When the big tobacco companies reached a legal settlement over health care costs with 46 states in 1998, those states were in line for a major windfall of money calculated in the billions.
Those states used the money in lots of different ways — some simply used it to plug budget holes — but only two used some of it to try to create a new economy in former tobacco-growing counties.
One of those two states was North Carolina; the other was Virginia. In the years since, the Tobacco Region Revitalization Commission has paid for miles of broadband fiber and other infrastructure improvements across Southwest and Southside.
While tobacco is a declining industry, oil and gas is not — at least not in North Dakota. When that prairie state found itself home to a gusher of boom towns, the state set up a special fund to receive its oil and gas taxes. That fund today pays for lots of other things: roads, bridges, tax relief and, perhaps most importantly, research into potentially new economic sectors the state could embrace once the oil and gas run out.
In both cases — one with tobacco, the other with oil and gas — states took the opportunity to use their momentary good fortune to look out for future generations.
Now comes this question: Should communities try to do the same with data centers?
That’s the question posed in a proposal released this week by the Brookings Institution, a Washington, D.C.-based think tank. In an 18-page paper entitled “Turning the data center boom into long-term, local prosperity,” authors Daniel Goetzel, Mark Muro and Shriya Methkupally advance a fascinating proposition that Virginia localities currently negotiating with, or hoping to negotiate with, data centers should read before they sign anything on the dotted line.
They make the case that localities hold more bargaining power with data centers than they realize and should use that power to negotiate for deals that help create a new economy in their regions.
The Brookings authors contend that artificial intelligence has changed the balance of power in localities’ favor because “hyperscale” data centers intended to do AI work require more land and more power — and both of those are in short supply. That means localities that want to attract data centers can be more aggressive in making demands, the authors say, and they lay out some of the things that localities ought to start asking for. Specifically, they should be asking for data center contributions that don’t just cover a one-time community need but can help jump-start an entirely new economic cluster.
Here’s how they lay out their argument:
“Traditionally, data centers have been built as standalone industrial facilities,” the Brookings paper says. “Accordingly, both developers and communities have treated data centers like warehouses — routine projects that produce a surge in short-term construction jobs followed by a steady stream of tax revenue without offering many direct contributions to or interactions with the local economy.”
Data centers also had more choices about where they could locate, which limited the power of localities to cut deals, because localities feared losing the tax revenue.
Enter artificial intelligence, which the Brookings authors see as a little-recognized game changer.
“Now the ChatGPT era is disrupting the standard model,” the Brookings paper says. “From the moment the AI titans committed to super-scaling their models to unprecedented sizes, they began to require unprecedented amounts of computing infrastructure to train and operate their datasets. What’s more, the faster the firms have sought to scale, the faster they have needed to strike deals with communities to build gargantuan computing facilities, often in the face of increased local concerns about electricity use, noise issues, and other side effects.”
That has shifted the dynamic some — maybe not a lot, but at least some.
“For years, data center development has pitted community against community in races to attract data centers and their perceived jobs,” the Brookings paper says. “Now, the new frenzy of competition between the hyperscalers themselves (and other actors) for the biggest sites with the best power connections has given communities at least a sliver of their own leverage to shape AI-focused data center deals into more forward-looking, mutually beneficial development projects.”
What do the authors have in mind?
The short version: They say localities should require data center companies to invest in the local economy — not just give a one-time contribution to pay for something.
The longer version: “More communities have the leverage to trade expedited permitting and approvals for substantive gains that truly advance the region’s goals, whether that involves donations of compute to local universities, R&D partnerships, support of local talent programs, or investments in AI startup intermediaries,” the paper says.
Vocabulary note: The phrase “compute” is not a typo that was meant to say “computers.” The word “compute” is the industry lingo for “computing power.” It rings odd to my ear, and not just because I spent too many years blasting Led Zeppelin at full volume, but our language evolves and that’s the word today. Copy editors, take note.
The Brookings authors offer some specifics.
Here’s one:
“Universities and states should seek to participate in AI players’ super-sized fundraising rounds, perhaps by asking for early access to cutting-edge models for their researchers in exchange for their investment. Using this approach, AI companies (at the scale of Anthropic and OpenAI), or smaller, high-growth companies, receive an inflow of much-needed capital and the ability to train their models on university datasets. Universities, meanwhile, need to receive near-term access to compute, which is critically important to their faculty recruitment and retention and broader R&D efforts. In the long run, universities can hold equity stakes in increasingly valuable companies, hoping to replicate Stanford’s windfall from its early equity stake in Google.”
Here’s a second:
“Communities or community-designated entities should seek to co-invest in AI-related real estate deals alongside developers of data centers. Both parties (impact investors and data center developers) would commit to transferring a portion of their illiquid equity shares into a separate fund called a community equity endowment (CEE). If (and only if) a given project is successful, then a portion of any windfall financial returns would be distributed to community members via the CEE, with direct community oversight on the uses of funds. The community shareholders could collectively decide if they wanted to use their CEE allocation to introduce universal basic income programs, develop a fund for one-time emergency expenses, provide micro-grants to small businesses, or even develop programs like local baby bonds.”
Here’s a third:
“Ask AI negotiating partners to commit to organizing AI-related pilot programs and testbeds as a condition of incentives for data center buildouts. By incorporating phased asks into these negotiations (such as calls for tech pilots, shared compute, and other benefits), states and regions can transform their interactions with the AI sector from a series of disconnected, isolated data center projects into AI hubs.”
Any successful negotiation requires some give-and-take; here’s what Brookings suggests states could offer: faster approval for various permits in exchange for “reduced energy usage during periods of peak usage; jointly funded pilots of AI-driven technologies that shift loads and energy mix during peaks and lulls in data center activities, while giving opportunities to locally sited data centers and proposed testbeds to incubate and stress-test these models; economic development commitments that expand beyond traditional short-term construction jobs; market commitments and conditional purchase agreements with geothermal and nuclear companies that are building projects within a given state, as a way to allow states to invest in new technologies and finance pilot projects without sticking ratepayers with the future bill; and commitments and investments that replicate at the state scale the arrangements emerging AI companies are making with federal agencies and national labs, such as Anthropic’s commitment to the Department of Energy’s Project Genesis or OpenAI’s and Nvidia’s agreements for sharing supercomputer and model resources with national laboratories.”
How practical are these suggestions? Some of this is already happening around the country. Botetourt County, which is getting a Google data center project, is exploring ways Google can make the Roanoke Valley a testbed for certain AI tools. The Brookings authors note that “Quietly tucked into the announcement with President Donald Trump hosted by Carnegie Mellon University (CMU) were $2 million in commitments from Anthropic to provide cybersecurity education to middle and high school students in Pennsylvania and funding for energy research at CMU, which is increasingly becoming a nexus of AI, energy, and industrial innovation.”
What the Brookings authors propose is that such deals become routine — that data center requests just aren’t about setback requirements or water permits, but more substantial investments to grow the local economy. The first step might be for state and local officials to read their report. They could even ask AI to summarize it for them.
After reading that report, you can also read West of the Capital, our weekly political newsletter that comes out on Friday afternoons. Sign up below.