Who Controls AI Compute?
The New Geopolitics of Chips, Data and Power
Artificial intelligence is often portrayed as a contest between algorithms or companies: OpenAI versus Google, Nvidia versus Advanced Micro Devices (AMD), the United States versus China. Yet the decisive struggle may be taking place somewhere far less visible. Behind every large language model sits an industrial infrastructure of advanced semiconductors, hyperscale data centres, electricity grids, minerals and cloud networks. The emerging AI race is therefore not simply a race to build better software, but a contest over who controls the physical resources required to make intelligence scalable.
This contest matters geopolitically because these resources are highly concentrated. The World Bank’s World Development Report 2026 warns that control over key parts of the AI value chain by the leading powers (especially the US and China) including raw materials, advanced chips and data centres, creates dependency risks for other countries (World Bank, 2026). Stanford’s AI Index 2026 finds that the US hosts 5,427 data centres, more than ten times any other country, while the Taiwan Semiconductor Manufacturing Company (TSMC) fabricates almost every leading AI chip (Stanford HAI, 2026). The apparent abundance of AI services therefore rests on a remarkably concentrated infrastructure.
From Silicon Valley to Silicon Geopolitics
Semiconductors have already become instruments of geopolitical strategy. Since 2022, the United States and its allies have progressively restricted China’s access to advanced AI chips and semiconductor-manufacturing equipment, explicitly linking technological capabilities to national security. These controls restricted trade and encouraged Beijing to accelerate its drive for technological self-reliance. Recent analysis by American think tank the Center for Strategic and International Studies suggests that US and allied controls have contributed to China’s push to localise semiconductor design and manufacturing (Shivakumar, Wessner and Howell, 2026).
The result is a feedback loop. Washington seeks to protect its technological advantage by limiting access to critical technologies; Beijing responds by investing more heavily in domestic alternatives. Both sides consequently treat semiconductor supply chains as strategic infrastructure rather than ordinary commercial networks. The partnership announced on 30 September between China’s DeepSeek and Huawei to develop programming tools optimised for Huawei’s AI processor Ascend chips illustrates how the competition is moving beyond individual processors towards entire technological ecosystems (Reuters, 2026).
The geography of AI matters. TSMC’s position is particularly consequential because advanced AI chips designed by companies such as Nvidia and AMD depend heavily on specialised semiconductor fabrication. Stanford’s AI Index notes that this leaves the global AI hardware supply chain dependent on a single foundry in Taiwan, even though TSMC expanded its US operations in 2025 (Stanford HAI, 2026). A disruption in the Taiwan Strait would not simply be a regional security crisis, but could reverberate throughout the infrastructure underpinning global AI development.
The Power Behind the Cloud
The other constraint is more basic: electricity. AI may exist digitally, but it is becoming increasingly dependent on physical power systems. The International Energy Agency (IEA) estimates that data centres consumed around 415 terawatt-hours of electricity in 2024, approximately 1.5 per cent of global electricity consumption. Under its base case, this figure could more than double to around 945 TWh by 2030, with AI-focused computing accounting for a substantial share of the increase (IEA, 2025).
The geopolitical implications are significant because electricity is not equally available, affordable or reliable everywhere. AI infrastructure gravitates towards locations where governments can combine cheap or dependable power, land, connectivity, cooling and investment incentives. The United States and China already dominate global data-centre electricity consumption, accounting for around 45% and 25% respectively in 2024 (IEA, 2025). Meanwhile, countries such as the United Arab Emirates and Saudi Arabia are attempting to turn abundant energy and capital into advantages in the AI infrastructure race.
These conditions create an emerging paradox. Governments want to attract data centres because they promise investment, technological prestige and economic activity. Yet the infrastructure places enormous demands on national grids. The issue is already politically contentious in the United States: on 30 September the Senate blocked a bill that would have required state utility regulators to consider whether data centres and other large electricity users should bear the incremental costs of the power infrastructure built to serve them. The competition for AI leadership is increasingly becoming a competition for power generation itself.
The Global South Faces a Different Race
For developing economies, the problem is not necessarily how to win the frontier AI race. It is whether they can participate in it without becoming permanently dependent on foreign infrastructure.
The World Bank’s Digital Progress and Trends Report 2025 argues that the foundations of AI participation are connectivity, compute, context and competency: reliable internet and electricity, access to computing infrastructure, relevant data and appropriate skills (World Bank, 2025). The organisation’s 2026 World Development Report goes further, concluding that advancing the AI frontier, which relies on semiconductors, data centres, AI-ready training data and top-level research talent, is out of reach for most developing countries in the near term. It recommends instead that they adopt available tools, adapt them to local conditions and, over time, advance towards frontier capabilities (World Bank, 2026).
