Conor here:
worth it to clear cut 30% of the amazon https://t.co/YkrMvSEiKv
— Yasha Levine (@yashalevine) November 23, 2024
By Irina Slav, a writer for Oilprice.com with over a decade of experience writing on the oil and gas industry. Originally published Oil Price.
- Data centers are turning into an unexpected obstacle that may well compromise the whole transition offensive against hydrocarbons.
- Wind and solar could supply power to data centers for certain periods of a day, week, or month, but the bulk of the round-the-clock supply must come from baseload generation facilities.
- The part about the affordability of the energy system likely has to do with emissions rather than money.
Until about a year ago, no one paid much attention to data centers. Everyone used them, of course, but they didn’t think about them. Then, the AI rush began. It was followed by a rush for energy supply. One year in, and data centers are threatening the very energy transition on which so many governments have staked everything.
Power utilities, regulators, and climate activists appear to be experiencing growing concern about the immediate outlook for oil and gas demand, Reuters reported this week, saying the fast growth in demand for electricity caused by the mushrooming of data centers had come as a surprise to many. There may have been some frustration, too, because wind and solar have been unable to scale up fast enough to cover this additional demand, per some of the people Reuters spoke to—even though there is no scale that would have covered that demand, not for intermittents.
The worry is real, for sure. The Washington Post also published a worry-laden article this week about the increase in electricity demand due to data center proliferation and the risk this is posing to “decades of progress cutting greenhouse gas emissions, as utilities lay plans for scores of new gas power plants to meet soaring electricity demand.”
Data centers are turning into an unexpected obstacle that may well compromise the whole transition offensive against hydrocarbons. Data centers need reliable, uninterrupted electricity around the clock, and there is no way either wind or solar, even with battery backup, can guarantee this to the extent that data centers need. No wonder their operators are turning to gas and coal generators. They’re even planning to build new nuclear and revive old nuclear. The race for electricity supply is on.
Natural gas generators and natural gas producers are only too happy to oblige, of course. After years of depressed prices, natural gas drillers are due some respite. They see it in the surge in electricity demand driven by data centers and their new and much greater needs originating with AI development. Data centers have become more power hungry. Gas is the most easily available source of that power.
Last month, S&P Global estimated that growth in data centers could add between 3 billion cu ft and 6 billion cu ft in new daily gas demand to the U.S. total by 2030. “We believe that combination of data center demand and ongoing security concerns will underpin hydrocarbon revenues, and particularly natural gas demand, for at least the next decade,” S&P analysts said.
Indeed, the ratings agency expects electricity demand from data centers to grow at an annual rate of 12% over the next six years, which is certainly a healthy pace from natural gas producers’ perspective. And there is no realistic way this demand can be reliably met by wind and solar, the pillars of the energy transition. Yet this does not mean that pressure on data center operators to green-up is not growing.
“I think everyone agrees that we need more and more renewable energy to keep up with a growing demand,” Meta spokesman Jim Cullinan said recently, as quoted by Reuters. “I think it is up to the utilities to comment on how they will fill the supply.”
Passing the ball to the power utilities, however, is not going to change the situation. That situation is that wind and solar could supply power to data centers for certain periods of a day, week, or month, but the bulk of the round-the-clock supply must come from baseload generation facilities, emissions and all. With that, the data center situation gives us a taste of what the energy transition would actually look like—because the central idea of the transition is total electrification.
“Data centers are just a warm-up act compared to the amount of electrification we’re going to have going forward,” the chief executive of transition advocacy outlet RMI told Reuters. “And if our first instinct is to start building gas plants and nuclear plants in order to do that, we’re just going to create an energy system we cannot afford.”
The part about the affordability of the energy system likely has to do with emissions rather than money, but the first part about data centers being a warm-up is spot-on. The surge in electricity demand from these facilities is an excellent illustration of the insurmountable obstacles to the success of the energy transition as envisioned by its net-zero champions. It also exposes the shortcomings of wind and solar, and dispels the myth that they can replace, instead of complement, gas and coal generation.
