All that power has to come from somewhere. The idea that all this AI is powered by “green” energy and unicorn farts is just a bunch of PR puffery from tech companies trying to divert attention from the environmental damage they’re causing.
The uncomfortable truth is that AI is the biggest setback on our path to energy sustainability we’ve seen in a generation.
We can power it all and then some with renewable and nuclear energy. We elected a regime openly hostile to that and openly pro fossil fuel. Like they literally ran on burning more coal, so it shouldn’t be a surprise that we are burning more coal.
AI doesn’t matter. If it’s not AI it’ll be EVs. Or if you’re pro immigration (as I am) then what do you think letting more people into the country does for power demand? It’s something like 5kW averaged out over 24/7 per head. That’s probably conservative when you do a full accounting of all demand per head. Every new immigrant is probably equivalent to a rack of GPUs.
Degrowth is political fantasy. It will establish a populist backlash every time. Or are you going to line up to be the first to become poorer?
I look at that stuff as a very privileged fantasy. Only the rich can romanticize poverty. The people who fantasize about green back to the land scenarios are usually wealthy middle or upper class people in developed nations who have zero first hand experience of what that actually means outside the Avatar films.
It is, but degrowth is an election losing proposition. Any talk like this needs to be transparently non-hostile to demand for political purposes. The solution should be something like requiring them to build nuclear or renewable energy, or tax them and put the money into a subsidy fund for clean energy.
By the time a nuclear plant comes online, Renewables have incrementally added 400 gigawatts. Granted, nukes generate 4 to 8 times more energy, but solar can significantly improve crop yields and soil health. They also make it easy to raise sheep and cattle. It's a good thing I like lamb (yum).
I favor public education, but let's not kid ourselves, there is not a polity on earth where degrowth would get more than 20% support. It's a weird social media echo chamber artefact that will exclusively sabotage efforts to decarbonize.
In a sane election system, 20% gives a party a significant position in the government that influences the coalition and drives some of the future decisions. Just not in the two-party circus.
> In a sane election system, 20% gives a party a significant position in the government that influences the coalition and drives some of the future decisions. Just not in the two-party circus.
There is no consensus among political scientists that either a two-party system or a multi-party/coalition system is inherently “better.” Each design produces different trade-offs in representation, stability, accountability, and policy outcomes.
...or everyone else decides to marginalize that 20% party and allies with the far right instead (I don't want to defend the US system, but proportional representation is not a panacea either).
> there is not a polity on earth where degrowth would get more than 20% support
Eh, I'm not so sure about that. Sustainability politics is mainstream in Europe in a way it isn't in the US. Aside from ethical concerns, a lot of people over here see climate change as a very real economic threat (likely to cause them material economic harm within their lifetimes).
You're probably right that a general degrowth strategy wouldn't ever be popular, but I bet a policy that say restricted AI and cryptocurrency with the aim of reducing electricity prices would be.
> You're probably right that a general degrowth strategy wouldn't ever be popular, but I bet a policy that say restricted AI and cryptocurrency with the aim of reducing electricity prices would be.
That's arbitrary. If you went back in time before AI and crypto, which industries would you pick to constrain growth or development of?
Is it whatever the latest industry is that is driving incremental emissions? If so, I don't know that it is a compelling mental model, because that is a degrowth mindset.
Intentionally reducing quality of life in the short term will never win elections, no matter how educated a populace is. The best strategy to reduce consumption that seems to be working is allowing below replacement total fertility rates.
>> But then you get an aging population and all the problems that that brings with it.
> Only for a generation (mostly \s but entirely true).
That's not accurate. The problems of population aging are not confined to a single generation. They are structural and persistent, unless the underlying institutions adapt.
Aging is a continuing demographic process, not a single event. Once a society enters sustained low fertility and longer life expectancy, each cohort is smaller than the one before it. Each cohort also lives longer. That means that today's workers support more retirees. Tomorrow's workers will support even more, unless something changes.
It can feel (but isn't) like a single generation problem if major structural changes happen like: raising retirement age in line with life expectancy, shifting pensions to funded, large-scale immigration, major productivity gains from technology, or cultural shifts to high fertility.
> Once a society enters sustained low fertility and longer life expectancy, each cohort is smaller than the one before it.
I mean, unless fertility completely collapses (to like less than 0.5) then it'll mostly be a single generation problem. Regardless of any future changes, the current generation (my kids etc) will be supporting a much larger older cohort, with problems arising from that. I am one of 4 siblings, have two kids, and as long as both of them have two kids, no more problems arise (obviously extrapolating to the population).
There's some amount of irreducible demand for kids so I'd be surprised to see TFR continue to decline on a generational basis. Mind you, I could be wrong (or alternatively, we could see a massive increase in TFR like we did post WW2).
