Which States Will Win the AI Race?
The AI race is really an overlapping set of races. Most notably, AI has become a theater of great power competition, namely between the U.S. and China. Within the American economy, the frontier labs are vying for dominance.
But there are levels of competition between nation-states and individual companies. In pursuit of economic development opportunities, U.S. states have traditionally competed for major private-sector projects. Homes for tech company operations, such as corporate campuses, have been especially sought after because these firms address massive markets and produce high-value-added jobs.
While the frontier labs remain fairly localized in the usual tech hubs, there are significant opportunities outside of these hubs on the physical infrastructure side. But not all states will benefit equally.
Virginia and the First Mover Advantage
The internet ranks with the Interstate Highway System, the Hoover Dam, and the Apollo Program as one of the great federal engineering projects of all time. In retrospect, it’s unsurprising that Northern Virginia, so close to the seat of government, has come to dominate computing infrastructure.
As early as the 1960’s, the federal government was laying fiber optic cables in Virginia. The state would become home to one of the original four internet exchange points. In the 90’s, America Online located in Loudon County and decades of data center construction began.
In 2008, the state attempted to lean into this edge to address unemployment in economically distressed counties, passing a sales tax break tailored to the equipment needs of data centers. The incentive worked, attracting Microsoft to Mecklenburg.
In 2010, after losing an Apple project to neighboring North Carolina, the state expanded the tax incentive to data center operations in all counties.
Today, data centers are a pillar of the state’s economy, accounting for half of Loudon County’s tax revenue. Geolocate any given IP address and the odds are good it originated in Ashburn.
Northern Virginia’s status as a tech hub extends beyond the computing infrastructure that put it on the map, as Amazon, Google, and Meta now have offices there, with Amazon maintaining a particularly large footprint.
Connectivity, energy capacity, and talent networks tend to cluster geographically. Virginia’s success is the product of decades of accumulated advantages that Virginia was able to compound during and even before the advent of the internet.
Similar policy choices are being made today that will shape the geographical future of the AI-enabled economy.
Problematic State Policies
Though computing infrastructure is inevitably situated within specific zip codes, the consequences of action and inaction are national in scope. The world — and the United States to the extent it wishes to remain competitive — will come to rely on AI more and more heavily for productivity growth, scientific discovery, and national security.
Even assuming states wish to act purely in their own — not the national — interest, many policy areas are just better handled at the federal level. To the extent states depart from national frameworks around AI — which is inherently national in scope — they will usually do so to the detriment of their own economies, becoming less interoperable with the rest of the nation and the world.
There are some extreme cases. As of April, 11 states are considering bans or conditional restrictions on new data centers. Other states are weighing more defensible proposals to ensure data centers contribute to new energy capacity and don’t deplete local resources, but even in these cases, legislators should exercise caution in erecting new veto points in an environment that already makes it difficult to build.
It’s harder to see the case for other state-level regulatory moves. States like Colorado, New York, California, and even (disappointingly) Texas have moved to regulate model outputs and to require developers to meet burdensome testing and reporting standards.
The specifics of these proposals are less important than the mismatch between the nature of the industry and the growing demands of state-level compliance.
AI systems are basically software products. The technical details, the capital inputs, and the capabilities have changed, but distribution has not. We don’t train one model for Colorado and another for California, and as discussed earlier, both models would probably run on equipment in Virginia anyway.
The Trump Administration has been clear-eyed about this and in March released a National Policy Framework for AI, designed explicitly to ward off the patchwork regime state-level regulation threatens to introduce. It covers many of the usual concerns about AI that have been in the ether, such as ratepayer protection and safeguards for children who interact with chatbots.
Federalism is a traditional strength of American governance and has often served to prevent rigid, one-size-fits-all policymaking. But the key insight behind our federal system isn’t that the most local process is best in every case. The American Constitution grants certain powers to the national government, such as conducting diplomacy and securing basic rights of citizenship. AI is another area where the federal government is the natural place for policy-making to happen.
But states that want to capitalize on the AI revolution do have a role to play on the infrastructure side.
AI’s Local Veto Problem
On April 7, voters in Festus, Missouri, ousted half of its city council over its support for a $6 billion data center project. The project, backed by both labor unions and Governor Mike Kehoe, was expected to create a large number of construction jobs in the area, as well as a smaller number of permanent roles.
Additionally, the city council negotiated several commitments to the community — including $5 million for a new fire station, upgrades to the city’s water system, and annual multi-million dollar community development payments — all funded by the developer.
Only a handful of homes were identified as being within a thousand feet of the planned development, and the homeowners were offered a voluntary buyout at above market value.
Residents began flooding city council meetings to protest the project. Overwhelmingly, opposition centered on process-related objections — that residents didn’t feel they had been consulted.
“It’s really the way the deal was handled that led to this kind of uprising,” said Rick Belleville, a Festus resident who ultimately sought and won a city council seat.
In the election that followed, four of the council members who voted to approve the project lost their seats, mostly to candidates hostile to the project.
None of this is to say some opponents didn’t have reasonable concerns, but endless process around every development is not a workable way to balance competing interests, some of which aren’t represented at all at the local level.
Compared to counties and municipalities, state officials are able to consider the broader impacts of data centers, namely jobs and revenue. They’re also better positioned than local governments to consider the long-term prospects of their states and how they might thrive in the coming AI-driven economy.
In Texas, counties have limited zoning authority. Additionally, local regulations that conflict with state law are generally void, resulting in a more predictable environment for development. Activists may organize, and local officials may object, but their ability to halt projects is constrained by design.
A recent conflict in Brazoria County illustrates how Texas’s state level development framework removes friction. Activists attempted to block a data center and associated energy infrastructure. The county denied the developer certain tax incentives, but it lacked the authority to stop the project itself.
As a county judge put it, “We cannot stop this project. We do not have the tools.”
As a result of its pro-growth development climate, Texas has become a magnet for developers such as The Stargate Project, a $500 billion joint venture by OpenAI and Oracle, which chose Abilene as the site of its first data center campus. Real estate firm Jones Lang LaSalle projected in February that Texas may overtake Virginia as the data center capital of the world.
Mississippi has taken an even more energetic approach: rather than merely limiting local veto power, the state has actively coordinated investment. Through the Major Economic Impact Act framework, the state structures incentive packages and aligns stakeholders — including local governments — early in the process.
Elon Musk, whose xAI is building a $20 billion data center and natural gas facility in Desoto County, praised the state’s “insane execution speed.”
Like Musk, Amazon has found Mississippi welcoming of data center projects and has committed $25 billion dollars to data center construction in the state, pledging to create 2,000 jobs and invest $300 million in grid improvements.
These success cases demonstrate that states can cut through stagnation not just by providing flexibility lacking in the federal government but also by operating with a bias toward action rarely found in local governments.
Which States Will Win the Future?
The future is unwritten and not always meritocratic. Virginia, which is weighing repealing its tax incentives for data centers, will continue to benefit from the computing infrastructure those tax incentives attracted to the state. California, its tendency to dream up overreaching legislation notwithstanding, will probably remain the tech capital of the world for some time to come.
But we can say who’s playing their cards well.
States like Texas with its local regulatory preemption and Mississippi with its MEIA legislation will carve out prosperous places for themselves and their workers in a world increasingly defined by AI progress.
Though in many ways the nature of AI confounds past assumptions about federalism, inter-jurisdictional competition will be fierce and the outcomes of this competition at least directionally just.