Breaking Geopolitics News & AnalysisWednesday, November 18, 2026
DiplomacyNorth America

Blocking US Data Centers Won’t Stop the Rise of an AI Oligarchy

The National Interest
November 18, 2026 at 4:09 PM
5 views
Blocking US Data Centers Won’t Stop the Rise of an AI Oligarchy

Halting new data centers won’t stop AI’s gains from concentrating in a few firms, so Congress should focus instead on spreading AI adoption across the economy. The post Blocking US Data Centers Won’t Stop the Rise of an AI Oligarchy appeared first on The National Interest.

Halting new data centers won’t stop AI’s gains from concentrating in a few firms, so Congress should focus instead on spreading AI adoption across the economy.

Following the November 3 midterms, Democrats took the House and the Senate. With Democrats in both chambers, the policy debate is likely to go well beyond the location of data centers to their social impact. It is entirely plausible that there will be increased congressional scrutiny of data center permitting, community consent, environmental disclosure, and whether these projects deliver broad economic benefits or primarily benefit a very small number of companies.

Data center politics were a big factor in Texas, Georgia, and Ohio, at least. The risk right now is that policymakers will get too fixated on where artificial intelligence (AI) is actually being run and end up with policies that don’t in any way address the distribution of the benefits or the wider economic impact. If Congress is still focused on location, then they’re going to miss the systemic barriers to adoption and are going to end up underinvesting in making the tools available and useful to a much wider range of industries and communities.

America has a choice. It can go down the path of an AI economy or an AI-garchy. The difference is vital. An AI economy is one in which the benefits of AI are broadly distributed across the economy (such as manufacturers, hospitals, schools, and government). An AI-garchy is one in which the benefits of AI are increasingly accruing to a very small number of frontier model developers and hyperscalers (AI oligarchs), and the rest of the economy sees only very modest productivity gains. AI-garchy is already happening. It’s not a new phenomenon in America, and blocking data centers in the United States won’t stop it or reduce the power of AI oligarchs.

Why Blocking Data Centers Won’t Curb Big Tech’s Power

A March 2026 Gallup survey found that 71 percent of Americans opposed building an AI data center in their local area, compared with 53 percent who opposed a nuclear power plant. Public opposition is already translating into political action: dozens of referendums restricting data-center development passed in the 2026 midterm elections.

While the midterm results show that voters’ acceptance of AI expansion can’t be assumed, nuclear plants and high-voltage lines faced similar public and regulatory opposition. National goals clashed with local safety and environmental concerns. These precedents may help predict policy and public responses to AI data centers: Goldman Sachs has warned that potential state-level moratoriums and tighter regulation could affect projects accounting for roughly 45 percent of planned US data center capacity expansion.

One consequence is that AI infrastructure investment is increasingly flowing to Asia. Microsoft has expanded cloud and AI capacity in Malaysia and Indonesia, while Google recently committed $15 billion to data center investments in Malaysia. Southeast Asia now accounts for a growing share of Asia-Pacific data center capacity under construction. As permitting hurdles and community opposition intensify, the region is positioning itself as a key destination for the next wave of global AI-computing infrastructure.

This is a paradox. Concentration is already in place in terms of who owns the models, and moving the energy-intensive infrastructure to Asia doesn’t change that at all. The same companies still own the frontier models. If Big Tech is going to avoid a much deeper political backlash, then greater transparency is absolutely going to be essential. 

Communities are increasingly demanding disclosure of projected electricity demand, water consumption, land use, tax incentives, grid upgrade costs, environmental impacts, community benefits, and the use of nondisclosure agreements (NDAs) in local negotiations. More open reporting would go a long way to separating out genuine economic development from the perception that communities are being asked to take on the cost without any proper information or accountability.

They’re still the gatekeepers, they still set the prices, they still determine portability and access. America may end up with fewer data centers and fewer jobs, but the concentration of power is exactly the same, just shifted to servers in Asia. That’s why their power is not going to be reduced if they’re required to move their data centers outside of US territory.

Why AI Adoption, Not Computing Power, Is the Real Constraint

Data center locations may be secondary. The key question is whether AI boosts productivity across the economy. Even if all proposed US data centers were built tomorrow, turning AI capability into economy-wide productivity would remain a much harder challenge. The constraint has changed. In the early days of the AI race, the constraint was compute. Today the constraint is adoption.

