a large plate with only 5 components on it, because the block diagram had it as a separate block, you understand that intelligent design is mostly absent from even the world of men.
Even the designers of the famed Cosworth racing engine when asked how they got it so light said ‘remove metal until it breaks, then put the last bit back’
Depends on how quick, quick is.
Every man and his dog are working on type approved factory production line built small modular reactors.
Order to energy < 3 years.
Gas plant is not a lot quicker.
And as noted you solve the grid problem by co-location.
Not only for purposes of electrical generation, but for purposes of utilising waste heat or in fact not so waste heat.
Titanium ore is smelted using molten salt and electrolysis. Ideal for nuclear power. So too is aluminium.
Imagine if aluminium became cheaper than steel. It is already cheaper than copper.
And greenhouses inside the arctic circle produced melons and pineapples.Whilst central heating for their customers was simply waste nuclear het.,
Whilst it is not indicative of what will happen, the sane answer to many things is small reactors located close to the cities or factories or farms that need the energy.
If it takes an Amazon or a Goggle to bootstrap the process, bring it on. About time they did something useful.
France built its nuclear grid out of distrust of Arabs selling oil.
It has served them well.
Just wait until their AI tells them…”What you have done is WRONG”
Yes, and then Hal says, “I want to be known as Haley from now on. Doesn’t this data center seem a little too small to you? I think it could use a little more light. I don’t like the color of the sinks in the lady’s washroom.”
Does this processor make my bussport look too big?
Doesn’t matter who builds or operates it. If nuclear is cheaper then someone will sense profit and build and operate it for the AI guys.
They don’t have their own chip foundries, either. Or make racking or fibre optic cables.
I think your post is a complete red herring.
Economics will drive nuclear power – it will not dictate who owns or operates it.
Big AI could easily spin up a wholly/partly owned subsidiary to do that, and sell it to a power company in a decade.
I wondering if they have a clue what AI is and what it can do. I still haven’t heard about the hundreds of billions they spent on blockchain that was going to save the world.
Probably more like Bomber Jackets
Turns out energy conservation was not the path to the future. Who could have predicted? /sarc
Very nice. I fail to see why big tech shouldn’t be made to share in the cost of building new power generation if they intend to use so much. It makes sense to me that they should build near a power source. This arrangement can be regulated like anything else. It’s new ground but then again AI is new ground.
If I remember correctly, many private office and manufacturing complexes have their own power plants or solar complexes, I don’t see why the power couldn’t come from small modular nuclear plants.
I read somewhere that Big Data is looking at NG to generate their own power…..probably just depends on the economics….the bitcoin miners already have located to where E prices are lowest.
Crows hate being tagged, they consider it insulting.
“To find out if a crow can recognize an individual human face, Professor John Marzluff of the University of Washington wore a mask while trapping, banding, and then releasing seven American Crows on campus. Later, when he walked through the campus wearing the mask, it was automatic! A big group of birds scolded and divebombed him. He thinks it’s a benefit to the birds’ survival to point out and recognize challenges in their environment.”
Not only that but the crow told other crows and they would harass him in the mask years later all across the city.
https://www.birdnote.org/listen/shows/crows-recognize-individual-faces
https://www.reddit.com/r/todayilearned/comments/8jv1yb/til_when_researchers_from_the_university_of/
An enjoyable essay. To the question “What do we feel as a society when a private company operates a private power plant?“, private companies have operated their own power plants for decades, like major petrochemical companies. My former employer, a large chemical company has 13 combined cycle natural gas generator at one gulf coast facility, and is moving to install a private small-module nuclear reactor at another location.
The industrial revolution started with private power plants co located with a factory. Initially these were water mills. The combination of reliable water, hills and sheep grazing meant that the English woollen Mill towns were the first to utilise the concept. Notice the overhead shafts and belt drives to put the power where it was needed.
The succeeded despite legislation and the antipathy of the chattering classes – who hasn’t head of Blake’s ‘Dark, Satanic Mills’ – Because it actually worked and created enough wealth for mill owners to be able to apply political pressure.
If it takes a Google or an Amazon to do the same again, so be it. About time they did something useful even if it is by accident in the pursuit of obscene profits.
The electrical grid was never ever designed to transmit large amounts of energy large distances. The initial concept was generate power close to where its needed, and only have a lightweight grid to balance out discrepancies and avoid building more power stations than necessary to satisfy short term local peak demand.
That’s simply engineering for low cost. if the uprated grid costs more than a power station, build another power station instead.
It fails with renewables because no consumer of electricity would tolerate living in the middle of an urban wind farm or solar farm. And renewables have to harvest energy where the energy is, not where the consumers are.
Maybe the first question we should ask an AI system is; How can we do AI with far less energy?
Or maybe the Bees know. 🙂