Whatever Big Tech corporate spokespeople say, the fact is that their industry needs a reliable electricity supply, and the poster tech of the transition falls short on that very condition. One can either have green or reliable power supply. The sooner this gets acknowledged by the transition leaders, the better for everyone—including the planet.
This is really being driven by AI which are power hungry beasts in the order of 10x power load. We’d typically spec a server for 3-6kW power supplies but now it’s driven up to 50KW for a SINGLE high end system, never mind an entire rack. Cooling is then also being driven into advanced water cooling solutions to dissipate the heat generated (Also requiring power albeit less but adding to the whole system draw). Advances in CPU core consolidation / increased density and improvements in network bandwidth are offset by far higher GPU/DPU demands. Off the scale levels, really! Most Datacenters can’t cope and some have per rack limits of 7kW where I’m from. Ha!
Some of the possible responses to this crisis.
A breakthrough in energy production. (Don’t hold your breath. Any truly “new” energy source will have gigantic infrastructure demands to be met.)
A breakthrough in room temperature super conductive materials. (Again, don’t hold your breath. This is just as elusive as the fabulous cold fusion methods.)
A reduction in Data Centre energy demands. (This could be done through mandates on energy per byte rules and or the collapse of the AI bubble.)
Placing the Data Centres offworld. (There, solar panels to supply power for the Data Centres would be both easy to build and maintain and almost limitless in area.)
The Jackpot. (Fewer pesky Terran humans means less energy demand.)
Stay safe.
Ok, good list, so no, no, no, no again (gigantic infrastructure demands to be met), which leaves us with The Jackpot.
Stay safe indeed.
I was thinking of Saint Elon and his Space Cadets when I put forward the space based data centres idea. It is doable, and for a billionaire concerned about his or her posterity, this idea has possibilities.
When life hands you Lemons, make Lemon aides.
The data centre and electricity issue is hugely over exaggerated. The IEA gives a good breakdown with projections here. Hannah Richie also goes in and crunches the numbers.
We’ve been here before, with predictions that IT would gobble up all our energy supplies. The reality is that a combination of choke points in expansion and the increasing energy efficiency of all data technology ensures that it’s never as bad as anticipated.
The issue is almost entirely a local one within particular grids where there is a concentration of data centres. In global terms these are largely negligible – the equivalent of a handful of big thermal plants (or a few thousand acres of solar panels).
I can’t help thinking that stories like these are let loose in the WSJ etc in order to distract attention from the core issue with energy use – our grossly inefficient homes, transport systems, and our taste for too much meat. It’s easy to point to something like data centres and say ‘oh, there is nothing we can do, AI will eat up all the gains’. That simply isn’t true, its a side show and it distracts from the real issues.
Thanks. Still, given what you say is true, isn’t AI a big waste of time and resources? I can’t see the point of it.
I doubt much will come from AI, at least not open AI platforms, but it’s not my area of expertise. I’ve already heard of a few companies who were initially enthusiastic who have backed away once they’ve realised they’ve left themselves open to numerous legal and technical problems with AI generated work. But I suppose there are some real applications somewhere, I would guess though that they are not the data harvesting ones, more applications for robots, etc.
What is the average person to do when society writ large engages in the kind of delusional thinking that would get an individual committed to a mental institution? Or at least an intervention from concerned family members, followed by a strict schedule of meds?
There is simply no way any of the ridiculous scenarios of AI pan out. There are real-world limits on energy, plus the legal and ethical issues, just to start. But that won’t stop the Wall Street grifters from projecting out a pie-in-the-sky future where there are no limits on growth.
And I haven’t even mentioned climate change.
ICYMI yesterday: Huge gas plant eyed to power mystery $5B Louisiana data center. 2254 MW is nothing to sneeze at.
And from a search “According to the 2023 Integrated Resource Plan (IRP) for Entergy Louisiana, the peak load for the company was 9,874 MW.” The data center additions are a 23% increase in peak load. Keep in mind Entergy Louisiana already serves huge customers at dozens oil refineries, petrochemical plants and LNG plants. The hundred miles on the Mississippi River between Baton Rouge and New Orleans is referred to as the chemical/cancer corridor.
