If you care about energy, or technology, the growth in electricity demand from data center operations is inescapable. If you care about both, it is elemental. AI-driven data center demand is a, if not the, main driver of electricity demand growth in many US electricity markets, and is one of the prime movers of global electricity demand growth this decade. The combined capital expenditures to develop AI from the four ‘hyperscalers’ (Alphabet, Amazon, Meta and Microsoft) will exceed $200 billion in 2024.
Demand is nuanced, of course. Electricity demand is only 3-6% of the total expense of training a new AI model. But, access to bulk, reliable, and ideally clean power at massive scale is essential. For AI, energy is nothing, and energy is everything. It is both a minor expense, and an inextricable necessity.
This quantum state leads hyperscalers, utilities, grid operators, and power producers to make strategic commitments that are both grand and somewhat vague. This week, both Amazon and Meta have announced strategies for nuclear power, in order to meet the significant growth in their electricity demand from data centers. The International Energy Agency is hosting its first-ever Global Conference on Energy and AI.
All good, and important for thinking about the future. But while strategic vagaries can be constructive, they do not necessarily lead to construction starts. For that, we need clear tactics on the ground in existing power markets, and real data from trusted market participants.
And in the state of Virginia, we have a bit of both right now. Virginia already operates the world’s largest cluster of large-scale data centers, and data centers are already more than one quarter of the state’s total electricity consumption. But ‘Data Center Alley,’ as its highest-concentration area is known, is already filling in, and communities, utilities, and the grid operator are already concerned about further growth. These conditions push data center development further afield. Specifically, further west, into the service territory of the Rappahannock Electric Cooperative (REC).
I know this because a recent Halcyon email notification on the Virginia State Corporation Commission highlighted an ongoing docket, Case No. PUR-2024-00025, from REC and a newly formed set of entities, Hyperscale Energy Services, LLC and Hyperscale Energy 1, LLC. The name alone suggested to me that this proceeding, and these new entities, would be data center-focused; I was not wrong.
REC is asking its state regulator for approval to create these new entities to ensure that it can meet electricity demand from data centers, but do so in a way that does not abuse the public trust within the utility’s duty to provide service to all customers. Essentially, one large data center’s demand could completely alter the operations of even a large electricity service territory, both operationally and commercially. And, it could do so whether the data center is operating (and putting massive demands on grid power) or if it ceases operations (and leaves a revenue hole that a utility must recoup from its remaining rate payers).
That’s still somewhat abstract, so I used Halcyon to dig further. First, I was curious just how big this power demand from data centers might be: a quick query located this text from another filing in the ongoing docket (p. 20 of 55): “other new data center customers, some of which may be up to four times the size of REC’s existing peak system load”
That’s clearly an enormous-but-still-abstract figure, because REC does not publish its system load anywhere. Fortunately another query can help: buried in another document (pp. 7-9 of 20) is REC’s statement that its 2022 peak load is 1,105 megawatts, and that it could add more than 8,000 megawatts of load if all contemplated developments are fully realized in the next two decades.
Those are concrete numbers! But, still not enough to be fully actionable. In the same proceeding, REC says that hyperscale data centers can be 200-500 megawatts in size, but that there is also interest in building data centers that are multiples of its current peak load — so, more than two gigawatts of demand from one entity. Those figures are directional, but not a proper forecast.
So, another query: What is REC’s forecast for data center demands, not just the big numbers it waves around? That query returns a chart from REC’s filed demand projections with PJM, its regional transmission operator. REC’s data center demand today: all of zero megawatts. Its forecast demand in 2040? More than three gigawatts.
Now we have a clear picture of what is underway in a new frontier for data center development. An electric cooperative that currently meets a 1.2 gigawatt electric load, including zero megawatts of data centers, is planning to meet almost three times that much demand from data centers alone in 15 years. In order to do that, it is creating a new commercial structure to serve these customers in a way that both meets their needs, and protects other customers. It’s all in the filing, so to speak — just not all in one place, or in one document, or at one point in time.
Announcements and pronouncements are one thing, but technically informed, clearly labeled forecasts are another. The truest sense of Rappahannock Electric Cooperative’s future data center demand is not on its website, or in the plans of its future customers. It is buried deep in a filing, public but not hidden, but also not easily discoverable.
Theoretical discussions of data center power demand help drive awareness, and awareness drives policy. Energy supply initiatives from the world’s biggest data center operators expand the sense of what can be built, by whom, and by when. But, much of the real work is here: in filings and testimonies, in hard-to-find forecasts, in dockets and proceedings, and in legalistic-but-essential contracts and entity proposals. I think of it as a cross-referencing of theory and practice, with the practical result of allowing builders and investors to better understand what could work fastest, and in highest volume, and where.
Meeting AI power demand requires innovation, and discerning that demand’s exact parameters needs new approaches too. If you would like to know more about how Halcyon’s alerts link a flow of state and federal energy proceedings to an AI-powered query system, we’d love to hear from you.
Comments or questions? We’d love to hear from you - sayhi@halcyon.eco, or find us on LinkedIn and Twitter