Congress should consider concrete policies such as investing in AI literacy across the board. It should involve workforce training in prompt design, data quality, human-in-the-loop oversight, model limitations, cybersecurity risks, procurement, and the measurement of outcomes. Success should be measured in real terms via metrics such as the proportion of firms using AI across multiple functions, reductions in administrative processing times, improvements in inventory accuracy, reductions in patient wait times, faster permitting times, and productivity gains per worker. The question isn’t whether a company has bought AI; it’s whether it can show that it’s actually achieved measurable improvements in performance as a result of its adoption.

Building a bigger model isn’t the same as helping a manufacturer get its orders filled, reducing the amount of time a patient has to wait at a hospital, or enabling a public body to respond to the public more effectively. The tech companies can provide the tools. They can’t restructure institutions to make effective use of those tools. The history of transformative technologies is that the biggest gains have come not when a small number of firms have the capability, but when thousands of firms have adopted it.

We’ve seen this before. Electric motors were being put into factories from the 1890s onwards, but the real productivity gains didn’t come until much later when owners redesigned plants around the new technology. Economists describe this as a J-curve: productivity initially rises slowly as organizations adapt to a new technology, before accelerating once processes and institutions are redesigned around it.

The early evidence is that we’re still very much at the bottom of that curve. Census Bureau data for November 2025 to January 2026 show that around 18 percent of firms were using AI in at least one function. Of those firms, 57 percent were using it in three or fewer functions. Access to capable systems is not the same as integration across the economy, and that’s where the productivity gains are either going to be made or lost.

This is also a question of competition. Countries rarely win technological races simply by inventing new technologies. They win by deploying them at scale. A country can be number one in the world in terms of frontier models and still be a laggard if those capabilities don’t get deployed to improve productivity in factories, hospitals, logistics firms, and the public sector.

What Congress Should Measure to Spread AI Across the Economy

Think of a small manufacturer looking at an AI scheduling tool. The real issue is whether the inventory records are accurate, whether the purchasing system actually feeds into the production system, and whether the recommendations can be acted upon. In many cases, the most valuable investment won’t be in the AI subscription at all but in the underlying organizational work required to make that subscription useful.

The same applies to public services. A hospital trialing AI to do admin work needs to know whether it actually reduces workload or just creates more errors. A government department rolling out a chatbot needs to know whether it’s actually solving people’s problems rather than simply giving them a faster response. Those are tests of organizational effectiveness, not tests of computing power.

Congress should be looking at American AI leadership in terms of two things: what our companies can invent and what our institutions can deliver. The metrics there could be the percentage of small and medium-sized enterprises (SMEs) using AI in their day-to-day operations, the number of federal and state agencies deploying audited AI systems, productivity growth in AI-intensive sectors, the percentage of the workforce completing AI training, and the percentage of AI projects that generate documented cost savings or service improvements. What gets measured is more likely to get deployed at scale.

This means Congress should set out specific, measurable metrics to guide policy and oversight, not just with respect to innovation at the frontier but with respect to whether the capability that’s being developed is actually spreading and creating real economic value.

Some will point out that future AI systems will be so sophisticated that they’ll be able to deal with a lot of the integration issues on their own. They won’t be able to fix inventory records that nobody actually keeps; they can’t possibly work out who made a decision or who is to blame when things go wrong. That’s the realm of people, of management and of authority.

Leadership in AI won’t be decided by who owns the most data centers or where they are located. It will be about who deploys AI most effectively across the economy. An AI-garchy can survive without a single data center right next door. Rather than halt the build (which is what some in DC seem to be advocating), we should be working to ensure the whole economy has a say in the intelligence it’s going to be running. The policy focus should be on the key stakeholders that are in a position to actually drive the adoption of AI across the economy, including SMEs, the public sector (such as hospitals and schools), local governments, and the workforce that’s going to be implementing and using the systems.

America’s AI future will be decided not by where the servers sit, but by whether AI spreads beyond a handful of oligarchs and into the factories, hospitals, schools, and public institutions where productivity is actually created. When a few brilliant AI companies lead the way while the rest of society’s institutions stand still, we don’t have an AI economy. We have an AI-garchy, and it runs the show wherever the data centers sit, whether at home or overseas.

About the Author: Bruno S. Sergi

Bruno S. Sergi is an instructor at Harvard University’s Sustainability and Global Development Practice Graduate Programs. He is also affiliated with the Harvard Center for International Development, the Davis Center for Russian and Eurasian Studies, and the Harvard University Asia Center. He has led the launch of multiple scholarly journals and book series, including the Cambridge Elements series at Cambridge University Press and Entrepreneurship and Global Economic Growth at Emerald Publishing.

The post Blocking US Data Centers Won’t Stop the Rise of an AI Oligarchy appeared first on The National Interest.