Cato Institute: The US needs to speed up investment in AI infrastructure and energy

The US risks hitting a wall not because it runs out of AI ideas, but because it runs short of electricity, sites, and coherent infrastructure policy. If the country truly wants to keep its lead in artificial intelligence, it has to do more than debate model regulation: it needs to accelerate investment in energy, grids, and data centers. Kevin Frazier, a visiting fellow at the Cato Institute, is drawing attention to exactly this gap between political ambition and physical capacity.

A shift in the AI conversation

According to Frazier, Washington is now trying to work out what a national AI framework should look like and which tools can back up its stated goal of technological leadership. That marks an important shift: the conversation about the AI market is gradually moving beyond debates over model risks, copyright, and safety. A more down-to-earth question is taking center stage — can the country quickly deploy the compute needed to train and serve modern AI systems? In that context, a national framework isn't just a set of rules for developers; it's also a clear signal to investors, data center operators, and energy companies.

Infrastructure as the bottleneck

This is where infrastructure becomes the main bottleneck. Large data centers need more than chips and servers: they require enormous amounts of electricity, grid connections, land, cooling, and construction permits. A single modern campus built for AI workloads can draw hundreds of megawatts, and in some cases its demand approaches that of a small city.

Meanwhile, new generating capacity, transmission lines, and substations take far longer to build than new models and services take to launch. The result is that the technology cycle keeps speeding up while the energy and construction cycles don't. That's why, even with capital and demand in place, bringing new capacity online can be delayed for years.

For companies, this means rising costs, pushed-back timelines, and more cautious investment decisions. For Washington, it turns AI from a purely digital issue into a matter of industrial policy. The national framework under discussion will likely need to cover not only rules for using AI but also the conditions for scaling it: access to power, predictable regulatory requirements, project approvals, and clear incentives for private investment.

Striking a balance between a federal approach and local constraints also matters, because many real-world barriers arise at the level of states, utilities, and municipalities.

Business needs a planning horizon

If the government wants American companies to build infrastructure at home, those companies need a planning horizon. Businesses can pour billions into compute clusters, but they won't keep up the same pace if grid interconnection, permitting, and local restrictions become unpredictable. Frazier's argument also matters because it shifts the focus of the debate over US leadership.

AI leadership doesn't go only to whoever has the strongest models, but also to whoever builds the entire supply chain fastest — from energy and data centers to network infrastructure and access to compute. Under this logic, the winners include not just AI developers but also energy companies, industrial site developers, equipment manufacturers, and regions that can move approvals through faster. The losers are jurisdictions where AI demand already exists but the physical base to support it can't keep up.

This also changes who benefits from the AI boom: part of the value created will flow not only into software but into heavy infrastructure.

Conclusion

The takeaway is fairly practical: the next phase of the AI race will be decided not only by the quality of algorithms but by the speed of construction. If the US wants its talk of technological leadership to be more than a declaration, it will have to align its AI strategy with energy, grids, and capital projects. Otherwise, a capacity shortage will become a constraint that no model can remove on its own. That's why the AI debate is increasingly a debate over who can turn compute demand into real megawatts, buildings, and connected server capacity the fastest.

In brief:

• According to Cato Institute researcher Kevin Frazier, US leadership in AI hinges on energy, sites, and infrastructure policy.
• A single campus for AI workloads can draw hundreds of megawatts, while grids and generating capacity take longer to build than models take to launch.
• The next phase of the AI race will be decided by the speed of infrastructure construction, not just the quality of algorithms.