Key takeaways
- Data centres are not just energy intensive; they are mineral intensive, requiring 60–75 tonnes of metals per megawatt (MW) of capacity as a significant driver of mining and refining activity.
- The United States, home to 44% of global data-centre buildout is deeply import-dependent, with China, Canada, Belgium, Germany, and Mexico dominating mineral exposure.
- Critical artificial intelligence-enabling minerals (gallium, germanium, tantalum) face structural supply constraints that mining timelines cannot easily resolve.
- Copper emerges as the system-level bottleneck, with data centres contributing materially to a projected global shortfall of 10 million metric tonnes per annum (Mmtpa) by 2040.
- The cloud’s physical foundation exposes geopolitical risk, linking digital growth directly to mining, refining, and trade policy.
Introduction: the cloud has a physical cost
Data centres run the “cloud” of the networked world. These are facilities with powerful computer servers responsible for processing and crunching our online data. These capabilities provide services which we have come to take for granted such as online gaming, secure banking transactions, smart home services, and other digital services, all of which give substance to our digital world.
As the adoption and demand of AI-powered services accelerates, data centres – which serve as the physical backbone of this digital world – are projected to grow to meet the expected demand. The number of data centres is expected to more than double from its current number of 12,000 to 28,000 by 2030.
While the impact of this growth is usually assessed from the perspective of the demand on energy and environmental resources (land, water), there is another angle to consider – the minerals/metals impact of this growth. The electrical/electronic circuitry that underpin the computer servers, power supply, and automation components of data centres are made from mineral elements.
As we shared in our article on Why the seventeen rare earth elements matter for energy, tech and security, 13 of the 17 REEs are found in electronics equipment. There are other elements that are not classed as REE but play critical roles in the electronics design of these facilities, including silver (Ag), gold (Au), platinum (Pt), and copper (Cu).
Given the expanded role that data centres are expected to play, even while acknowledging warnings about the risk of an AI bubble burst from economically feeble AI-business models, this article gives us a glimpse into the mineral foot print of data centres.
Anatomy of a data centre
The schematic of a data centre shown in Figure 1, captures the core physical infrastructure elements of its electrical distribution system. This system, which is foundational to the operation of the centre, supports the centre’s hardware and software. These include power subsystems, uninterruptible power supplies (UPS), ventilation, cooling systems, fire suppression, backup generators, and connections to external networks.
Figure 1 – 3D schematic illustration of electrical distribution system in a data centre. Source: https://green-data.blogspot.com/2020/03/electrical-distribution-system-data-.html
Typically, a data centre consists of three main areas:
- Server room: This area houses the racks and server cabinets, storage systems, and networking equipment including routers, switches, firewalls, storage systems, servers, and application delivery controllers.
- Power room: Here the electrical supply infrastructure is housed and managed – the uninterruptible power supply (UPS), backup generators, and electrical power distribution units (PDU).
- Network operations centre (NOC): The area from which the centre is monitored and controlled to ensure stable, 24/7 operation of IT devices.
A survey of related literature indicates a cost range of US$6 million/MW to US$15 million/MW to build a data centre, but can rise to US$40 million/MW for an AI-application data centre. Electrical system and cooling account for 60%+ of the total construction costs. Operating costs are driven by maintenance (~40%), electricity (~15-25%), labour, water, general/administration, and others.
The global data centre landscape
Based on data from Synergy Research Group, the United States had 5,427 data centres, as of November 2025, making it the world’s largest concentration of data centres. Out of approximately 12,000 operational data centres globally, the U.S. led the share at 45%, followed by Germany at 4.4%, the United Kingdom at 4.3%, then China at 3.7% as per Table 1.
| List of top countries by data centre count | |||
| Rank | Country | Number of data centres | Global share (%) |
| 1 | United States | 5,427 | 44.7% |
| 2 | Germany | 529 | 4.4% |
| 3 | United Kingdom | 523 | 4.3% |
| 4 | China | 449 | 3.7% |
| 5 | Canada | 337 | 2.8% |
| 6 | France | 322 | 2.7% |
| 7 | Australia | 314 | 2.6% |
| 8 | Netherlands | 298 | 2.5% |
| 9 | Russia | 251 | 2.1% |
| 10 | Japan | 222 | 1.8% |
| Total top 10 | 8,672 | 72.3% | |
| Global total | 12,148 | ||
Table 1 – The United States leads the top 10 countries by data centre count (Nov 2025).
The top 10 countries, by the number of data centres, account for nearly three-quarters of the global total.
