Since the dawn of large‑scale digital infrastructure, compute (the ability to process, store, and serve data) has grown from a technical resource into a strategic asset. As argued in previous analyses, the contemporary AI boom isn’t just about algorithms or models; it is equally about chips, data centers, power, supply chains, and infrastructure. This structural view of AI infrastructure, detailed in our previous article, “How AI is Rewriting Geopolitics,” shows how compute access itself is reshaping global power dynamics.

This article explores the geopolitics of compute. It examines how compute access is evolving as a form of power, analogizes, but also challenges, the notion of compute as akin to oil reserves, and considers whether compute should instead be treated more like a public utility, akin to shared infrastructure. The analysis draws on reports and data from 2022–2025, including new empirical work by the OECD, recent developments in semiconductor policy such as the CHIPS and Science Act in the United States, and industry data‑center capacity forecasts.

Compute (in practice, data centers, cloud infrastructure, and AI compute resources) has rapidly moved to the center of global economic and strategic competition.

A 2025 report by McKinsey & Company estimates that global data center power capacity has grown more than 200% over the past decade, rising from 26 gigawatts (GW) in 2015 to 81 GW in 2024. McKinsey projects that by 2030, global demand could push capacity to between 171 GW and 219 GW. Roughly 70% of this increase is expected to serve AI workloads [1][2]. This expansion demands not only servers and networking gear, but power delivery systems, cooling infrastructure, and real estate. The underlying infrastructure (energy supply, cooling, connectivity) is now as critical as compute hardware itself [3].

Recent work by OECD treats compute as a “fourth factor of production,” alongside labor, land, and traditional capital. In the 2025 working paper, the authors develop a methodology to map the global distribution of public‑cloud compute availability. This reflects a growing consensus that compute capacity exerts macroeconomic influence by shaping who can deploy AI applications, at what scale, and with what latency [4]. In turn, countries and firms that build deep compute capacity, data centers, cloud infrastructure, chip manufacturing, accrue strategic advantage akin to industrial power. Access to compute increasingly defines economic competitiveness, national security posture, and technological sovereignty.

Control over compute has concentrated significantly, especially via hyperscale cloud providers and major data center operators. A 2025 study of 775 non U.S. data centers revealed that U.S.-based companies operate approximately 48% of all non U.S. data center projects, when weighted by investment value [5]. This concentration means that even data centers built outside the United States may still be subject to the legal and regulatory influence of U.S.-based operators. Consequently, compute infrastructure becomes a channel of influence and control, rather than an evenly distributed global good. In this framework, compute resembles a strategic resource, like oil, minerals, or energy pipelines, rather than a public amenity. But that raises normative questions about equity, access, and long-term global development.

The concentration of compute power and its geopolitical implications prompt a critical question: Should compute be treated more like a private strategic asset, or like a global public utility, similar to libraries, roads, or public grids?

Building compute infrastructure (data centers, cooling, power, high speed interconnects) entails massive capital. According to McKinsey, global investment required by 2030 could reach $6.7 trillion to meet compute demand [1]. Without such investment, global compute capacity will lag behind demand.

Because compute powers AI, cloud services, governments view control over compute as a strategic asset. That justifies treating compute capacity as industrial infrastructure subject to regulation, export controls, and national security policy, not as a public good. The export control regimes instituted by the United States illustrate this logic. Since 2022, the U.S. has used export controls to restrict advanced semiconductors and AI chips from reaching certain foreign actors, notably China [5].

The compute industry exhibits signs of concentration and monopoly style control. Hardware bottlenecks (e.g., dominance by certain chip foundries), dominance of certain cloud providers, and large capital barriers create a high-risk environment for smaller firms or under resourced regions. This concentration may lead to inefficiencies, market distortions, and exclusion of emerging players, especially from the Global South.

Given these factors, some treat compute as a form of strategic capital, a commodity whose value lies in being controlled by a small number of actors with capacity to invest heavily.

