The AI concentration problem in the U.S.

There’s a lot at stake in the artificial intelligence race, and although it may feel like it’s everywhere, the U.S. AI race is primarily playing out in just a few places — specifically, hubs that offer AI entrepreneurs advantages like capital, talent and more.

That helps explain why so many AI companies, patents, job opportunities and so much else are concentrated in Silicon Valley and other very expensive, mostly coastal U.S. cities, said Mark Muro, a senior fellow at Brookings Metro.

Marketplace’s Lily Jamali spoke with Muro about that concentration and what it means for AI development going forward.

The following is an edited transcript of their conversation

Mark Muro: (When it comes to) AI in general, two-thirds of all assets and capabilities are in 15 what we call “early adopter metros.” San Francisco and San Jose, you know, (have) a quarter or more of all activity, job postings, patents, etc. And then places like Seattle, Los Angeles, Boston, account for huge portions of this. For generative AI, you know, ChatGPT and other applications, so-called large language models, we’re talking about 60% in just 15 places.

Lily Jamali: And how worried are you that we’re going to keep seeing this kind of concentration as more and more AI startups and development projects pop up?

Muro: I’m quite concerned about the concentration dynamic. We call it a winner-take-most geography. And I think until proven otherwise, that is the standard way that technology economies evolve. And some of it is highly valuable, it’s an efficient way for places to innovate. Key assets, research and development, finance, deep pools of workers, lead to what’s called clustering. But if we cluster the economy in just a few places, most places don’t have deep connection to it, and especially with AI, which is expected to be an important benefit to regional economies. So it behooves us to find ways for more places to participate, and for more ideas and more talented people to be contributing.

Jamali: So if that kind of clustering continues unabated, what are your thoughts on how that affects the direction of AI development in the U.S.?

Muro: I think for one thing, it leads to a kind of monoculture where the same relatively narrow group of Big Tech companies and leading experts develop the technology in the way they see fit, which may not solve problems that are, you know, evident across the country and other markets. Another concern is more of America becoming suspicious of a technology — or even innovation in general — that is occurring somewhere else. So I think there are some very serious concerns, and that we should be thinking about how we might more diversify technology in general and AI specifically.

Jamali: So what is an up-and-coming AI company to do? I mean, you’ve touched on this a little bit, but we can’t ignore that places like Silicon Valley and Seattle have their advantages. They’ve got the resources, the money, the people, and that matters if you’re trying to compete in the AI race.

Muro: If a great idea sprouts in Indianapolis, I think there’s the hope that it can tie into local strengths. Regions and states, I think, need to help with making it feasible to not only have an idea come out, but then to grow it and scale it. And I think there is a role for, you know, federal and state local policy and even philanthropy to help make sure that there are fertile grounds for scaling AI companies all across the country.

Jamali: Yeah, talk to me more about that, because in this report, you are advocating for spreading the wealth to other places that can support AI development. How does that happen? Is that a job for government incentives? Is that something we see from the private sector?

Muro: Yeah, our thought is that these trends, to themselves, will just get us more of what we’re already getting from technology, which is intense concentration. So it is going to take some active steering and nudging and incentivizing. One thing I think that likely the federal government needs to do is to work hard on making sure that new emerging companies in all places have access to two things that are critical: big data sets for training and other testing, and then computational resources. Another thing is making sure that more places have top quality, really compelling research on interesting use cases. Training [is] very important, and then states can help with subsidies and support for clustering. Some places will begin to emerge with really meaningful groups of AI firms, and that’s also a very promising way to proceed. I think there’s a reason though that suspicion around technology in general, and maybe AI, has arrived at a kind of high level. And I think that may have to do with the fact that AI is concentrated into just a few places, and most places are not participants. And I think a degree of diffusion of AI into new places and more places could be really beneficial for the industry.

Jamali: In terms of selling itself, getting buy in from not just the coastal U.S., from the rest of the country?

Muro: Absolutely. And not just as a political matter, making sure that we’re surfacing talented AI researchers who have figured out great ideas that would help with relevant issues across the country, whether it’s in agriculture or whether it’s in health services, or what have you. And I think we need to access more talent and more ideas across the country.

More on this

As Mark Muro mentioned, state and federal government intervention would play a key role in any attempt to spread AI innovation.

U.S. lawmakers in the Congressional Artificial Intelligence Caucus recently revealed the CREATE AI Act, which would establish a national artificial intelligence research center.

No details yet on just how much money an effort like that would need to get off the ground, but wouldn’t be surprised to see a billion figure in there.