In NY State several retired coal plants have been converted to natural gas and have adjacent data centers, all with utility commission approval. The state mandates all new residential construction is electric and replacing fossil heating and cooking gets banned in 2030. There are minimal incentives for upgrading insulation of housing stock. My house has R19 walls and R30 roof; upgrading would cost over $100K easily.
Prices for electricity and natural gas are set at the margins. Building out base load generation to serve data centers can only increase marginal rates, which ultimately are paid by smaller industrial/commercial and residential users that don’t have contracted rate protection. Eventually the US will have EU energy prices; California and much of the northeast already have them.
A small price to pay for AI cat videos and deep fakes!
As my link indicates, there are major stresses from data centres, but they are localised, not worldwide. Hannah Richie explains this in more detail. There is nothing particularly new on this. The article is garbled – they would not be building a power station for a data centre, that makes no sense as what they need is reliable energy, which is why they are usually located where there are a number of circuits crossing. They would be co-locating a gas plant for peaking power supply, which is a common thing – this replaces the need for diesel or other back up. Plus, this is Kinsey talking their book, as the IEA goes through in detail, those projections are not realistic.
A 1500 MW plant is not a peaker. There are multiple peaker plants in south Louisiana and all are a few hundred MW and near loads. The location of this project is far from load centers. Entergy was a client old contacts confirm Meta is behind this monstrosity. Surely data center operators look for redundancy; they is why some in NYS are located at old coal plants bc of abundant transmission in multiple directions.
Yes these are localized issues. Localized in some of the most densely populated areas or energy intense regions like Louisiana. AI is just ramping up and this is what’s coming. The magnificent 7 are all domiciled in the US and the spooks love being traffic cops.
That would be the typical size of a modern peaker plant if its intended to take the full load in a mid sized system and a requirement for an immediate compensation for the loss of one or more major thermal plants. It all depends on local context and the likely demand from the data centre. From the data I’ve seen (in very different grids), such plants might only be used for a few days a year, usually when an unexpected outage meets a high peak demand.
Data centres absolutely should be located where there is redundancy, but for a range of reasons which I don’t fully understand the companies are very hung up on certain locations – its mostly so far as I know related to the reliability of grid supplies, and this tends to push towards areas already under a high load. They seem to calculate that the cost of these back up plants is worth it. In my experience, they are also often pairing data centres with gas powered plants precisely to minimise regulatory obstacles (i.e. submitting them as an integrated solution, not one-off developments).
The point about it being localised is that there aren’t actually many data centres being built in most of the world. It is a local issue in that it is creating very significant problems for particular grids – something I’m very aware of as I live in one of the worst affected areas and I see the rise every day in the online power trackers. But, as both the IEA and Richie go through in detail, its nowhere near the global issue that is being hyped.
I’m old enough to have lived through at least two of these ‘IT will eat up the worlds energy supplies’ scares. None came to pass for reasons Richie explains. She could be wrong, but nearly everyone I know in the business – and I know quite a few – think that the rise in demand has peaked already and is likely to balance out over the next decade or so.
PK, I read all your comments with great interest and respect your opinions. But you are simply wrong on this one. 1500 MW is not a typical peaker in the US that runs only a few days per year. $3.2B is a huge investment by Entergy. That is a base load plant and certainly if grid tied. The data center and Entergy plant site is 1400 acres FFS. Peakers are almost always near load centers and are a few hundred MW maximum.
Last week Virginia came out with energy usage numbers & data centers consumed 26% of all electricity use in the state. In my county. Prince William, data center growth is exploding & projections are that in less than 10 years we will have more than double the data center square footage of Loudoun, currently “the data center capital of the world.”
Art Berman & others are warning that the tight oil “revolution” is shortly coming to bitter end, implying that the glut of cheap natural gas will likewise.
My experience makes me agree with the first comment by Michael, and not yours, PlutoniumKun
Six to eight years ago I was getting a big UPS refurbed, and the tech said that at the big datacenter he had just refurbed, it used about one-third of the electricity it used for the last refurb … despite the march of computers taking over the world.