Yep, someone was bragging recently they used 13B tokens last year. At 8mg CO2/token that's ~100t of CO2. Consumption of 5 households (or 200 NYC-London flights) just for vibe coding!
Thank you for acknowledging the elephant in the room. I've literally seen people on HN argue that AI's increased power demand isn't bad for climate goals, because the money will encourage renewables.
It's astounding how people don't see it, even when it's the invisible hand of the market that's choking them to death.
It's that the majority of AI deployments are happening in a country which has a has had very poor renewable adoption and is now actively sabotaging renewable projects with an active opposition to climate goals because a particular group wants to protect their existing revenue.
Renewables are cheap and highly profitable, and money talks - even in the US, as can be seen in Texas. But it's hard to fight against your government when they want to force you to buy their rich friends' fossil fuels instead...
This is a pretty gross mis characterization of what’s happening. There’s been a lot written about the fluff that is a lot of these AI company “purchases” of “green” energy. In practice there’s no way to get that power from (insert middle of nowhere location with green energy plant) to (insert location of AI datacenter) so to actually power the data center the utility is forced to power on some clunky old coal plant to keep the chips powered.
The AI company is issuing press releases saying how they bought all this clean power but in practice they just forced some old clunky power plants back online to meet their demand.
What your are describing is purchasing certificates from renewable energy vendors, which while technically a small investment (more money to the renewable energy vendor → renewable business growth → more renewable energy projects) has very little to do with renewable energy projects like those I was talking about.
It is technically possible for the AI companies to decide to become self-sufficient or enter into the energy production market if things tilt far enough in favor of that, but it is somewhat unlikely and unexpected.
Big renewable projects are run by electricity producers, not consumers, and they are the ones being actively sabotaged in all sorts of ways.
"At BigGridCo we're proud to switch AI to 100% renewable power. On paper we just send all the dirty power to (scoffs) pesky houses and industry, leaving the clean power for AI."
Nuclear power works too, it’s clean and low carbon impact.
Can Microsoft and Google not afford to build a battery factory or nuclear power plant? Are they broke or something?
Why is the solution to scarcity of supply to bend over backwards and roll back regulations? The scarcity of supply itself should be a hint to society to stop supporting unfettered growth. Or maybe these mega-corporations need to get over it and pay fair market value for the projects they want to build.
Why do we have to breathe coal power emissions so that we can have one more ChatGPT wrapper nobody asked for?
> Nuclear power works too, it’s clean and low carbon impact.
You want an AI company to invest in a project that takes decades to complete? What are the chances they're around when it completes and what powers their datacenters while that takes place
Just to be pedantic: The median construction time is 7 years. With very slow planning, it is a decade, not decades. It can be done faster though.
Our power consumption won't be going down, and it generally wouldn't be the AI company itself running the project but the electricity companies that earn money supplying power that see dollar signs in all that extra electricity consumption.
Even if the AI companies all die, our global electricity consumption will keep going up margins will be better than the retired plants, so it's a good investment regardless.
I think you should look up actual construction times on reactors in developed countries. Be VERY happy if you can do it in less than 15 years.
> Even if the AI companies all die, our global electricity consumption will keep going up margins will be better than the retired plants, so it's a good investment regardless
If the company putting up the money goes bankrupt, what happens to the project? Maybe it's picked up by someone else?
I think AI companies should try to make it to 2030, my guess is at least a few of them won't make it. Don't commit to projects that won't even complete in the 2030s
I think you should look that up. I was even being conservative: Korea and China seems to be managing consistently around the 6 year time scale, while Japan has done it in less than four years from construction start till operation.
Granted, the US would have to import professionals to do it at that speed, and politicians will of course try to hinder the process with endless bureaucracy as their sponsors would rather sell fossil fuels...
> If the company putting up the money goes bankrupt, what happens to the project?
If people didn't start such medium-length projects out of fear of hypothetical future bankruptcy, there would never have been any infrastructure projects. Investors do not worry about them going bankrupt, they worry about losing momentum and would generally rather light money on fire than stagnate. We live in a time where business people start space programs out of bloody boredom.
However, what happens in these cases is just that other investors flock the carcass and takes over for cheap, allowing them to reap the benefits without having to have footed the whole bill themselves. Bankruptcy is not closure for a company, but a restructuring often under new ownership.
The only realistic scenario where such project would be dropped is if the world situation changed enough such that it would no longer be considered profitable to complete, such as due to other technology massively leapfrogging it to the point where investing in that from scratch is better than continuing investment, or demand being entirely gone such that the finished plant would be unproductive. Otherwise the project would at most change hands until it was operational.
