Sam Altman has been a big backer of Helion partly because he hopes their smallish fusion machines will be able to be installed behind the meters at data center/AI clusters. Of course, one's probability estimate of whether and when they will succeed may be pessimistic. But they have a contract with Microsoft already, pending the success of their technology (currently aiming at a commercial prototype by 2025 or 2026).
Thanks Lynne. Like many.. am also studying the dynamic.
Particularly interested in how datacenter demand drives prices. How traditional customer segments compete to buy power in open markets against insatiable buyers with those margins. How utilities price discriminate and shifting profit/cost centers toward different customer segments. How this reframes the role and benefit of DER. Which jurisdictions take power public.
Data center demand is centered in a select handful of States. It should be feasible to set up a simple study that tracks IRPs, rate cases, and policy changes.
This was foreseeable for quite a while and have pushed the argument since 2008 to little avail. Information Age+Big Data+Compute+Energy Physics is directionally a straight forward equation. With EVs clearly coming since 2005 plus Compute, it make no sense that demand forecasts were often lower than GDP growth. The Industry has been insular for too long.
I'm going to try to address your pricing questions in my subsequent posts, particularly with a focus on both flexibility and DER. Although my follow up question would be whether the capex and regulatory limits on utility expansion will lead to the erosion of vertical integration as more data centers build their own power.
On your study point: do you know Cameron Brooks at E9 Insight? His company catalogs the issues in all 50 states, so they would be a good group of folks to engage to perform such a study.
"California Air Resources Board’s (CARB) order, that would ban locomotives that are 23 years old or older from running in the state, requiring them all to be replaced by zero-emission locomotives by 2029."
Governments will require that all personal and commercial vehicles run on batteries. Am I supposed to believe that if I drive my EV SUV 50 miles a day it will use way less electricity than my GPU using computer? Or that an EV locomotive, using no diesel, pulling a mile-long freight train over the Rockies, will use way less electricity than a data center?
The media recently been claiming an energy crisis will be caused by AI. (But not of course by EV.)
Will the increase in electricity use from AI be a drop in the bucket compared to the increase caused by EV? Where am I wrong?
I also think it matters that EVs behave differently from stationary batteries and from data centers, and that different types of data centers behave differently. Different scales, timing, degrees of flexibility. If we had/can have better price formation and price discovery in good, informative markets, then we can use price signals to coordinate all of these diverse uses of the electric system.
The Grid Strategies report linked in my post has some discussion of the relative magnitude of other forms of electrification, so I'd look there for more information, and they may also cite some more sources.
In your definition of a data center are you only considering Amazon Web Service centers (and similar) where the info model is storage lots of data (purchase transactions) and retrieving what is needed when a new transaction occurs? Do you consider this different from generative AI queries? A parallel but different topic is crypto mining/ledger maintenance. I am guessing this is not covered in the subject discussion, but I would like your thoughts on this as well.
You are right, my observations are at a more general level and if you look at the first graphic, the bar chart of forecast data center investment by data center type, it does delineate among these different types (although it doesn't include bitcoin mining). I didn't want to get too deep in the weeds on the differences between what is now standard cloud computing and AI computing. I may do so in a subsequent post because the temporal patterns of their energy use differ, and that has really important system-level implications in electricity.
Excellent framing post. I am looking forward to the follow ups.
Thank you! I hope you find it useful :-).
Sam Altman has been a big backer of Helion partly because he hopes their smallish fusion machines will be able to be installed behind the meters at data center/AI clusters. Of course, one's probability estimate of whether and when they will succeed may be pessimistic. But they have a contract with Microsoft already, pending the success of their technology (currently aiming at a commercial prototype by 2025 or 2026).
Thanks Lynne. Like many.. am also studying the dynamic.
Particularly interested in how datacenter demand drives prices. How traditional customer segments compete to buy power in open markets against insatiable buyers with those margins. How utilities price discriminate and shifting profit/cost centers toward different customer segments. How this reframes the role and benefit of DER. Which jurisdictions take power public.
Data center demand is centered in a select handful of States. It should be feasible to set up a simple study that tracks IRPs, rate cases, and policy changes.
This was foreseeable for quite a while and have pushed the argument since 2008 to little avail. Information Age+Big Data+Compute+Energy Physics is directionally a straight forward equation. With EVs clearly coming since 2005 plus Compute, it make no sense that demand forecasts were often lower than GDP growth. The Industry has been insular for too long.
I'm going to try to address your pricing questions in my subsequent posts, particularly with a focus on both flexibility and DER. Although my follow up question would be whether the capex and regulatory limits on utility expansion will lead to the erosion of vertical integration as more data centers build their own power.
On your study point: do you know Cameron Brooks at E9 Insight? His company catalogs the issues in all 50 states, so they would be a good group of folks to engage to perform such a study.
Thanks, very helpful!
"California Air Resources Board’s (CARB) order, that would ban locomotives that are 23 years old or older from running in the state, requiring them all to be replaced by zero-emission locomotives by 2029."
Governments will require that all personal and commercial vehicles run on batteries. Am I supposed to believe that if I drive my EV SUV 50 miles a day it will use way less electricity than my GPU using computer? Or that an EV locomotive, using no diesel, pulling a mile-long freight train over the Rockies, will use way less electricity than a data center?
The media recently been claiming an energy crisis will be caused by AI. (But not of course by EV.)
Will the increase in electricity use from AI be a drop in the bucket compared to the increase caused by EV? Where am I wrong?
I also think it matters that EVs behave differently from stationary batteries and from data centers, and that different types of data centers behave differently. Different scales, timing, degrees of flexibility. If we had/can have better price formation and price discovery in good, informative markets, then we can use price signals to coordinate all of these diverse uses of the electric system.
"If we had/can have better price formation and price discovery..."
I guess that's why you call it the knowledge problem. Thank you for your answer.
The Grid Strategies report linked in my post has some discussion of the relative magnitude of other forms of electrification, so I'd look there for more information, and they may also cite some more sources.
In your definition of a data center are you only considering Amazon Web Service centers (and similar) where the info model is storage lots of data (purchase transactions) and retrieving what is needed when a new transaction occurs? Do you consider this different from generative AI queries? A parallel but different topic is crypto mining/ledger maintenance. I am guessing this is not covered in the subject discussion, but I would like your thoughts on this as well.
You are right, my observations are at a more general level and if you look at the first graphic, the bar chart of forecast data center investment by data center type, it does delineate among these different types (although it doesn't include bitcoin mining). I didn't want to get too deep in the weeds on the differences between what is now standard cloud computing and AI computing. I may do so in a subsequent post because the temporal patterns of their energy use differ, and that has really important system-level implications in electricity.