Avoiding AI-induced "energy gentrification"
As in housing, the path to stability is through abundance.
The list of reasons to support radical energy abundance is very long. Let’s add another to the list: avoiding what I’ll call “energy gentrification.”
The popular conception of gentrification in the context of housing sounds something like this: developers put up new “luxury” units in low-income neighborhoods, causing prices to rise and involuntarily driving locals out of their long-time neighborhoods. This is why while so much of NIMBYism is resistance to development by parochial self-interest, much of it is also a genuine concern for low-income people.1
Yet we know that this popular conception is wrong. Housing scarcity, rather than growth, displaces low-income families. Picture an influx of wealthy outsiders buying and fixing up dilapidated brownstones; incumbent homeowners sell and earn a windfall, but without an outlet like new construction to absorb this new demand, existing renters or prospective buyers with fewer resources are trapped in a bidding war they stand no chance of winning. There are no completely independent affordable and high-end housing markets; all housing markets are connected. That’s why even building so-called “luxury” housing quickly leads to more abundant housing options for low-income renters through the filtering effect.
Kate Pennington, an economist at the Census Bureau, has a clever working paper looking at the effect of new construction in San Francisco, taking advantage of random new construction induced by building fires. She finds that rather than pushing existing residents out into poorer neighborhoods, new construction actually reduces displacement and lowers nearby rents. Change, not stasis, ends up being the source of stability and security.
Why the long housing windup for a blog nominally about energy? Rapid advances in artificial intelligence could quickly and substantially increase energy demand in the United States, both through the explosive growth in data centers directly used to train and run AI models, and through accelerated economic growth. The Boston Consulting Group estimates that data centers’ share of U.S. electricity demand will triple by 2030:
But these are just the mere short-term, first-order effects. What if AI, as many believe, turns out to be a game-changer for economic growth in the medium to long term? I’m not referring to wild projections of explosive growth on the order of 20 or 30% per year (though you should check out this Asterisk Mag debate on this possibility). But say AI sparks a productivity boom that brings growth up to 5% for a sustained period.
In such a scenario, the race between energy efficiency and economic growth that has kept U.S. electricity use relatively flat since the mid-2000s would likely end, with growth emerging victorious. Add this to the challenge of electrifying transportation and other fossil fuel-guzzling sectors and we could see huge growth in absolute electricity demand.
We should expect this potential, massive growth in demand to vary widely between regions. The growth in huge, energy-hungry data centers will accelerate in exurbs or suburbs on the outskirts of larger metro regions featuring affordable, buildable land. Needing only a handful of security staff and maybe a technician or two on-site, these facilities won’t be especially tied to clusters of skilled labor. The landing spots for many of these new data centers intersect significantly with emerging remote work migration patterns, potentially putting these two trends on a collision course.
Here’s the problem: as with housing, we have made it very difficult to quickly build new energy infrastructure in the United States. And that’s in a world of steady-but-slow 2% GDP growth year after year, where meeting growing energy needs rarely requires big expansions of energy infrastructure. In a future of substantially faster growth, the costs of highly inelastic electricity supplies may start to compound quickly.
This raises not just the possibility of escalating data center NIMBYism, but also the prospect of a wave of energy cost-induced displacement. In a housing market in which government has barred expansions in supply, low-income and working-class households are pitted against wealthier newcomers with much deeper pockets. As we’ve covered, that’s a bidding war the former will lose. A similar scenario could play out if energy demand from AI proves truly explosive. Rather than facing off against high-income newcomers, families may find themselves bidding for a fixed energy supply against even richer corporations—either AI companies themselves or businesses whose productivity is turbo-charged by integrating AI into their processes. Tamping down the energy demand of the marginal crypto mining operation is one thing, but OpenAI? Not so fast.
As in housing, the answer is abundance. Just as building more housing is the key to ensuring families can stay in the neighborhoods they want, building an energy grid that can more easily scale will be necessary to do the same in a world of more rapid, AI-induced growth. Relentless R&D to further bend cost curves, radically expanding interregional transmission capacity, and yes, permitting reform, are all in order.
If AI is “eating the world,” then it is, indeed, time to build.
Yes, I know I’m conflating gentrification and displacement here. That’s one of my pet peeves, too.
Should AI dramatically increase our energy needs in the coming years, I see no real alternative to nuclear energy. Solar, wind, and natural gas will help, but nuclear reactors will probably be a necessity.
This means that the public and the government are going to need to let go of the suffocating regulations that have imperiled the industry since the 1970s.