This problem results in a difficult policy choice. Importing computing power through foreign cloud providers can give countries affordable access to sophisticated AI without the enormous capital cost of constructing their own infrastructure. But excessive dependence on external providers can create strategic vulnerabilities over data, pricing, service continuity and technological autonomy. Building domestic capacity, meanwhile, can be extraordinarily expensive and risks producing infrastructure that is underutilised.
The danger is that an AI divide could reinforce existing economic inequalities. Countries without reliable electricity, high-speed connectivity, specialised skills or access to affordable compute may become consumers of AI designed elsewhere, while a small group of states and corporations capture a disproportionate share of the economic and strategic value.
From AI Race to Infrastructure Diplomacy
The next phase of AI competition will be determined by more than who develops the most impressive model. It will depend on who can secure chips, electricity, data, minerals, cloud capacity and the talent necessary to combine them. UN financial agency the International Monetary Fund (IMF) has highlighted how AI’s expanding electricity requirements are already creating new pressures on energy systems (IMF, 2025a; IMF, 2025b), and estimates that AI-related technology investment added about 0.5 percentage points to US GDP growth in 2025 (IMF, 2026).
For governments, this realisation should redefine of AI policy. Industrial strategy, energy security, semiconductor policy and digital infrastructure can no longer be treated as separate portfolios. They are becoming components of the same geopolitical equation.
The answer is not necessarily technological autarky. Attempting to manufacture every chip, build every data centre and develop every model domestically would be unrealistic for most states. The more practical objective is resilience: diversified supply chains, reliable energy, competitive cloud markets, interoperable systems and international arrangements that prevent access to critical AI infrastructure from becoming an instrument of coercion.
AI may ultimately transform economies and societies in ways that remain difficult to predict. But one reality is already becoming clear. The future of artificial intelligence will not be determined in the cloud alone. It will be determined in semiconductor fabs, power plants, data centres, mineral supply chains and undersea networks. In the emerging geopolitical order, whoever controls those foundations will have considerable influence over who gets to build, deploy and benefit from the intelligence of the future.
Bibliography
Bogmans, C., Gomez-Gonzalez, P., Ganpurev, G., Melina, G., Pescatori, A. and Thube, S. (2025) ‘Power Hungry: How AI Will Drive Energy Demand’. IMF Working Paper WP/25/81, 22 April. Washington, DC: International Monetary Fund. https://doi.org/10.5089/9798229007207.001
Ganpurev, G. and Pescatori, A. (2025) ‘AI Needs More Abundant Power Supplies to Keep Driving Economic Growth’. IMF Blog. https://www.imf.org/en/Blogs/Articles/2025/05/13/ai-needs-more-abundant-power-supplies-to-keep-driving-economic-growth
International Energy Agency (IEA) (2025) Energy and AI. World Energy Outlook Special Report. Paris: IEA. https://www.iea.org/reports/energy-and-ai
International Monetary Fund (IMF) (2026) ‘AI: Deployment and Disruption’. IMF Annual Report 2026. https://www.imf.org/annual-report/2026/in-focus/ai-deployment-and-disruption/
Reuters (2026a) ‘DeepSeek partners with Huawei to develop chip programming tools, reducing reliance on Nvidia’. https://www.investing.com/news/stock-market-news/deepseek-partners-with-huawei-to-develop-chip-programming-tools-reducing-reliance-on-nvidia-4924051
Shivakumar, S., Wessner, C. and Howell, T. (2026) ‘China’s Localization Drive in Semiconductors Gains Impetus from Allied Chip Export Controls’. Center for Strategic and International Studies. https://www.csis.org/analysis/chinas-localization-drive-semiconductors-gains-impetus-allied-chip-export-controls
Stanford Institute for Human-Centered Artificial Intelligence (Stanford HAI) (2026) The 2026 AI Index Report. Stanford University. https://hai.stanford.edu/ai-index/2026-ai-index-report
World Bank (2025) Digital Progress and Trends Report 2025: Strengthening AI Foundations. Washington, DC: World Bank. https://www.worldbank.org/en/publication/dptr2025-ai-foundations
World Bank (2026) World Development Report 2026: The Promise of Artificial Intelligence. Washington, DC: World Bank. https://doi.org/10.1596/978-1-4648-2331-2
