Two issues not addressed. First, do we really need all this compute? Probably we do for most mundane business services. But do we need crypto? Agentic AI, super AI, the surveillance state? Second, what is powering the grid that is charging the batteries? It seems to me that if there is no commitment to making sure the power to run these data centers is zero carbon then this is just a strategy to move the responsibility for climate destruction onto the backs of the utilities. Perhaps it will be possible to coerce the utilities to install zero carbon to provide the needed power on their grids, but the record to date is not encouraging.
As a person living close to the proposed data center this company wants to build in Salem Oregon. I am very disappointed that you gave this air time. Ask him about the NDA’s he had city staff sign. He is not exactly lying to you. But pretty close.
Loved the moment where 12 million gallons became "three Cheesecake Factories." That's exactly the kind of number people can't feel, so I ran it through a perspective tool I've been building. It's about 18 Olympic pools, 24 municipal water towers, or the fuel in about 200 Boeing 747 tanks. My favorite: it's just 2.4% of what Los Angeles uses in a single day, so a full year of this data center's water equals roughly 35 minutes of LA's. https://app.datapalatable.ai/p/264c7328 Seems like data center developers will win or lose these local fights partly on whether residents can picture the numbers.
I understand the objection to high water use in the desert, but evaporative cooling is much more efficient cooling so air-cooled vs water cooled chillers is a choice to use more energy rather than more water, which makes no sense for the grid, or especially in Detroit, when you're sitting next to a lake. Leaning into air-cooled chillers everywhere feels like the wrong response to the bad PR around water-use. Also the cheesecake factory analogy is terrible. What does that even mean? Who knows how big a cheesecake factory is? Do cheesecake factories use 12 million gallons per day? I'd rather know more detail on the golf-course or agriculture water use comparisons.
Secondly, I loved the framing "this is no longer industrial real estate that happens to consume power. It’s really an energy asset that happens to consume real estate." The next question then should be "how should the real-estate neighbors make use of that energy asset?" Segue to a discussion of thermal energy networks... In cold climates, new data centers are a thermal asset that needs to be harvested. It doesn't need to be waste-heat, new data centers should plan in order to provide future connection of district heating and thermal energy networks.
Verrus is answering every electron objection — batteries instead of diesels, closed-loop cooling, flexible load that holds down power prices — and the backlash shows up anyway, which says the real constraint was never the machine. It's the permission to site the machine: scarcity rent rotating from the grid to the permit office, the slowest layer for capital to resolve. In The Thermodynamics of Thought I trace why inference gets repriced as infrastructure rather than software, and why the 3–7 year interconnect queue is where that physics gap gets paid. https://seldondance.substack.com/p/the-thermodynamics-of-thought #DataCenters #AIEnergy
This is such an important piece of the grid conversation because I think we often frame the problem as “we need more electricity” when the more interesting question is when we need it.
A more flexible grid could change the equation quite a bit. If some of the biggest new loads, especially data centers and other flexible demand can respond to what the grid is actually doing, we may not have to solve every demand spike by simply building more generation.
I’ve been looking at the AI–energy connection from the climate side, and this is one of the pieces I find most fascinating: AI is creating enormous new electricity demand, but it could also push us to rethink how flexible the electricity system itself needs to become.
The grid isn’t just a supply problem. It’s increasingly a timing problem.
The four hour battery plus indefinite compute scale down is a bet on the shape of the emergency, not only its length. RTO demand response strips are two to four hours because that is what the programme pays for, not because that is what a cold snap does, and Uri ran for days. Day two and day three are when the interconnection agreement gets read out loud. The part I would want in writing is who eats the shortfall when a shed event overruns the strip, the tenant taking an SLA credit or the developer carrying the capacity penalty, because that allocation decides whether this is a grid asset or a very expensive option the utility was handed for free. Interruptible industrial tariffs solved a version of this decades ago, and the smelters that took them learned that the curtailment clause is the whole contract.
Two issues not addressed. First, do we really need all this compute? Probably we do for most mundane business services. But do we need crypto? Agentic AI, super AI, the surveillance state? Second, what is powering the grid that is charging the batteries? It seems to me that if there is no commitment to making sure the power to run these data centers is zero carbon then this is just a strategy to move the responsibility for climate destruction onto the backs of the utilities. Perhaps it will be possible to coerce the utilities to install zero carbon to provide the needed power on their grids, but the record to date is not encouraging.
