Nitpicking a Utility Dive Article on AI Power Needs
A little bit of a back and forth from Secretary of the Interior Burgum during his confirmation hearing and Senator Cortez Masto (D-NV)
Sen. Cortez: “Let me ask you this, and maybe I’m wrong about this, but isn’t the combination of renewables plus battery storage baseload?”
Sec. Burgum: “Yup, that is certainly the future.”
He then goes on to say that without storage, wind and solar are unreliable, and that batteries are a few years out. This hearing was last year. Today, you are not going to see a single AI data center built that is not using some form of renewable plus storage, yet the same unreliable and intermittent framing is still used.
He also mentioned that the U.S. needs to adopt an all-of-the-above energy policy, a concept he has largely disregarded for anything not hydrocarbon related.
Anyways, here is an example from today illustrating this almost meme-like narrative around solar.
In a Utility Dive article about AI power needs, a friend sent it to me who, like myself, is invested in a company covered in the article, Fermi America. Normally, I do not follow Utility Dive.
“Solar and wind are essential but intermittent.”
You are not going to see an AI facility using solar without storage, so this framing is asinine.
“Natural gas may be having a moment, but the winning model will be hybrid: gas for firmness, renewables for optics and cost, and storage for stability.”
A recent study looking at how AI sites can curtail demand laid out the optimal mix for several sites located in PJM’s RTO footprint and here is an example of one:
“At Koala, the less constrained site, the optimal mix included 26MW of solar PV, 49MW of four-hour battery storage totaling 196MWh, and 11MW of on-site gas generation.”
Depending on the constraint, whether local generation or transmission, the main tool used to create flexibility of demand will be storage.
But what else did Utility Dive get wrong? At the core of the public narrative around AI energy is the idea that there is not enough power being generated to supply these sites. This is true, but the timeframe where this is true is very small. These large sites are designing generation and storage infrastructure to curtail demand for a few hours a year.
This is because most of the time the grid operates below peak capacity, but service providers like PJM plan generation to meet infrequent peak demand levels.
Without AI sites providing their own capacity through onsite generation and storage for these peak periods, local utilities would need to factor in that additional load in their peak planning, potentially requiring more conventional generation to be online and increasing system stress. This is where the narrative that AI sites will raise consumer power bills comes from.
The AI sites that plan to be integrated with the grid, unlike the ones that plan to be 100% self-contained, will during peak demand periods rely on onsite generation and, more importantly, storage to reduce their load on the system. Once peak demand subsides, they resume normal grid usage while recharging their storage, smoothing their impact on the grid without adding stress to overall capacity.
Due to the long lead times on gas turbines, the fastest and realistically best way to provide what is called baseload during these periods of curtailment is with solar plus storage.
Now back to Fermi America. While it has plans to provide 11 GW, the bulk of it from gas-fired generation with a transition to nuclear over time, it is still not going to be 100% self-contained.
The site is essentially doing a role reversal: instead of the grid supplying power to a massive AI campus, the campus is bringing its own generation and storage to handle the majority of its load. This includes up to 1.75 GW of solar capacity and roughly 500 MWh of four-hour battery storage. It is the flexibility of this solar and storage combination that will allow Fermi to launch the site with the proposed 1.1 GW capacity while only using the 200 MW of grid-supplied power provided by Southwestern Public Service when needed.
If you found this article valuable, consider becoming a subscriber. The Critical Materials Bulletin is supported by readers, and for $5 a month or $55 a year you can help fund research that produces clear, no nonsense reporting that informs and advances the discussion on critical materials and battery metals.
DISCLAIMER: This article should not be construed as an offering of investment advice, nor should any statements (by the author or by other persons and/or entities that the author has included) in this article be taken as investment advice or recommendations of any investment strategy. The information in this article is for educational purposes only. The author did not receive compensation from any of the companies mentioned to be included in the article.