In power capacity terms, global data centre capacity reached 122.2 gigawatts (GW) of installed IT power capacity as of Q1 2025, according to Synergy Research Group. Once again, the U.S. leads with 53.7 GW capacity (44% of global total), followed by China with 31.9 GW (26%), and European Union (EU) countries with 11.9 GW of capacity (9.7%). These three regions represent 80% of global data centre power capacity.
Figure 2 – The U.S. accounts for 44% of global data centre power. Source: Visualcapitalist
Expressed in per-unit GW capacity, the global average data centre power capacity is estimated at 10.1 MW/unit. Regional variation stretches from 3.05 MW/unit in Brazil to 71 MW/unit in China, estimated by cross-referencing the regional number of data centres against their corresponding IT power capacity.
| Global IT power capacity of data centres | ||
| Region | IT power capacity (GW) | % of global total |
| U.S. | 53.7 | 43.9% |
| China | 31.9 | 26.1% |
| EU | 11.9 | 9.7% |
| Japan and Korea | 6.6 | 5.4% |
| India | 3.6 | 2.9% |
| Other Asia Pacific | 3.1 | 2.5% |
| UK | 2.6 | 2.1% |
| Australia and New Zealand | 1.6 | 1.3% |
| Other North America | 1.5 | 1.2% |
| Africa | 1.5 | 1.2% |
| Other Central and South America | 1.4 | 1.1% |
| Eurasia | 1.2 | 1.0% |
| Middle East | 1.1 | 0.9% |
| Brazil | 0.6 | 0.5% |
| Global total | 122.2 GW | 100% |
Table 2 – Global data centre IT power capacity by region. Source: International Energy Agency
According to International Energy Agency (IEA) projections, data centre electricity demand will more than double to 945 terawatt-hours (TWh) globally by 2030, approaching 3% of total global electricity consumption. According to the regional distribution of the data electricity demand shown in Table 3, China’s data centre electricity demand will increase most aggressively at 170%, exceeding U.S. growth of 133%.
| Change in data centre power consumption by region | |||
| Region | 2024 (TWh) | 2030 (TWh) | Growth |
| United States | 183 | 426 | +133% |
| China | 65 | 175 | +170% |
| Europe | 64 | 109 | +70% |
| Japan | 19 | 34 | +80% |
| Rest of world | 84 | 201 | +140% |
| Global total | 415 | 945 | +128% |
Table 3 – China leads growth in projected data centre electricity consumption by 2030 vs 2024. Source: IEA
Europe’s data centre power demand is expected to grow by 70%, while Japan’s growth is forecast to be 80% through 2030. Globally, data centre electricity demand is expected to rise by 128% over the next half-decade. As of 2024, data centres consumed 415 TWh of electricity globally, which is 1.5% of global electricity demand. For context, this is more than what the UK consumed (319 TWh) in a year, slightly lower than what France consumed (450 TWh), and 20% less than Canada’s consumption (503 TWh).
Going by these growth rates, if we assumed the same power efficiency and data centre capacity utilization levels, this would imply that the number of data centres would increase from ~12,000 currently to ~28,000 over this period with a capacity of 278 GW.
Deconstructing the server – a periodic table of critical minerals
Demand for data centres is indeed rising sharply, expected to more than double over the next five years, largely to support new and computationally intense AI technologies. Data centres consume large amounts of energy (415 TWh in 2024), require vast amounts for cooling (70,000 litres/day/MW), and consume significant mineral resources to build (>60 tonnes/MW).
The chart from the United States Geological Survey (USGS) below illustrates minerals found in data centre infrastructure.
Figure 3 – Minerals used in data centres and U.S. net import dependence. Source: USGS and USGS Mineral Commodity Summaries 2025
As Figure 3 is intended to convey U.S. critical minerals exposure to foreign supply, it includes the key minerals used in each part of a data centre, and the percentage the U.S. imports to meet consumption demand for that mineral.
The U.S. relies on imports from other countries for many of these minerals. Looking over the graphic, net import dependence ranges from 45% for the metal copper to 100% for metalloid arsenic and metal gallium.
To facilitate the examination of mineral intensity, we look at the minerals contained in the key components of the data centre infrastructure, the U.S. net import as of 2024, source countries, and an indication of the criticality of the mineral as per the U.S.
Server boards and circuitry
Server boards connect the essential electrical components of a server. The intricate circuitry found here requires minerals that are able to efficiently conduct electricity and are resistant to corrosion – especially copper. Other metals found in server boards are silver, gold, and platinum.