The reason was Virtualization, the ability to run many servers inside a single piece of hardware; while the compute power on that one piece of hardware is much higher than if it were running a single “instance”, having several (or several dozen) servers on a single piece of hardware is a huge electric input savings.
That was the Great Leap Forward in technology, which objectively improved the world.
So now that we are much more efficient, we can take more Great Leaps Forward; just like crypto mining erased and reversed efficiency gains of Moore’s Law and networking, AI is erasing and reversing the efficiency gains of Virtualization.
On Le Show yesterday Harry Shearer highlighted an article about how AI slows productivity, because of the extra steps required to “control” the result. The snappy AI pictures/video in the Anitidote du Jour in Links for the past few weeks are a fantastic result, which come at a price, learning and effort to steer the AI, and the energy input.
The energy input is not just the raw compute power, but extra code and extra servers to connect the AI to everything else (and I mean everything else, I am stunned how EVERY company of which I have any knowledge are already fully converted to AI for daily operations).
You’re correct of course that the electric supply nearest to the datacenter will supply the extra load, but all of our computers (personal and server farms) are working harder to supply the data for MachineLearning, and to insert AI as a middleman in every computer transaction, so the follow-on effects are significant.
Feels like a continuation of the cryptocurrency mining bubble. Probably kept the same plans and just replaced crypto with AI.
While she’s sort of right on this topic, I was deeply unimpressed by her book. She’s selling something that a lot of people desperately want to believe – that we can have our cake and eat it.
The problem facing us is far tougher than any of the Green New Deal types want to face. I’d maybe have some confidence that it could be done if people were honest about the scale of the problem, but they’re not. Instead we have people like Hannah pretending (or maybe believing) that the problem is far simpler than it is.
20% of existing energy usage is electricity. There’s a reason that the other 80% mostly isn’t electrical. Nobody in the greentech world even seems to be aware of this, let alone have any serious solutions to it. And that’s one of hundreds of problems.
I’m not a fan of her worldview (although I suspect this is more her ‘unique selling point’, rather than her true beliefs), but her number crunching is in my experience usually precise and well grounded and is properly contextualised.
I don’t know how you think nobody is focusing on electricity. Every major technical report on the problem I’ve read over 2-3 decades has emphasised the urgent need to electrify transport, infrastructure and heating and this has been central to published policy in most countries that I’m aware of. It is absolutely central to any form of transition.
The company I worked for in 1998-2002 made an absolute fortune on Y2k. All that fear, was rather dear.
“”Whatever Big Tech corporate spokespeople say, the fact is that their industry needs a reliable electricity supply, and the poster tech of the transition falls short on that very condition. One can either have green or reliable power supply. The sooner this gets acknowledged by the transition leaders, the better for everyone—including the planet.””
Uhh, the CA electeds are not listening and disagree. We have had 4 battery storage fires in San Diego County this year. But…
“”Siva Gunda, vice chair of the California Energy Commission, said policymakers have faith that advancements in battery chemistries will reduce the risk of thermal runaway and battery fires. “As we move forward, technology is really getting better,” he said. But at the same time, “we take (safety concerns) very, very seriously.”
So green or reliable or safe. Pick two!
“”Five years ago, the state counted a mere 770 megawatts of battery storage available to the California Independent System Operator, which manages the grid for 80 percent of the state and a small part of Nevada. By the end of this year, 10,379 megawatts are expected to be online and the state aims to grow that number to 52,000 megawatts by 2045.””
Land Use battles in CA reach runaway status. Housing shortage, data center shortage and peak use shortage vs electric cars, population growth and the elimination of natural gas use in the state.
I think the movie “the Matrix” might have a solution for the data center energy woes, and as a bonus, a solution for the immigration problem. The u.s. could utilize the influx of immigrants as a source of human energy just as the machines did in the Matrix. I see a win-win! And! We could extend this solution to include the homeless, those currently on welfare, and perhaps some tranche of the retirees drawing Social Security. This might also provide some relief to the growing mental health issues afflicting so many in our population — they could find a better life in the Matrix.
Give us your tired, your poor, your huddled masses, and the u.s. could add them to our own tired, poor, and huddled masses and apply them to solving the energy problems of the 21st Century.\s
I wonder what colour Soylent AI would be.