As a person living close to the proposed data center this company wants to build in Salem Oregon. I am very disappointed that you gave this air time. Ask him about the NDA’s he had city staff sign. He is not exactly lying to you. But pretty close.
What is he lying about, specifically?
Loved the moment where 12 million gallons became "three Cheesecake Factories." That's exactly the kind of number people can't feel, so I ran it through a perspective tool I've been building. It's about 18 Olympic pools, 24 municipal water towers, or the fuel in about 200 Boeing 747 tanks. My favorite: it's just 2.4% of what Los Angeles uses in a single day, so a full year of this data center's water equals roughly 35 minutes of LA's. https://app.datapalatable.ai/p/264c7328 Seems like data center developers will win or lose these local fights partly on whether residents can picture the numbers.
I understand the objection to high water use in the desert, but evaporative cooling is much more efficient cooling so air-cooled vs water cooled chillers is a choice to use more energy rather than more water, which makes no sense for the grid, or especially in Detroit, when you're sitting next to a lake. Leaning into air-cooled chillers everywhere feels like the wrong response to the bad PR around water-use. Also the cheesecake factory analogy is terrible. What does that even mean? Who knows how big a cheesecake factory is? Do cheesecake factories use 12 million gallons per day? I'd rather know more detail on the golf-course or agriculture water use comparisons.
Secondly, I loved the framing "this is no longer industrial real estate that happens to consume power. It’s really an energy asset that happens to consume real estate." The next question then should be "how should the real-estate neighbors make use of that energy asset?" Segue to a discussion of thermal energy networks... In cold climates, new data centers are a thermal asset that needs to be harvested. It doesn't need to be waste-heat, new data centers should plan in order to provide future connection of district heating and thermal energy networks.
You may want to check out this framework for building a more ecologically sound, socially just, and sustainable data center that goes beyond energy and water. https://www.biohabitats.com/wp-content/uploads/2026/08/Data-Center-Framework.pdf.
Verrus is answering every electron objection — batteries instead of diesels, closed-loop cooling, flexible load that holds down power prices — and the backlash shows up anyway, which says the real constraint was never the machine. It's the permission to site the machine: scarcity rent rotating from the grid to the permit office, the slowest layer for capital to resolve. In The Thermodynamics of Thought I trace why inference gets repriced as infrastructure rather than software, and why the 3–7 year interconnect queue is where that physics gap gets paid. https://seldondance.substack.com/p/the-thermodynamics-of-thought #DataCenters #AIEnergy
This is such an important piece of the grid conversation because I think we often frame the problem as “we need more electricity” when the more interesting question is when we need it.
A more flexible grid could change the equation quite a bit. If some of the biggest new loads, especially data centers and other flexible demand can respond to what the grid is actually doing, we may not have to solve every demand spike by simply building more generation.
I’ve been looking at the AI–energy connection from the climate side, and this is one of the pieces I find most fascinating: AI is creating enormous new electricity demand, but it could also push us to rethink how flexible the electricity system itself needs to become.
The grid isn’t just a supply problem. It’s increasingly a timing problem.
The four hour battery plus indefinite compute scale down is a bet on the shape of the emergency, not only its length. RTO demand response strips are two to four hours because that is what the programme pays for, not because that is what a cold snap does, and Uri ran for days. Day two and day three are when the interconnection agreement gets read out loud. The part I would want in writing is who eats the shortfall when a shed event overruns the strip, the tenant taking an SLA credit or the developer carrying the capacity penalty, because that allocation decides whether this is a grid asset or a very expensive option the utility was handed for free. Interruptible industrial tariffs solved a version of this decades ago, and the smelters that took them learned that the curtailment clause is the whole contract.
Sorry Nelson we just don't trust you. Because your legally obligated to f$c7 the public for higher profits for investors.
If you were a B Corp with a 5:1 or 10:1 compensation cap you might get more trust.
50 years of neoliberalism and Reaganomics is a lot of trauma for people to get through.
We need to switch to metric. It's too hard to memorize that 1 Cheesecake Factory is 4 million gallons...
I guess, "better" is better, but when Nelson calls a 500 MW data center an "energy asset" I wondered how many EVs could be powered by that.
500 MW * 1000 kW/MW * 6500 full load hours/yr / 5000 kWh/yr for an inefficient US EV = about 700,000 passenger EVs. Unless I slipped a decimal place.
That was the kind of "energy asset" load growth I was hoping for not too long ago.
Skynet is where the terminator was born!