Table 4 details the critical minerals used in this application.
| Minerals/metals found in data centre server boards and circuitry | |||||
| Key components | Critical minerals used | Mineral intensity (kg/MW) | U.S. net import reliance (%) – 2024 | Source countries1 | Ranking on criticality table |
| Server boards and circuitry | Silver | 5 | 64% | Mexico, Canada, Republic of Korea, Poland | 37 |
| Gold | 1 | 0% | 84 | ||
| Copper, refined | 2,000 | 45% | Chile, Canada, Mexico, Peru | 29 | |
| Tin | 120 | 73% | Peru, Bolivia, Indonesia, Brazil | 39 | |
| Tantalum | NA | 100% | China, Australia, Germany, Indonesia | 45 | |
| Palladium | NA | 36% | Russia, South Africa, Belgium, Italy | 24 | |
| Platinum | NA | 85% | South Africa, Belgium, Germany, Italy | 30 | |
| Boron | NA | 0% | NA | ||
| Rare earth elements | 2 | 80% | China, Malaysia, Japan, Estonia | ||
| 1: Top 4 import sources (2020–23) in descending order of import share. Source: USGS Mineral Commodity Survey [net import, source countries and criticality ranking], World Economic Forum – From Minerals to Megawatts [Mineral Intensity] NA = not available | |||||
Table 4 – Critical minerals found in data centre server boards and circuitry. Source: USGS
For server boards and circuitry, U.S. import dependence on the materials range from 0% for boron and gold, to 80% for rare earth elements, and 100% for tantalum.
Heatsinks – metals that move heat
Heatsinks are designed to conduct heat away from sensitive electronics and thus prevent servers from overheating. The metals chosen for this task are highly conductive, malleable, and resistant to corrosion. Aluminum (Al) and copper (Cu) fit the bill here.
| Minerals/metals found in data centre heatsinks | |||||
| Key components | Critical minerals used | Mineral intensity (kg/MW) | U.S. net import reliance (%) – 2024 | Source countries1 | Ranking on criticality table |
| Heatsinks | Aluminum | 6,700 | 47% | Canada, United Arab Emirates, Bahrain, China | 30 |
| Copper, refined | 3,600 | 45% | Chile, Canada, Mexico, Peru | 29 | |
| Iron | 18,000 | 87% | China, Germany, Brazil, Canada | 80 | |
| 1: Top 4 import sources (2020–23) in descending order of import share. Source: USGS Mineral Commodity Survey [net import, source countries and criticality ranking], World Economic Forum – From Minerals to Megawatts [Mineral Intensity] NA = not available | |||||
Table 5 – The U.S. is 45%+ reliant on aluminum and copper imports, which are sourced from Canada, UAE, China, and Mexico. Source: USGS
Semiconductors and microchips
The computational heft of data centres comes down to microchips and processors. Semiconductor materials are at the core of their functioning. Materials such as silicon (Si) and germanium (Ge) that conduct electricity under specific conditions are found here. Others include arsenic (As), fluorspar (CaF2), gallium (Ga), and platinum (Pt).
| Minerals/metals found in data centre semiconductor and microchips | |||||
| Key components | Critical minerals used | Mineral intensity (kg/MW) | U.S. net import reliance (%) – 2024 | Source countries1 | Ranking on criticality table |
| Semiconductors and microchips | Arsenic | NA | 100% | China, Morocco, Malaysia, Belgium | 61 |
| Fluorspar | NA | 100% | Mexico, Vietnam, South Africa, China | 47 | |
| Gallium | 2 | 100% | Japan, China, Germany, Canada | 6 | |
| Germanium | 0.3 | 100% | Belgium, Canada, China, Germany | 7 | |
| Indium | NA | 100% | Republic of Korea, Japan, Canada, Belgium | 22 | |
| Palladium | NA | 36% | Russia, South Africa, Belgium, Italy | 24 | |
| Platinum | NA | 85% | South Africa, Belgium, Germany, Italy | 30 | |
| Silicon | 160 | <50% | Brazil, Russia, Canada, Malaysia | 18 | |
| Tantalum | NA | 100% | China, Australia, Germany, Indonesia | 45 | |
| 1: Top 4 import sources (2020–23) in descending order of import share. Source: USGS Mineral Commodity Survey [net import, source countries and criticality ranking], World Economic Forum – From Minerals to Megawatts [Mineral Intensity] NA = not available | |||||
Table 6 – the U.S. is highly exposed on semiconductor elements used in development of microchips. Source: USGS
The U.S. is at least 85% import dependent on all metals found in microchips and processers other than silicon (<50%) and palladium (36%).