An alternative, increasingly voiced by academics and policymakers, argues that compute should be treated as a public good or utility, with equitable access and regulatory frameworks that ensure broader participation. Several reasons support this perspective:

As much of economic activity moves online, digital services, cloud‑based workflows, AI‑powered applications, compute becomes foundational infrastructure, not a luxury. Denying equitable access to compute can entrench global inequalities, limiting which nations or companies can participate in the AI economy. The 2025 report by World Bank on AI adoption and digital progress notes that while cloud computing has diffused widely in high‑income countries, developing countries remain at a disadvantage, both in access to data centers and in cost, regulatory burden, or latency [6].

Treating compute like a public utility encourages decentralized, distributed infrastructure. This can improve resilience, avoid concentration risks, reduce dependency on a small number of providers, and support local digital ecosystems. Distributed compute infrastructure also aligns with sustainable development goals by enabling local innovation rather than forcing reliance on foreign cloud providers.

Without intervention, market-driven compute infrastructure may under‑serve regions with weak infrastructure, limited connectivity, or small economic returns, even where compute could offer social value. If compute is left purely to market forces, many communities or developing nations might be permanently excluded. Public‑utility framing could help correct such market failures.

To manage compute geopolitics, supply‑chain policy, export controls, and infrastructure investments have become core tools.

The 2022 CHIPS and Science Act in the U.S. aims to revitalize domestic semiconductor manufacturing. The law provides funding, subsidies, and tax credits to encourage chip manufacturing on U.S. soil, including advanced logic chips essential for AI accelerators [7][8]. By boosting domestic semiconductor capacity, the Act seeks to reduce reliance on foreign supply chains, particularly amid tensions with key chip-manufacturing hubs like Taiwan. As of 2024, the Act had already catalyzed hundreds of billions in announced investments by private firms [7][8]. This demonstrates how governments are viewing compute infrastructure, including chips, as strategic assets, with public funds deployed to shape who leads in compute. By influencing chip supply, a country can influence global compute capacity, cloud services, and AI capability.

Alongside domestic investment, export controls have become a key tool in compute geopolitics. The U.S. has progressively tightened restrictions on advanced semiconductors, with the aim of restricting competitor nations’ access to AI‑enabling hardware [7]. However, research shows that hardware‑centric controls may be insufficient. A 2024 analysis argues that firms in restricted countries have increasingly exploited software and modeling optimizations, or leveraged older, less advanced chips, to bypass hardware restrictions, a phenomenon the paper calls “Whack‑a‑Chip” [9]. In addition, as public cloud compute becomes more widespread, export controls on hardware do not prevent foreign entities from accessing compute capacity indirectly, via cloud-based services hosted in countries without restrictions. This implies that restricting chips alone does not guarantee control over compute. Critics argue that expanding controls to cover cloud services risks undermining legitimate research and innovation, especially in academia. The tension between these two framings, compute as commodity vs. compute as public utility, underpins the geopolitics of compute.

Recognizing the growing importance of compute availability, the OECD in 2025 released a working paper titled Measuring Domestic Public Cloud Compute Availability for Artificial Intelligence, which develops a methodology to map the global distribution of public‑cloud compute capacity [4]. The OECD’s work highlights that only a few economies control the majority of public‑cloud compute infrastructure, reinforcing global asymmetries. Such asymmetries shape who can deploy AI at scale, which countries can host critical workloads, and which remain marginalized. As a result, compute access is now emerging as a dimension of geopolitical inequality. By offering a standardized metric for compute availability per economy, the OECD’s framework may help policymakers, and civil‑society actors, argue for more equitable or diversified compute infrastructure.

Given these dynamics, should compute be treated like oil reserves, a scarce, strategic, highly contested resource, or as a public utility, a shared infrastructure supporting broad social and economic development?

Treating compute like oil reserves emphasizes scarcity, strategic value, and competition. The arguments in favor of this framing include:

• Strategic competition: just as oil once defined national power, now compute infrastructure, chips, data centers, cloud capacity, defines power in the digital era. Nations invest heavily, restrict access (export controls), and compete for dominance.

