{ "title": "Inside the AI-Led Resource Race", "sourceUrl": "https://www.imf.org/en/publications/fandd/issues/2025/12/inside-the-ai-led-resource-race-thijs-van-de-graaf", "canonical": "https://www.imf.org/en/publications/fandd/issues/2025/12/inside-the-ai-led-resource-race-thijs-van-de-graaf", "overlayPath": "/en/publications/fandd/issues/2025/12/inside-the-ai-led-resource-race-thijs-van-de-graaf/index.md", "summary": "Material demands—for energy, chips, and minerals—will determine who dominates data", "sections": [ { "heading": "Overview", "content": "- Material demands—for energy, chips, and minerals—will determine which countries and companies dominate AI and data.\n- AI’s infrastructure includes servers, cooling systems, chips, and mined minerals; this physical backbone is rapidly expanding with data centers multiplying in number and size.\n- The contest in algorithms is increasingly a contest for energy, land, water, semiconductors, and minerals, shaping who can harness the AI revolution." }, { "heading": "Energy and data-center growth", "content": "- Data centers already use about 1.5 percent of global electricity supply.\n- The International Energy Agency (IEA) expects data center demand to more than double by 2030, with AI responsible for much of the increase.\n- AI accounts for less than a tenth of added power demand this decade globally, but national impacts can be large:\n - In the US and Japan, data centers could account for nearly half of new demand by 2030.\n - In Ireland, data centers already use more than a fifth of the country’s electricity.\n- Hyperscale centers have power needs in the tens of megawatts; new projects are moving to gigawatt scale, including a planned 5-gigawatt campus in Abu Dhabi that spans 10 square miles.\n- Localized strains:\n - In northern Virginia, data centers consume about one-quarter of the state’s power, affecting utilities’ ability to serve other customers.\n - Ireland’s grid operator froze new projects in 2022 except those that could generate their own power.\n - Singapore halted approvals in 2019 and now allows facilities only under strict efficiency rules." }, { "heading": "Big Tech and power markets", "content": "- Major technology companies are among the world’s biggest corporate buyers of renewable energy.\n - Microsoft, Amazon, and Google have each signed multibillion-dollar power purchase agreements.\n- Firms are experimenting with on-site generation and advanced technologies:\n - Microsoft has explored nuclear, including small modular reactors and possible acquisitions of mothballed plants.\n - Google is backing advanced geothermal.\n - Amazon is testing hydrogen for backup power.\n- Geographic concentration: the US is home to more than 40 percent of the world’s data centers.\n- Implications:\n - Big Tech’s capital could accelerate clean-energy innovation or cement dependence on fossil fuels.\n - Europe has seen AI boost renewables, while US demand still leans heavily on natural gas." }, { "heading": "Efficiency, software, and semiconductor constraints", "content": "- AI can both increase electricity demand and help manage it (balancing grids, forecasting renewables, optimizing energy use, piping waste heat into district heating).\n- Efficiency gains in hardware and software:\n - New chips such as Nvidia’s Blackwell processors and Google’s tensor processing units (TPUs) aim for more operations per watt.\n - China’s DeepSeek, released in January 2025, was trained at a fraction of the cost and energy of comparable models from OpenAI and Google.\n- Jevons paradox: greater efficiency may spark increased use and higher aggregate demand.\n- Semiconductors are a strategic choke point:\n - Training state-of-the-art models requires thousands of specialized chips, most designed by Nvidia and manufactured almost exclusively in Taiwan Province of China by the Taiwan Semiconductor Manufacturing Company (TSMC).\n - Geopolitical responses include US export controls on advanced chips and subsidies for domestic fabrication; China is building domestic capabilities; Europe, Japan, and India are investing in their industries." }, { "heading": "Mineral footprint and supply risks", "content": "- Fabrication and data centers require substantial minerals and water:\n - Components include gallium, germanium, silicon, rare earths, copper; a single hyperscale campus can contain nearly as much copper as a midsize mine produces in a year.