The writer makes this statement without backing it up: “ Data centers need reliable, uninterrupted electricity around the clock, and there is no way either wind or solar, even with battery backup, can guarantee this to the extent that data centers need.” I would like to see more proof the combination of renewables plus storage will not be sufficient. Especially when renewables are drawn from broad geographies, and storage is widespread. The emergence of mass vehicle-to-grid is also not considered. For limited times when this is not sufficient, yes, perhaps gas peakers. But emerging low-heat geothermal might also supply base load. I wonder if this is not more a case of utilities stuck in old mindsets going back to solutions to which they are accustomed. The climate crisis demands more innovative thinking.
With all due respect, it’s been extensively discussed that there will be a base load problem that is simply not solvable with wind or solar alone, even with storage. In the Northern Hemisphere, winter days are short, ambient light levels are lower, and it takes only a week of persistent clouds to cause problems. Plenty of articles have modeled that. So your demand for proof of a widely accepted problem is not reasonable.
The point which this article may not state clearly enough is that growth in data center demand will greatly increase base load requirements.
This issue is NOT “more innovative thinking”. It’s a lack of sufficient bloodymindedness in greatly restricting energy use, as in hard caps and rationing.
If you want to see a fairly rigorous analysis of why we can’t get by on renewables and storage alone, see this analysis by Roger Andrews at https://euanmearns.com/the-cost-of-wind-solar-power-batteries-included/.
Adding enough energy storage to get through “real-life” periods of bad weather is, to quote, “ruinously expensive”. Andrews studied scaled-up versions of both Germany’s and California’s renewable energy infrastructure and compared it to real-world demand to estimate the true energy storage requirements. LCOE estimates rose 10 to 20X (!!) once this storage was added.
For true 24/7 reliability, storage isn’t just a little add-on. It is (by far) the dominant cost of the system. Throw in real-world constraints on the critical minerals required, like lithium, nickel, and cobalt, and the conclusion becomes clear. It will never happen.
[Sigh…] I really miss the Energy Matters website at https://euanmearns.com/. Euan Mearns and Roger Andrews’ ability to analyze grids at the system level (and to properly account for supply vs demand mismatches in fine-grained time-domain analyses) was the best I’ve ever seen. Alas, updates stopped shortly after Andrews passed away in 2019.
Long-term storage via hydrogen might be one solution for much of our energy needs. There are numerous technical problems for utilizing it on a large scale, but hydrogen is essentially infinite, ie., using renewables to split water into green H and O2, with the hydrogen then returning to water when used to produce energy. Not so oil, gas, etc. Its also non-toxic, unlike methane. But of course, any realistic change to a sustainable economy is going to involve restricting our penchant for endless consumerism. And good luck with that.
I would suggest that those people who don‘t understand the limitations of renewables get a solar system and not just talk in the abstract about how they are going to save the world. EVERY solar system owner in higher latitudes is keenly aware of the fact that you have the highest energy use when the least comes in. It is ludicrous to think you can solve this with storage . Have you ever installed a system like that? The scale of resource input (copper & batteries) is mind blowing!!!!! Only rich westerners and fossil fuel subsidised resource production can afford us this. Even if you change your use patterns, ration your energy use AND pour western level salaries into it, this is at best a transitional technology for rich people to take the bite out of a transition to an energy starved future. Nothing to do with saving the planet. I produce in the summer about 3 times of what we use and in the winter even with substantial LiFe PO battery Banks am nowhere near able to even cover 70 %. The current expectations to computing is 99.999% uptime, you can‘t run payment systems in datacenters that maybe up or not. All of our IT is built from the ground on up, like a house of cards, on this kind of reliability and all the business processes driven by it on top. The level of complexity is a function of energy availability and stability. Why do you think as a developer and IT professional I make damn sure that i have a wood stove in my kitchen and a cistern in my garden? Having contributed to some of that tech garbage myself I am keenly aware of what a liability it is and would NEVER EVER entrust my daily needs such as food, water and heating to an IT driven system without having a completely manual alternative.