Key Takeaways
- • A Carnegie Endowment for International Peace paper published June 8, 2026 by Alasdair Phillips-Robins, Teddy Tawil, and Sam Winter-Levy argues that time to power is the single most important driver of AI data center economics, ahead of energy prices and tax incentives.
- • The authors estimate a one-year delay in operation would cost an illustrative 100 MW U.S. data center more than $500 million over its life cycle, or more than 5 percent of its total value.
- • Big U.S. technology companies will spend about $670 billion this year on compute clusters, roughly 2 percent of U.S. GDP, while worldwide spending on data centers will approach $1 trillion.
- • The United States hosts roughly three-quarters of the world's advanced AI compute clusters, but the authors warn that domestic constraints on grid capacity, permitting, and politics are tightening that lead.
- • The paper proposes a democratic compute coalition, including fast-track approvals, grid reform, critical minerals cooperation, and shared AI safety and incident reporting standards among allied nations.
Summary
A new Carnegie Endowment for International Peace paper, published June 8, 2026, argues that the composition of global AI infrastructure will shape the balance of power and that democracies have a narrow window to secure their position. According to the authors, America's largest technology companies will spend about $670 billion this year building compute clusters, roughly 2 percent of U.S. GDP, and companies and governments worldwide will invest close to $1 trillion in data centers. As of May 2025, almost three-quarters of the world's advanced AI computing clusters were located on American soil, and U.S. projects move faster than most peers. The authors caution, however, that this lead is fragile given constraints on grid capacity, permitting, and political opposition, while China is mobilizing and Gulf states are offering energy and capital to attract developers. Several traditional allies lag sharply: the authors note that every major publicly reported European AI data center combined appears to contain less computing power than the Amazon-Anthropic cluster in New Carlisle, Indiana, and Australia, Italy, and South Korea have no major publicly known operational AI chip concentrations.
The paper's central finding, drawn from a financial model of a hypothetical 100 MW AI data center across ten countries, is that time to power matters more than energy costs or tax incentives. The authors estimate a one-year delay would cost more than $500 million over a twelve-year life cycle. On the authors' country rankings, the United States and the UAE lead, while Germany, with the longest timelines, ranks last. A one-year delay would drop the United States to fifth, behind the UAE, Finland, Canada, and India. The authors state that doubling U.S. electricity prices would cost less than a one-year delay, and that removing typical state tax incentives equals about 60 percent of that delay's cost.
On policy, the authors argue that governments should avoid spending scarce public funds on energy subsidies or tax cuts and instead focus on removing bottlenecks. Recommended domestic measures include fast-track review for projects meeting clean energy, grid investment, or tax criteria, interconnection queue reform, support for bottlenecked equipment production, and clean behind-the-meter power such as solar microgrids. Internationally, the paper proposes a democratic coalition with mutual fast tracks for allied infrastructure, common industrial policy for critical minerals where chokepoints lie outside democracies, joint frontier AI safety research, and common transparency and incident reporting standards. The authors note that the study relies primarily on public data and that its conclusions are tentative.
The paper's central finding, drawn from a financial model of a hypothetical 100 MW AI data center across ten countries, is that time to power matters more than energy costs or tax incentives. The authors estimate a one-year delay would cost more than $500 million over a twelve-year life cycle. On the authors' country rankings, the United States and the UAE lead, while Germany, with the longest timelines, ranks last. A one-year delay would drop the United States to fifth, behind the UAE, Finland, Canada, and India. The authors state that doubling U.S. electricity prices would cost less than a one-year delay, and that removing typical state tax incentives equals about 60 percent of that delay's cost.
On policy, the authors argue that governments should avoid spending scarce public funds on energy subsidies or tax cuts and instead focus on removing bottlenecks. Recommended domestic measures include fast-track review for projects meeting clean energy, grid investment, or tax criteria, interconnection queue reform, support for bottlenecked equipment production, and clean behind-the-meter power such as solar microgrids. Internationally, the paper proposes a democratic coalition with mutual fast tracks for allied infrastructure, common industrial policy for critical minerals where chokepoints lie outside democracies, joint frontier AI safety research, and common transparency and incident reporting standards. The authors note that the study relies primarily on public data and that its conclusions are tentative.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.