• Concentration and barriers to entry: The high capital cost, resource requirements (power, cooling), and regulatory overhead create high barriers to participation. Only a few actors, major cloud providers, hyperscalers, and governments, can marshal the necessary resources.

• Supply‑chain and security considerations: Semiconductors, power supply, and data‑center infrastructure have become matters of national security, supply‑chain resilience, and geopolitical leverage. Governments may view compute as critical strategic infrastructure akin to energy, defense, or manufacturing.

From this perspective, compute is a contested strategic asset whose control shapes global hierarchies.

Alternatively, treating compute as a public utility frames it as essential infrastructure that should be broadly accessible, regulated for fairness, and distributed to avoid concentration. The arguments for this framing include:

• Broad economic development: As digitalization advances, compute becomes fundamental for social services, education, small business, public administration, and more. Widespread access promotes inclusivity, reduces inequality, and fosters innovation beyond a handful of tech giants.

• Avoiding monopoly and concentration risks: Public‑utility treatment encourages decentralization, shared infrastructure, and possibly regulated or subsidized access, reducing the dominance of large incumbents.

• Stability, resilience, and sustainability: Utility‑style planning can help integrate compute infrastructure with energy grids, renewable energy, regional power planning, and ensure long‑term grid health rather than ad‑hoc competitive buildouts that may stress the grid or lead to unsustainable energy consumption.

In sum, viewing compute as public infrastructure encourages a collective, long‑term, equitable approach to digital infrastructure, especially important as digital access becomes as crucial as roads or electricity.

Given the trade‑offs, a hybrid approach may offer the best path, treating compute partly as strategic infrastructure and partly as public good. Such a framework might involve:

1. National Compute Planning & Strategic Investment: Governments should invest in domestic compute capacity (data centers, chips), especially in underserved regions, to reduce over reliance on a few providers and foster domestic digital ecosystems. Policies like the CHIPS Act are examples of this approach.

2. Transparent Global Compute Availability Metrics: Building on the work of OECD’s “Measuring Domestic Public Cloud Compute Availability for AI,” governments and multilateral institutions should track and report compute capacity distribution globally. Transparency can highlight gaps, inequities, and over-concentration, enabling targeted interventions.

3. Regulatory Frameworks for Fair Access and Local Ownership: Encourage local ownership and operation of data centers; provide incentives for publicly accessible or shared infrastructure; regulate large cloud providers to avoid monopolistic dominance; ensure open‑source, nonprofit, or academic access to compute.

4. Energy and Environmental Integration: Align compute development with sustainable energy planning, grid capacity expansion, renewable energy adoption, and efficient cooling systems. This ensures compute growth does not compromise environmental sustainability.

5. Hybrid Export/Export‑Control Policies and Cloud Governance: While strategic export controls will remain relevant for military and security use cases, policymakers must recognize that cloud compute access, even if physical hardware resides in one country, can circumvent restrictions. Governance frameworks should consider cloud‑level regulation, data sovereignty, transparency, and fair‑use provisions.

Compute, once a niche technical concern, has become central to geopolitics. The global race to build data centers, secure power supply, manufacture chips, and dominate AI infrastructure is underway. At its core, this race transforms compute into strategic capital, shaping national competitiveness, digital sovereignty, and global economic power. But treating compute purely as a commodity risks deepening inequality, concentrating power, and undermining global inclusivity. As compute becomes foundational to economic development and public services, there is strong rationale for viewing it, at least partly, as a public utility: shared, regulated, and accessible. The ideal path lies in a hybrid approach: strategic investment and national planning, combined with policies that promote equity, transparency, sustainability, and broad access. As compute infrastructure continues to scale in the coming years, the decisions made now will shape not just who builds the biggest data centers, but who builds the future of the global digital economy.