\n- Projected 2030 mineral demand from data centers (IEA estimates cited):\n - More than half a million metric tons of copper each year.\n - 75,000 tons of silicon each year.\n - Data centers’ share of global demand could reach 2 percent for copper and silicon.\n - For gallium, data centers could account for more than a tenth of total demand.\n- Concentration and export controls:\n - China controls 80–90 percent of global refining of silicon, gallium, and rare earths.\n - In 2023 it restricted exports of gallium and germanium; since late 2024 new curbs followed on tungsten, tellurium, bismuth, indium, and molybdenum.\n- Responses by importing regions include recycling programs and alliances with resource-rich countries in Africa and Latin America.\n- Price and supply implications: prices for many of these metals have spiked." }, { "heading": "Land, water, and siting considerations", "content": "- Hyperscale data centers locate where cheap power, abundant water, and fast fiber-optic links converge. Land is rarely the limiting factor.\n- Water constraints:\n - Cooling demands millions of gallons a day.\n - Two-thirds of new US centers since 2022 have been built in water-stressed regions.\n - Local disputes have arisen (Arizona, Spain, Singapore) over competing claims on scarce water supplies.\n - Most of AI’s water footprint is indirect: power plants supplying data centers consume far more water than the centers themselves.\n- Latency and climate considerations influence siting:\n - Ireland’s cluster reflects its role as a transatlantic cable hub.\n - Abu Dhabi’s planned 5-gigawatt campus was chosen partly to minimize delays with Asia and Europe.\n - Colder countries (Norway, Iceland) offer climate advantages via reduced cooling needs." }, { "heading": "Policy challenges and implications", "content": "- Planning uncertainty:\n - Forecasts of data center demand for 2030 diverge widely; the highest published estimate is nearly seven times the lowest.\n - Governments must expand electricity systems quickly but face risks of overbuilding and locking in fossil fuels.\n- Transparency gap:\n - Little public reporting exists from the industry on data center use of electricity, water, or minerals.\n - Greater disclosure would help regulators, utilities, and communities plan.\n- Sustainability and equity:\n - Expanding grids and supply chains without safeguards risks repeating boom-and-bust cycles of past commodity races.\n - Benefits could be tilted toward the rich world if developing economies remain suppliers of raw materials and face higher implied costs for energy and capital.\n- Potential outcomes:\n - Managed well: the AI boom could accelerate clean energy and foster more resilient supply chains.\n - Managed poorly: the AI boom risks locking in new emissions and deepening resource dependence." }, { "heading": "Key findings and policy recommendations", "content": "- Key factual findings:\n - Data centers use about 1.5 percent of global electricity.\n - The largest hyperscale projects have power needs in the tens of megawatts; new projects are moving to gigawatt scale (example: a planned 5-gigawatt campus covering 10 square miles).\n - The US hosts more than 40 percent of the world’s data centers.\n - By 2030, data centers could consume more than half a million metric tons of copper and 75,000 tons of silicon each year; data centers’ share of global demand could reach 2 percent.\n - China controls 80–90 percent of global refining for several critical materials.\n - Two-thirds of new US centers since 2022 have been built in water-stressed regions.\n- Policy recommendations implied by the analysis:\n - Treat power plants, grids, water, and minerals as integral parts of digital/AI policy.\n - Expand and modernize electricity systems rapidly while avoiding lock-in of fossil-fuel infrastructure.\n - Increase industry transparency on electricity, water, and mineral usage to aid planning and regulatory oversight.\n - Implement environmental and social safeguards across energy and mineral supply chains to avoid boom-and-bust dynamics.\n - Promote equitable outcomes so developing economies do not become solely raw-material suppliers while bearing higher costs.