Data Centers are already changing electricity consumption
Big upward revisions to the data center states
10 states drove the national increase in electricity demand
actuals beating estimates, and upward revisions
natural gas is back, baby
👉👉👉Reminder to sign up for the Weekly Recap only, if daily emails is too much. Find me on twitter, for more fun.
Data Centers are already changing electricity consumption
Data centers are already driving an increase in electricity consumption:
States with active hyperscaler data centers have seen the biggest increases in electricity consumption.
Consumption of electricity in the U.S. commercial sector has recovered from pandemic levels, with annual U.S. sales of electricity to commercial customers in 2023 totaling 14 billion kilowatthours (BkWh), or 1%, more than in 2019. However, the growth in commercial demand for electricity is concentrated in a handful of states experiencing rapid development of large-scale computing facilities such as data centers . . . Commercial electricity demand in the 10 states with the most electricity demand growth increased by a combined 42 BkWh between 2019 and 2023, representing growth of 10% in those states over that four-year period. By contrast, demand in the forty other states decreased by 28 BkWh over the same period, a 3% decline.
A 10% increase in the top 10 states “experiencing rapid development of large-scale computing facilities,” with a 3% decrease in the bottom 40.
Likewise, actuals are exceeding estimates, and forecasts are upward revised:
We made our largest revisions to the forecast in the South Atlantic and West South Central census divisions, which together account for 40% of U.S. commercial electricity demand. We now expect that commercial consumption in the South Atlantic will increase by 5% in 2024 and 2% in 2025 and in West South Central by 3% this year and 1% next year.
Actually, 2-3% increases in consumption are pretty modest, but that’ll compound quickly, if we keep this up.
Going au naturale gas
What’s powering all this energy consumption?
Well, it’s natural gas for the most part:
Natural gas consumption hit an all-time high, driven by electricity use.
No surprise there, as natural gas is reliable and plentiful, and ramps upwards more easily than say, nuclear. Let’s just hope the winter doesn’t get too cold.
Meanwhile, solar and wind, will have to find other ways to contribute.
In terms of why this matters, well:
that AI-related computation consumes a lot of energy is old news; that
the current grid—from power generation to storage to transmission to routing—might be unprepared for the increased demand, is also old news. Electricity demand has been basically flat for decades, so sudden and urgent demands for MOAR POWER, is a bit of a shock to a previously sleepy sector; and that
the lack of sufficient electricity might be a bottleneck to the continued growth of AI-related computation, is also old news.
So, none of those are the main reasons while this data matters.
The open question is/was whether AI (and the related electrification of everything) would actually overwhelm existing capacity.
The reason energy consumption might not be that big a deal (outside of AI just fizzling) is either because the hype-machine was (a) overstating actual long-term demand growth by extrapolating from a recent trend; and/or (b) understating inevitable efficiency gains (at every step of the value chain) because necessity is the mother of invention, after all.
It is obviously way too early to tell, but these early returns fall into “the hype-machine” isn’t over- or under-stating anything, just yet.1
Other links
Rural Michigan grows. Flight from metro isn’t over yet. See also The US moves South. The South is the only census region to gain net domestic migrants, and the only non-Southern sub-region to gain is the Mountain West (including Arizona, Idaho, Nevada, Montana, Utah, Colorado, Wyoming and New Mexico).
Remote work is tanking home prices in the heart of London. Is it remote work, or is it a 5 year mortgage that’s ~2x more expensive than it was three years ago?
2035: an allocator looks back over the previous 10 years. Cliff Asness peers into the future. Pulls no punches.
Good thread on recent unemployment claims release. Not much change.
Reinsurance prices fall, as capital flies to sector. I don’t hate to say I told you so. It never had anything to do with ‘billion dollar hurricanes’ or some such nonsense.
Previously, on Random Walk
Random Walk is an idea company dedicated to the discovery of idea alpha. Find differentiated data, perspectives and people, and keep your information mix lively. A foolish consistency is the hobgoblin of small minds. Fight the Great Idea Stagnation. Join Random Walk. Follow me on twitter. Follow me on substack:
As an aside, I found this story about how data center growth is overwhelming Atlanta to be mildly amusing, if only because (as predicted) “build more housing yimbyism” is giving way to “actually land use regulations are good again because when we said ‘yimbyism’ we really meant ‘change regulations to build just the stuff we want.’”
Nice. Asness is in on FartCoin too (or will be). That gives me even more confidence in it.
let’s connect!