1. The Cost of Compute: A $7 Trillion Race to Scale Data Centers | McKinsey & Company (2025)
   https://www.mckinsey.com/industries/technology-media-and-telecommunications/our-insights/the-cost-of-compute-a-7-trillion-dollar-race-to-scale-data-centers

1. Data Centres: Powering the Growth of AI and Cloud Computing | Macquarie Asset Management (2024)
   https://www.macquarie.com/uk/en/about/company/macquarie-asset-management/institutional/insights/data-centres-powering-the-growth-of-ai-and-cloud-computing.html

1. Beyond Compute: Infrastructure That Powers and Cools AI Data Centers | McKinsey & Company (2025)
   https://www.mckinsey.com/industrials-and-electronics/our-insights/beyond-compute-infrastructure-that-powers-and-cools-ai-data-centers

1. Measuring Domestic Public Cloud Compute Availability for Artificial Intelligence | Lehdonvirta, V., Wu, B., Hawkins, Z., Caira, C., & Russo, L. (2025), OECD Artificial Intelligence Papers https://www.oecd.org/content/dam/oecd/en/publications/reports/2025/10/measuring-domestic-public-cloud-compute-availability-for-artificial-intelligence_39fa6b0e/8602a322-en.pdf

1. How Sovereign Is Sovereign Compute? A Review of 775 Non-U.S. Data Centers | Richardson, A. et al. (2025), arXiv
   https://arxiv.org/abs/2508.00932

1. Digital Progress and Trends Report 2025: Strengthening AI for Development | World Bank (2025)
   https://documents1.worldbank.org/curated/en/099112525160536089/pdf/P505350-59c98ca8-0803-4f23-b470-17f3dab010ab.pdf

1. The CHIPS Act: How U.S. Microchip Factories Could Reshape the Economy | Council on Foreign Relations (2025) https://www.cfr.org/in-brief/chips-act-how-us-microchip-factories-could-reshape-economy

1. 2024 State of the U.S. Semiconductor Industry | Semiconductor Industry Association (2024) https://www.semiconductors.org/wp-content/uploads/2024/10/SIA_2024_State-of-Industry-Report_2024.pdf

1. Whack-a-Chip: The Futility of Hardware-Centric Export Controls | Gupta, R., Walker, L., & Reddie, A. W. (2024), arXiv
   https://arxiv.org/abs/2411.14425

Since the dawn of large‑scale digital infrastructure, compute (the ability to process, store, and serve data) has grown from a technical resource into a strategic asset. As argued in previous analyses, the contemporary AI boom isn’t just about algorithms or models; it is equally about chips, data centers, power, supply chains, and infrastructure. This structural view of AI infrastructure, detailed in our previous article, “How AI is Rewriting Geopolitics,” shows how compute access itself is reshaping global power dynamics.

This article explores the geopolitics of compute. It examines how compute access is evolving as a form of power, analogizes, but also challenges, the notion of compute as akin to oil reserves, and considers whether compute should instead be treated more like a public utility, akin to shared infrastructure. The analysis draws on reports and data from 2022–2025, including new empirical work by the OECD, recent developments in semiconductor policy such as the CHIPS and Science Act in the United States, and industry data‑center capacity forecasts.

Compute (in practice, data centers, cloud infrastructure, and AI compute resources) has rapidly moved to the center of global economic and strategic competition.

A 2025 report by McKinsey & Company estimates that global data center power capacity has grown more than 200% over the past decade, rising from 26 gigawatts (GW) in 2015 to 81 GW in 2024. McKinsey projects that by 2030, global demand could push capacity to between 171 GW and 219 GW. Roughly 70% of this increase is expected to serve AI workloads [1][2]. This expansion demands not only servers and networking gear, but power delivery systems, cooling infrastructure, and real estate. The underlying infrastructure (energy supply, cooling, connectivity) is now as critical as compute hardware itself [3].

Recent work by OECD treats compute as a “fourth factor of production,” alongside labor, land, and traditional capital. In the 2025 working paper, the authors develop a methodology to map the global distribution of public‑cloud compute availability. This reflects a growing consensus that compute capacity exerts macroeconomic influence by shaping who can deploy AI applications, at what scale, and with what latency [4]. In turn, countries and firms that build deep compute capacity, data centers, cloud infrastructure, chip manufacturing, accrue strategic advantage akin to industrial power. Access to compute increasingly defines economic competitiveness, national security posture, and technological sovereignty.