(Particular AI companies making it to 2030 is not really that important when it is electricity producers making these investments and running these projects to earn money from AI companies, EV charging, heatpumps, etc.)
Finland, Olkiluoto, license application 2000, construction started in 2005, planned operation in 2010, actual operation 2023.
France Flammanville 3, construction started in 2007, planned operation in 2012, actual operation 2024, so 17 years
Hinkley point UK, construction began 2017 projected commissioning is in 2029/2030.
Vogtle USA, permits 2006, construction started 2013, operation 2023/2024.
South Korea, shin kori 3 and 4 took 7 and 10 years. And those aren't new designs.
Japan, the newest commissioned reactor is from 1997? Sure, France built really fast in the 80s... Different requirements/rules/public opinion.
And this is all from the start of construction. The beginning of the project is actually waaaay before that.
Please send me some links when you've done your research to prove me wrong. And yes, I did leave out china because I don't see the us building a Chinese design reactor... And even if that was possible it wouldn't meet us standards so you can effectively start over.
> If people didn't start such medium-length projects out of fear of hypothetical future bankruptcy, there would never have been any infrastructure projects.
And who finances that? Not banks by themselves, governments always have to give out some loan guarantees or favorable treatment. No private investor can deal with that amount of risk. So the bureaucracy that you speak of, without it no nuclear plant would exist.
So why don't you point me to a commercial nuclear power plant that was privately funded without loan guarantees by a government and all of that.
> We live in a time where business people start space programs out of bloody boredom.
So if you're referring to SpaceX, no Musk started that to make life multi planetary. And he understood that no one will finance that so the company needs to first make money to finance that mars shot.
Bezos I'm less familiar with but I know he has a collection of space artifacts so I think it's an interest of his and he probably wants to show he can do what musk can.
Google/MS/Meta will be around, probably. The other AI companies? Certainly not all of them.
I wouldn't rule out the current expenditure on AI to be a risk to the big players either. They're putting so much money in this. And with all the off balance sheet tricks that are happening now it'll be hard to know the real exposure.
Again it's a supply chain problem. Regardless of how much cash you have, you can't just order a new battery factory or nuclear power plant and have it up and producing in a couple years. We have eviscerated our supply chains for those things and no matter how much money we throw at the problem now it's going to take decades to reindustrialize. Rome wasn't built in a day.
If the concern is over externalities such as CO2 emissions and other types of pollution then sure, let's tax those directly. That will help accelerate solutions through free market mechanisms.
It's the same song as with crypto. Just as silly as then - of course many people will burn whatever is the cheapest fuel right now, even if they maybe invest in something else in the future. But the total goes up anyway.
>The idea that all this AI is powered by “green” energy and unicorn farts is just a bunch of PR puffery from tech companies trying to divert attention from the environmental damage they’re causing.
Do we have a solid breakdown ala Our World in Data for the energy mix mused to power AI Datacenters?
The only thing I have seen is the facility that Musk acquired in Memphis for Grok is illegaily emitting more pollutants than allowed because of Musk's insane drive for speed and it is causing health problems in the underlying poor community.
Its the reason I will never use Grok but i've been curious about where ChatGPT, Claude and Gemini are hosted. Google has had a history of efficient data centers and they are running custom silicon so i'd assume they are the best here?
I don't like the social harms related to AI but I think the energy is a silly emphasis. No one has ever thought twice about any heavy industry or absurdist garbage for consumers, home heating, etc.
If we were on track for everything else a serious uptake of AI might have put us barely off track.. But this is like blaming the wafer thin mint for the fat guy exploding.
I think it's still worthwhile, though. AI, given its current trajectory, will be able to help immensely with science and engineering challenges. Degrowth isn't a recipe for sustainable reduction of CO2 emissions.
The big engineering challenges right now are electrifying everything (which means convincing people that it's the right thing to do and that gas powered vehicles belong to the trashbin of history, amongst others) and banning production of "virgin" plastic items, especially single use items (which also required a whole lot of convincing).
Most of that is convincing is done in the exact opposite direction with... you guessed it... AI.
Pumping even more CO2 into the air hoping the magic box spits out a solution to remove the CO2 from the air doesn't seem like a sustainable recipe either.
This is broadly more PR puffery. We don’t need some magic AI model to tell us how to cut emissions. We just need to execute things we already know work.
All that power has to come from somewhere. The idea that all this AI is powered by “green” energy and unicorn farts is just a bunch of PR puffery from tech companies trying to divert attention from the environmental damage they’re causing.
The uncomfortable truth is that AI is the biggest setback on our path to energy sustainability we’ve seen in a generation.