\n\nThijs van de Graaf, associate professor of international politics at Ghent University and energy fellow at the Brussels Institute for Geopolitics; December 2025.\n\n---\n\n Content in this bundle\n\n- Inside the Ai-Led Resource Race\n - Inside the Ai-Led Resource Race (Markdown version){rel=\"alternate\" type=\"text/markdown\"}\n - Inside the Ai-Led Resource Race (PDF){rel=\"external\" type=\"application/pdf\"}\n\n---\n\nSource: https://www.imf.org/en/publications/fandd/issues/2025/12/inside-the-ai-led-resource-race-thijs-van-de-graaf" } ], "bullets": [ "[Markdown version](/en/publications/fandd/issues/2025/12/inside-the-ai-led-resource-race-thijs-van-de-graaf/index.md)", "[Structured JSON version](/en/publications/fandd/issues/2025/12/inside-the-ai-led-resource-race-thijs-van-de-graaf/index.json)", "[Bundle manifest](/en/publications/fandd/issues/2025/12/inside-the-ai-led-resource-race-thijs-van-de-graaf/bundle-manifest.json)", "Authors: THIJS VAN DE GRAAF", "Published: November 17, 2025", "Material demands—for energy, chips, and minerals—will determine which countries and companies dominate AI and data.", "AI’s infrastructure includes servers, cooling systems, chips, and mined minerals; this physical backbone is rapidly expanding with data centers multiplying in number and size.", "The contest in algorithms is increasingly a contest for energy, land, water, semiconductors, and minerals, shaping who can harness the AI revolution.", "Data centers already use about 1.5 percent of global electricity supply.", "The International Energy Agency (IEA) expects data center demand to more than double by 2030, with AI responsible for much of the increase.", "AI accounts for less than a tenth of added power demand this decade globally, but national impacts can be large:", "Hyperscale centers have power needs in the tens of megawatts; new projects are moving to gigawatt scale, including a planned 5-gigawatt campus in Abu Dhabi that spans 10 square miles.", "Localized strains:", "Major technology companies are among the world’s biggest corporate buyers of renewable energy.", "Firms are experimenting with on-site generation and advanced technologies:", "Geographic concentration: the US is home to more than 40 percent of the world’s data centers.", "Implications:", "AI can both increase electricity demand and help manage it (balancing grids, forecasting renewables, optimizing energy use, piping waste heat into district heating).", "Efficiency gains in hardware and software:", "Jevons paradox: greater efficiency may spark increased use and higher aggregate demand.", "Semiconductors are a strategic choke point:", "Fabrication and data centers require substantial minerals and water:", "Projected 2030 mineral demand from data centers (IEA estimates cited):", "Concentration and export controls:", "Responses by importing regions include recycling programs and alliances with resource-rich countries in Africa and Latin America.", "Price and supply implications: prices for many of these metals have spiked.", "Hyperscale data centers locate where cheap power, abundant water, and fast fiber-optic links converge. Land is rarely the limiting factor.", "Water constraints:", "Latency and climate considerations influence siting:", "Planning uncertainty:", "Transparency gap:", "Sustainability and equity:", "Potential outcomes:", "Key factual findings:", "Policy recommendations implied by the analysis:", "**Inside the Ai-Led Resource Race**" ], "related": [ { "title": "Inside the Ai-Led Resource Race", "role": "document", "sourceUrl": "https://www.imf.org/-/media/files/publications/fandd/article/2025/12/van-de-graaf.pdf", "summary": { "path": "/-/media/files/publications/fandd/article/2025/12/van-de-graaf.pdf.md", "mime": "text/markdown" }, "binary": { "path": "/-/media/files/publications/fandd/article/2025/12/van-de-graaf.pdf", "mime": "application/pdf" } } ], "alternates": { "markdown": "/en/publications/fandd/issues/2025/12/inside-the-ai-led-resource-race-thijs-van-de-graaf/index.md", "json": "/en/publications/fandd/issues/2025/12/inside-the-ai-led-resource-race-thijs-van-de-graaf/index.json", "bundleManifest": "/en/publications/fandd/issues/2025/12/inside-the-ai-led-resource-race-thijs-van-de-graaf/bundle-manifest.json" }, "generatedAtUtc": "2026-09-15T20:12:02.158Z" }