Control over compute has concentrated significantly, especially via hyperscale cloud providers and major data center operators. A 2025 study of 775 non U.S. data centers revealed that U.S.-based companies operate approximately 48% of all non U.S. data center projects, when weighted by investment value [5]. This concentration means that even data centers built outside the United States may still be subject to the legal and regulatory influence of U.S.-based operators. Consequently, compute infrastructure becomes a channel of influence and control, rather than an evenly distributed global good. In this framework, compute resembles a strategic resource, like oil, minerals, or energy pipelines, rather than a public amenity. But that raises normative questions about equity, access, and long-term global development.

The concentration of compute power and its geopolitical implications prompt a critical question: Should compute be treated more like a private strategic asset, or like a global public utility, similar to libraries, roads, or public grids?

Building compute infrastructure (data centers, cooling, power, high speed interconnects) entails massive capital. According to McKinsey, global investment required by 2030 could reach $6.7 trillion to meet compute demand [1]. Without such investment, global compute capacity will lag behind demand.

Because compute powers AI, cloud services, governments view control over compute as a strategic asset. That justifies treating compute capacity as industrial infrastructure subject to regulation, export controls, and national security policy, not as a public good. The export control regimes instituted by the United States illustrate this logic. Since 2022, the U.S. has used export controls to restrict advanced semiconductors and AI chips from reaching certain foreign actors, notably China [5].

The compute industry exhibits signs of concentration and monopoly style control. Hardware bottlenecks (e.g., dominance by certain chip foundries), dominance of certain cloud providers, and large capital barriers create a high-risk environment for smaller firms or under resourced regions. This concentration may lead to inefficiencies, market distortions, and exclusion of emerging players, especially from the Global South.

Given these factors, some treat compute as a form of strategic capital, a commodity whose value lies in being controlled by a small number of actors with capacity to invest heavily.

An alternative, increasingly voiced by academics and policymakers, argues that compute should be treated as a public good or utility, with equitable access and regulatory frameworks that ensure broader participation. Several reasons support this perspective:

As much of economic activity moves online, digital services, cloud‑based workflows, AI‑powered applications, compute becomes foundational infrastructure, not a luxury. Denying equitable access to compute can entrench global inequalities, limiting which nations or companies can participate in the AI economy. The 2025 report by World Bank on AI adoption and digital progress notes that while cloud computing has diffused widely in high‑income countries, developing countries remain at a disadvantage, both in access to data centers and in cost, regulatory burden, or latency [6].

Treating compute like a public utility encourages decentralized, distributed infrastructure. This can improve resilience, avoid concentration risks, reduce dependency on a small number of providers, and support local digital ecosystems. Distributed compute infrastructure also aligns with sustainable development goals by enabling local innovation rather than forcing reliance on foreign cloud providers.

Without intervention, market-driven compute infrastructure may under‑serve regions with weak infrastructure, limited connectivity, or small economic returns, even where compute could offer social value. If compute is left purely to market forces, many communities or developing nations might be permanently excluded. Public‑utility framing could help correct such market failures.

To manage compute geopolitics, supply‑chain policy, export controls, and infrastructure investments have become core tools.

The 2022 CHIPS and Science Act in the U.S. aims to revitalize domestic semiconductor manufacturing. The law provides funding, subsidies, and tax credits to encourage chip manufacturing on U.S. soil, including advanced logic chips essential for AI accelerators [7][8]. By boosting domestic semiconductor capacity, the Act seeks to reduce reliance on foreign supply chains, particularly amid tensions with key chip-manufacturing hubs like Taiwan. As of 2024, the Act had already catalyzed hundreds of billions in announced investments by private firms [7][8]. This demonstrates how governments are viewing compute infrastructure, including chips, as strategic assets, with public funds deployed to shape who leads in compute. By influencing chip supply, a country can influence global compute capacity, cloud services, and AI capability.

