A new Carnegie Endowment paper argues that the race to build AI infrastructure is no longer just an economic competition. It is becoming a question of geopolitical influence, with the speed of getting data centers online emerging as the most important factor in determining which countries gain an edge.
The paper, written by Alasdair Phillips-Robins, Teddy Tawil and Sam Winter-Levy, says democracies have a limited window to establish leadership in the buildout of advanced compute. The authors contend that the countries hosting the most powerful AI clusters will help shape how future AI systems are governed, what values they reflect and how they are used.
The report comes as countries and companies around the world are spending heavily on data centers and compute clusters. It says U.S. technology companies alone are expected to invest hundreds of billions of dollars this year in AI infrastructure, while global spending on data centers is approaching the trillion-dollar mark.
Carnegie says the United States remains the clear leader in advanced AI computing capacity, hosting the majority of the world’s most powerful clusters. But the paper warns that the advantage could narrow quickly. Domestic limits on electricity grids, permitting and political opposition are slowing projects in the U.S., while China is pushing to close the gap and Gulf states are trying to lure developers with cheap power and capital.
The paper also says many U.S. allies risk falling behind. It points to Europe as a region where even the largest publicly known AI data center plans appear to trail major U.S. projects, and it notes that some allies have no major publicly known operational AI chip concentrations at all.
The central conclusion of the paper is that the most important variable is not electricity cost or tax policy, but time to power, meaning how quickly a project can move from planning to operation.
Using a financial model of a hypothetical 100-megawatt AI data center across 10 countries, the authors say that delays are far more expensive than most other policy levers. A one-year delay, they estimate, can reduce the life-cycle value of a U.S. data center by more than $500 million. The report says that is costlier than a doubling of electricity prices, typical state tax incentives, or moderate tariffs on AI servers.
The authors argue that the reason is straightforward. Every month a facility is delayed is a month of forgone revenue. They also say capital tied up in unfinished projects cannot earn returns, and that future revenue is less valuable than revenue generated sooner.
In the paper’s country rankings, the United States and the United Arab Emirates come out on top largely because projects can move quickly there. Germany ranks worst because timelines are longest. But the paper says the rankings can change sharply if delays increase or if other countries speed up their permitting and grid connection processes.
The paper concludes that no single democracy can build the world’s AI infrastructure alone. Instead, it calls for a broader coalition of democratic countries that can pool advantages in land, capital, power generation and supply chains.
Among the recommendations are faster review processes for qualifying projects, reforms to grid interconnection systems, steps to improve flexibility and resilience, and support for behind-the-meter clean power such as solar microgrids and wind farm arrangements.
Internationally, the authors urge democracies to coordinate more closely on AI infrastructure and supply chains. They suggest mutual fast tracks for allied investment, joint industrial policy on chokepoint materials, shared safety and security research, and common standards for transparency and incident reporting.
The paper frames the issue as one of relative progress, not an open-ended push to build more AI capacity everywhere. Its argument is that democracies need to move faster than authoritarian rivals if they want the center of gravity in advanced AI to remain in the democratic world.