Whether a data centre is a "good" or "bad" thing ultimately depends on the value of the services it provides to society vs the environmental damage it leads to. Reducing the environmental damage is one thing, but as David says here, efficiencies are likely to translate into increased demand. The question is - increased demand for what? AI can be used for valuable things e.g. medical research etc. The problem is that the vast majority of AI will actually be used for the same thing most of the rest of the internet has evolved into - the advertising industry and the political manipulation industry. The more energy AI-driven advertising, algorithms and manipulation get, the more energy they will use, and the environmental impact, of the energy use, of the building of the data centres, and of the unneccessary mass consumption, air frieght of fast consumer goods etc, which is driven by the advertising, will be disregarded, as it has been to date. Unless we can have a radically different way of permitting and providing energy to data centres which involves transparency to society about the real value they are providing in society - what exactly those servers in those corridors are doing, for whom and for what. Society needs transparency on this so that society can make informed decisions about what deserves energy and what doesn't.
Perhaps before worrying about the increased in electricity demand from AI data centers, we should first look at the business model of Open AI and the rest of them. We need to understand how these companies are going to make a profit with these astronomical investments. So far this is not being explained by the AI companies and all we have is smoke and mirrors.
> efficiencies are likely to translate into increased demand
I believe this is known as the [Jevons Paradox](https://en.wikipedia.org/wiki/Jevons_paradox). It seems like a general problem with trying to solve some of the energy transition problems using efficiency. It doesn't always apply (there is a limit to how much I can run my heat pump!) but I agree that without policy constraints, savings here will simply be used to try to capture attention.
One of the most important laws we have had since 1970 is the National Environmental Policy Act (NEPA), which requires federal agencies to analyze the potential impacts of their proposed actions, come up with a reasonable range of alternatives, and include the public in the decision-making process. NEPA created the Council on Environmental Quality (CEQ) to oversee the act's implementation. The CEQ promulgated regulations effective in 1978. The regulations were stable until July 2020 when they were rewritten in such a way as to ignore the purposes and intent of the act, which was to prevent or eliminate damage to the environment. As of April 11, 2025, the CEQ rescinded its regulations as per the instructions of Executive Order 14154, Unleashing American Energy, January 20, 2025. Agencies are now on their own to create NEPA implementing guidance, not regulations, and the administration has issued a template based on the 2020 CEQ regulations to help them. NEPA helps us make informed decisions. The NEPA statute is still here but somewhat weakened with the recission of the CEQ regulations and the addition of Section 112 included in the reconciliation bill that passed both houses of congress and was signed into law on July 4, 2025.
David’s valiant effort to highlight that this *could* allow greener energy for data centers finally paid off after the guest made abundantly clear to prospective customers that fossil fuel add-ons are still his go-to. The challenges of highlighting tech driven by profit rather than public good.
I recently heard a discussion on another podcast (Catalyst) about the many things constraining grid expansion in the near term, and it made me think of the previous Volts podcast about the question of biofuels as necessarily being bad for the environment because land is a finite resource. In this analogy, the grid is a limited resource in the short term, so since we ALSO want to electrify industry and transportation and heating, even smart, demand sensitive AI is going to crowd out other uses we want.
Whether a data centre is a "good" or "bad" thing ultimately depends on the value of the services it provides to society vs the environmental damage it leads to. Reducing the environmental damage is one thing, but as David says here, efficiencies are likely to translate into increased demand. The question is - increased demand for what? AI can be used for valuable things e.g. medical research etc. The problem is that the vast majority of AI will actually be used for the same thing most of the rest of the internet has evolved into - the advertising industry and the political manipulation industry. The more energy AI-driven advertising, algorithms and manipulation get, the more energy they will use, and the environmental impact, of the energy use, of the building of the data centres, and of the unneccessary mass consumption, air frieght of fast consumer goods etc, which is driven by the advertising, will be disregarded, as it has been to date. Unless we can have a radically different way of permitting and providing energy to data centres which involves transparency to society about the real value they are providing in society - what exactly those servers in those corridors are doing, for whom and for what. Society needs transparency on this so that society can make informed decisions about what deserves energy and what doesn't.
Perhaps before worrying about the increased in electricity demand from AI data centers, we should first look at the business model of Open AI and the rest of them. We need to understand how these companies are going to make a profit with these astronomical investments. So far this is not being explained by the AI companies and all we have is smoke and mirrors.
> efficiencies are likely to translate into increased demand
I believe this is known as the [Jevons Paradox](https://en.wikipedia.org/wiki/Jevons_paradox). It seems like a general problem with trying to solve some of the energy transition problems using efficiency. It doesn't always apply (there is a limit to how much I can run my heat pump!) but I agree that without policy constraints, savings here will simply be used to try to capture attention.
One of the most important laws we have had since 1970 is the National Environmental Policy Act (NEPA), which requires federal agencies to analyze the potential impacts of their proposed actions, come up with a reasonable range of alternatives, and include the public in the decision-making process. NEPA created the Council on Environmental Quality (CEQ) to oversee the act's implementation. The CEQ promulgated regulations effective in 1978. The regulations were stable until July 2020 when they were rewritten in such a way as to ignore the purposes and intent of the act, which was to prevent or eliminate damage to the environment. As of April 11, 2025, the CEQ rescinded its regulations as per the instructions of Executive Order 14154, Unleashing American Energy, January 20, 2025. Agencies are now on their own to create NEPA implementing guidance, not regulations, and the administration has issued a template based on the 2020 CEQ regulations to help them. NEPA helps us make informed decisions. The NEPA statute is still here but somewhat weakened with the recission of the CEQ regulations and the addition of Section 112 included in the reconciliation bill that passed both houses of congress and was signed into law on July 4, 2025.
David’s valiant effort to highlight that this *could* allow greener energy for data centers finally paid off after the guest made abundantly clear to prospective customers that fossil fuel add-ons are still his go-to. The challenges of highlighting tech driven by profit rather than public good.
My goodness, almost an hours worth of discussion on powering data centers, and uninterruptible power supplies were never mentioned.
I recently heard a discussion on another podcast (Catalyst) about the many things constraining grid expansion in the near term, and it made me think of the previous Volts podcast about the question of biofuels as necessarily being bad for the environment because land is a finite resource. In this analogy, the grid is a limited resource in the short term, so since we ALSO want to electrify industry and transportation and heating, even smart, demand sensitive AI is going to crowd out other uses we want.