Alongside domestic investment, export controls have become a key tool in compute geopolitics. The U.S. has progressively tightened restrictions on advanced semiconductors, with the aim of restricting competitor nations’ access to AI‑enabling hardware [7]. However, research shows that hardware‑centric controls may be insufficient. A 2024 analysis argues that firms in restricted countries have increasingly exploited software and modeling optimizations, or leveraged older, less advanced chips, to bypass hardware restrictions, a phenomenon the paper calls “Whack‑a‑Chip” [9]. In addition, as public cloud compute becomes more widespread, export controls on hardware do not prevent foreign entities from accessing compute capacity indirectly, via cloud-based services hosted in countries without restrictions. This implies that restricting chips alone does not guarantee control over compute. Critics argue that expanding controls to cover cloud services risks undermining legitimate research and innovation, especially in academia. The tension between these two framings, compute as commodity vs. compute as public utility, underpins the geopolitics of compute.

Recognizing the growing importance of compute availability, the OECD in 2025 released a working paper titled Measuring Domestic Public Cloud Compute Availability for Artificial Intelligence, which develops a methodology to map the global distribution of public‑cloud compute capacity [4]. The OECD’s work highlights that only a few economies control the majority of public‑cloud compute infrastructure, reinforcing global asymmetries. Such asymmetries shape who can deploy AI at scale, which countries can host critical workloads, and which remain marginalized. As a result, compute access is now emerging as a dimension of geopolitical inequality. By offering a standardized metric for compute availability per economy, the OECD’s framework may help policymakers, and civil‑society actors, argue for more equitable or diversified compute infrastructure.

Given these dynamics, should compute be treated like oil reserves, a scarce, strategic, highly contested resource, or as a public utility, a shared infrastructure supporting broad social and economic development?

Treating compute like oil reserves emphasizes scarcity, strategic value, and competition. The arguments in favor of this framing include:

• Strategic competition: just as oil once defined national power, now compute infrastructure, chips, data centers, cloud capacity, defines power in the digital era. Nations invest heavily, restrict access (export controls), and compete for dominance.

• Concentration and barriers to entry: The high capital cost, resource requirements (power, cooling), and regulatory overhead create high barriers to participation. Only a few actors, major cloud providers, hyperscalers, and governments, can marshal the necessary resources.

• Supply‑chain and security considerations: Semiconductors, power supply, and data‑center infrastructure have become matters of national security, supply‑chain resilience, and geopolitical leverage. Governments may view compute as critical strategic infrastructure akin to energy, defense, or manufacturing.

From this perspective, compute is a contested strategic asset whose control shapes global hierarchies.

Alternatively, treating compute as a public utility frames it as essential infrastructure that should be broadly accessible, regulated for fairness, and distributed to avoid concentration. The arguments for this framing include:

• Broad economic development: As digitalization advances, compute becomes fundamental for social services, education, small business, public administration, and more. Widespread access promotes inclusivity, reduces inequality, and fosters innovation beyond a handful of tech giants.

• Avoiding monopoly and concentration risks: Public‑utility treatment encourages decentralization, shared infrastructure, and possibly regulated or subsidized access, reducing the dominance of large incumbents.

• Stability, resilience, and sustainability: Utility‑style planning can help integrate compute infrastructure with energy grids, renewable energy, regional power planning, and ensure long‑term grid health rather than ad‑hoc competitive buildouts that may stress the grid or lead to unsustainable energy consumption.

In sum, viewing compute as public infrastructure encourages a collective, long‑term, equitable approach to digital infrastructure, especially important as digital access becomes as crucial as roads or electricity.

Given the trade‑offs, a hybrid approach may offer the best path, treating compute partly as strategic infrastructure and partly as public good. Such a framework might involve:

1. National Compute Planning & Strategic Investment: Governments should invest in domestic compute capacity (data centers, chips), especially in underserved regions, to reduce over reliance on a few providers and foster domestic digital ecosystems. Policies like the CHIPS Act are examples of this approach.

2. Transparent Global Compute Availability Metrics: Building on the work of OECD’s “Measuring Domestic Public Cloud Compute Availability for AI,” governments and multilateral institutions should track and report compute capacity distribution globally. Transparency can highlight gaps, inequities, and over-concentration, enabling targeted interventions.

