The Compute Coalition: How to Build the Future of AI in the Free World

Report an error

Summary

AI-generated

This summary is written by a language model reading the source document. It is not the publisher's words and is not a substitute for the original.

Learn more about AI enrichment

The report identifies 'time to power'—the speed at which data centers can be permitted and connected to electricity—as the primary driver of economic returns and competitiveness for AI infrastructure. Using a financial model, the authors find that the United States and the UAE are currently the most competitive locations, while Germany is the least. The authors argue that the U.S. cannot maintain its lead alone due to grid constraints and political opposition, necessitating a coalition of democracies to pool geography, energy, and capital. Recommended domestic reforms include fast-track permitting and grid flexibility, while international efforts should focus on securing critical minerals and harmonizing investment rules across the 'free world'.

Key insights

AI-generated

These insights are written by a language model reading the source document. They are not the publisher's words and are not a substitute for the original.

Learn more about AI enrichment
  • The speed of getting a data center operational ('time to power') is the most critical factor for economic returns, far outweighing the impact of electricity costs, tax incentives, or tariffs. For a typical 100 MW U.S. data center, a one-year delay in operation reduces its life cycle value by approximately $543 million, or about 5.5%.
  • The United States and the United Arab Emirates (UAE) are the most competitive locations for AI data centers due to faster project timelines. In contrast, Germany is the least competitive, with the longest timelines. The UAE's average time from project approval to first operation is approximately 22 months, while the U.S. average is 24 months and Germany's is 42 months.
  • IT equipment, primarily GPUs, dominates the capital expenditures of AI data centers, accounting for roughly 70% to 80% of total costs. This means that non-IT costs, such as land and construction labor, have a relatively small impact on overall project economics compared to the cost of delays.
  • Behind-the-meter (BTM) power, such as on-site natural gas turbines, serves as a critical temporary bridge to bypass long grid connection queues, potentially shortening timelines by a year or more. However, BTM is not a permanent solution due to higher operating costs, environmental emissions, and scale limitations for gigawatt-level clusters.
  • The geopolitical distribution of compute is a strategic imperative because the countries hosting the infrastructure can control access, regulate use, and embed specific values into AI systems. A lack of domestic compute creates vulnerabilities in military intelligence, fiscal stability, and the ability to shape international safety standards.
  • While the U.S. currently leads in AI compute (hosting roughly 75% of advanced clusters as of May 2025), this lead is threatened by China's massive state-directed mobilization and the U.S.'s own domestic constraints, including grid capacity and rising political opposition to data center construction.
  • The U.S. should lead a 'Compute Coalition' of democracies to pool resources and bypass domestic limits. Such a coalition would combine Canada's hydroelectric resources, Northern Europe's cool climates, and Australia's solar potential to create a more resilient and scalable infrastructure than any single nation could provide.
  • To improve competitiveness, democratic governments should avoid expensive energy subsidies and instead focus on regulatory reforms: creating fast-track permitting 'lanes', implementing 'connect and manage' grid policies to allow earlier hookups, and incentivizing the production of bottlenecked equipment like large transformers.

Cite the original document

APA
Phillips-Robins, A., Tawil, T., & Winter-Levy, S. (2026). The Compute Coalition: How to Build the Future of AI in the Free World. Carnegie Endowment for International Peace. https://assets.carnegieendowment.org/files/Phillips-Robins%20et%20al._Compute_final.pdf
Chicago
Phillips-Robins, Alasdair, Teddy Tawil, and Sam Winter-Levy. The Compute Coalition: How to Build the Future of AI in the Free World. Carnegie Endowment for International Peace, 2026. https://assets.carnegieendowment.org/files/Phillips-Robins%20et%20al._Compute_final.pdf.
Wikipedia
{{cite report |last1=Phillips-Robins |first1=Alasdair |last2=Tawil |first2=Teddy |last3=Winter-Levy |first3=Sam |title=The Compute Coalition: How to Build the Future of AI in the Free World |publisher=Carnegie Endowment for International Peace |date=June 2026 |url=https://assets.carnegieendowment.org/files/Phillips-Robins%20et%20al._Compute_final.pdf |access-date=5 November 2026 |via=Climate Insights Directory (document carnend-003277)}}
BibTeX
@techreport{phillipsrobins2026compute, author = {Phillips-Robins, Alasdair and Tawil, Teddy and Winter-Levy, Sam}, title = {{The Compute Coalition: How to Build the Future of AI in the Free World}}, institution = {Carnegie Endowment for International Peace}, year = {2026}, month = jun, url = {https://assets.carnegieendowment.org/files/Phillips-Robins%20et%20al._Compute_final.pdf}, urldate = {2026-11-05}, note = {Indexed by Climate Insights Directory, document carnend-003277} }

Full text

Collected · Record updated