4 ms·

It's not the power demand that is the problem. It's that the majority of AI deployments are happening in a country which has a has had very poor renewable adop

It's not the power demand that is the problem. 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...

dangus 9mo ago
Exactly this. Powering all AI data centers with renewable energy is actually trivially easy. You could even legislate it and make big tech companies responsible for providing the power themselves. One stroke of the pen resolves the issue. If OpenAI can afford to “spend $1 trillion” on AI they can afford to build some wind/solar/battery power plants.
- badgersnake 9mo ago
  Not if your government refuses to let you build any renewable capacity.
  - dangus 9mo ago
    Hence the stroke of the pen. That’s all a policy choice.
- nradov 9mo ago
  In what sense is it trivially easy? The battery supply chain is still backlogged and will be for years to come.
  - dangus 9mo ago
    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?
    - hvb2 9mo ago
      > 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
      - arghwhat 9mo ago
        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.
        hvb2 9mo ago
        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
        arghwhat 9mo ago
        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.)
- exabrial 9mo ago
  That’s power that could have been used to shut off coal plants. Instead now you’ve extended their lives.
- schiffern 9mo ago
  >You could even legislate it Spoiler alert: "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."
cmiles8 9mo ago
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.
- arghwhat 9mo ago
  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.

dangus 9mo ago

Exactly this. Powering all AI data centers with renewable energy is actually trivially easy. You could even legislate it and make big tech companies responsible for providing the power themselves. One stroke of the pen resolves the issue. If OpenAI can afford to “spend $1 trillion” on AI they can afford to build some wind/solar/battery power plants.

badgersnake 9mo ago
Not if your government refuses to let you build any renewable capacity.
- dangus 9mo ago
  Hence the stroke of the pen. That’s all a policy choice.
nradov 9mo ago
In what sense is it trivially easy? The battery supply chain is still backlogged and will be for years to come.
- dangus 9mo ago
  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?
  - hvb2 9mo ago
    > 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
    - arghwhat 9mo ago
      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.
      - hvb2 9mo ago
        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
        arghwhat 9mo ago
        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.)
exabrial 9mo ago
That’s power that could have been used to shut off coal plants. Instead now you’ve extended their lives.
schiffern 9mo ago
>You could even legislate it Spoiler alert: "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."

cmiles8 9mo ago

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.

arghwhat 9mo ago
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.