3. Regulatory Frameworks for Fair Access and Local Ownership: Encourage local ownership and operation of data centers; provide incentives for publicly accessible or shared infrastructure; regulate large cloud providers to avoid monopolistic dominance; ensure open‑source, nonprofit, or academic access to compute.

4. Energy and Environmental Integration: Align compute development with sustainable energy planning, grid capacity expansion, renewable energy adoption, and efficient cooling systems. This ensures compute growth does not compromise environmental sustainability.

5. Hybrid Export/Export‑Control Policies and Cloud Governance: While strategic export controls will remain relevant for military and security use cases, policymakers must recognize that cloud compute access, even if physical hardware resides in one country, can circumvent restrictions. Governance frameworks should consider cloud‑level regulation, data sovereignty, transparency, and fair‑use provisions.

Compute, once a niche technical concern, has become central to geopolitics. The global race to build data centers, secure power supply, manufacture chips, and dominate AI infrastructure is underway. At its core, this race transforms compute into strategic capital, shaping national competitiveness, digital sovereignty, and global economic power. But treating compute purely as a commodity risks deepening inequality, concentrating power, and undermining global inclusivity. As compute becomes foundational to economic development and public services, there is strong rationale for viewing it, at least partly, as a public utility: shared, regulated, and accessible. The ideal path lies in a hybrid approach: strategic investment and national planning, combined with policies that promote equity, transparency, sustainability, and broad access. As compute infrastructure continues to scale in the coming years, the decisions made now will shape not just who builds the biggest data centers, but who builds the future of the global digital economy.

1. The Cost of Compute: A $7 Trillion Race to Scale Data Centers | McKinsey & Company (2025)
   https://www.mckinsey.com/industries/technology-media-and-telecommunications/our-insights/the-cost-of-compute-a-7-trillion-dollar-race-to-scale-data-centers

1. Data Centres: Powering the Growth of AI and Cloud Computing | Macquarie Asset Management (2024)
   https://www.macquarie.com/uk/en/about/company/macquarie-asset-management/institutional/insights/data-centres-powering-the-growth-of-ai-and-cloud-computing.html

1. Beyond Compute: Infrastructure That Powers and Cools AI Data Centers | McKinsey & Company (2025)
   https://www.mckinsey.com/industrials-and-electronics/our-insights/beyond-compute-infrastructure-that-powers-and-cools-ai-data-centers

1. Measuring Domestic Public Cloud Compute Availability for Artificial Intelligence | Lehdonvirta, V., Wu, B., Hawkins, Z., Caira, C., & Russo, L. (2025), OECD Artificial Intelligence Papers https://www.oecd.org/content/dam/oecd/en/publications/reports/2025/10/measuring-domestic-public-cloud-compute-availability-for-artificial-intelligence_39fa6b0e/8602a322-en.pdf

1. How Sovereign Is Sovereign Compute? A Review of 775 Non-U.S. Data Centers | Richardson, A. et al. (2025), arXiv
   https://arxiv.org/abs/2508.00932

1. Digital Progress and Trends Report 2025: Strengthening AI for Development | World Bank (2025)
   https://documents1.worldbank.org/curated/en/099112525160536089/pdf/P505350-59c98ca8-0803-4f23-b470-17f3dab010ab.pdf

1. The CHIPS Act: How U.S. Microchip Factories Could Reshape the Economy | Council on Foreign Relations (2025) https://www.cfr.org/in-brief/chips-act-how-us-microchip-factories-could-reshape-economy

1. 2024 State of the U.S. Semiconductor Industry | Semiconductor Industry Association (2024) https://www.semiconductors.org/wp-content/uploads/2024/10/SIA_2024_State-of-Industry-Report_2024.pdf

1. Whack-a-Chip: The Futility of Hardware-Centric Export Controls | Gupta, R., Walker, L., & Reddie, A. W. (2024), arXiv
   https://arxiv.org/abs/2411.14425