Magnets and data storage
Specialized magnetic materials are employed in the hard disk and solid-state drives to store and retrieve data. Barite (BaSO4), boron (B), and rare earth elements typically find application here.
| Minerals/metals found in data centre magnets and data storage | |||||
| Key components | Critical minerals used | Mineral intensity (kg/MW) | U.S. net import reliance (%) – 2024 | Source countries1 | Ranking on criticality table |
| Magnets and data storage | Barite | NA | >75% | India, China, Morocco, Mexico | 20 |
| Boron | NA | 0% | NA | ||
| Rare earth elements | NA | 80% | China, Malaysia, Japan, Estonia | ||
| 1: Top 4 import sources (2020–23) in descending order of import share. Source: USGS Mineral Commodity Survey [net import, source countries and criticality ranking], World Economic Forum – From Minerals to Megawatts [Mineral Intensity] NA = not available | |||||
Table 7 – U.S. import dependence on barite and REE, which are applied in magnets and data storage.
Other than boron, the U.S. is at least 75% import dependent on the key elements that go into the manufacturing of magnets and data storage for data centres.
Admittedly, these tables are U.S.-centric in that they focus on data centre minerals/metals for which the U.S. is import dependent and the top countries from where those metals are sourced by the US. However, the tables reveal the wide spectrum of minerals embedded within data centres more generally, and the intensity of the minerals (expressed as embedded mass for every 1MW of data centre capacity).
Overall, each megawatt of data centre capacity embeds 60-75 tonnes of minerals and metals, mainly in power and cooling systems, rather than servers. By 2030, the number of data centres is expected to be 28,000, with an estimated installed capacity of 278 GW.
Mapping U.S. mineral exposure by country
Extracting and ranking the countries to which the U.S. has the most exposure results in the Table 8 that proxies exposure by country across the data centre mineral supply chain. The top countries that lead as U.S. import sources for data centre minerals are China, Canada, Belgium, Germany, and Mexico/South Africa.
| U.S. exposure to data centre mineral supply chain | ||
| Rank | Country | Frequency |
| 1 | China | 9 |
| 2 | Canada | 8 |
| 3 | Belgium | 7 |
| 4 | Germany | 6 |
| 5 | Mexico | 5 |
| 5 | South Africa | 5 |
| 7 | Italy | 4 |
| 8 | Peru | 3 |
| 8 | Indonesia | 3 |
| 8 | Russia | 3 |
| 8 | Malaysia | 3 |
| 8 | Japan | 3 |
| 13 | Republic of Korea | 2 |
| 13 | Chile | 2 |
| 13 | Brazil | 2 |
| 13 | Australia | 2 |
| 13 | Morocco | 2 |
| 18 | Poland | 1 |
| 18 | Bolivia | 1 |
| 18 | UAE | 1 |
| 18 | Bahrain | 1 |
| 18 | Vietnam | 1 |
| 18 | India | 1 |
| 18 | Estonia | 1 |
Table 8 – Proxy for U. S. exposure concentration by country across the data centre mineral supply chain.
China ranks number 1 as it appears more frequently than any other country, reinforcing that the underlying U.S. data centre infrastructure is deeply entangled with Chinese mineral and material flows, even where final assembly occurs elsewhere.
Canada’s near parity with China in frequency, as well as Mexico’s presence in the top 5 underscore these countries as the most strategically important allied upstream suppliers. This should ordinarily position both Canada and Mexico at the heart of U.S. critical-minerals policy, realigned supply chains, and “friend-shoring” proposals.
While data centres are often described as “digital infrastructure”, their physical foundations reveal a tightly coupled, geopolitically exposed mineral supply chain – with China, Canada, and a small group of processing hubs accounting for a disproportionate share of U.S. exposure.
From mine to machine – the data centre supply chain
Tables 4 to 7 point us to the minerals and metals that underlie the equipment that goes into the data centres. The equipment itself is manufactured as part of a long, complex, and interdependent supply chain that stretches from miners to refiners, component makers, and end-product owners.
Servers and processing units are a significant subset of equipment deployed in data centres. In 2023, trade in servers and processing units (classified as HS code 8471) totaled $403 billion, $271 billion (67% ) of which were exports from Asia, led by China ($177 billion), Taiwan ($28.8 billion), and Vietnam ($17.7 billion). On the import side, United States led imports at $104 billion, followed by Germany at $27.7 billion, and Hong-Kong at $24.3 billion. Of the U.S.’s $104 billion in imports, ~40% was from China and ~14% from Taiwan.
Moving upstream, we find that according to S&P Global modelling, copper, with demand currently at 28 MMtpa, is the most critical mineral in data centre infrastructure. Its demand is projected to increase by 14 MMtpa to 2040, 10% of which is anticipated to be due to data centre expansion. However, only 32 MMtpa can be supplied on our current trajectory by 2040, thus leaving a shortfall of 10 MMtpa unless key bottlenecks such as slow mine permitting and restrained investment are addressed.
Further, lesser-known elements such as gallium, germanium, indium, palladium, and tantalum enable the high-performance microchips at the heart of every graphics processing unit (GPU) and data centre (see Table 6). Given its role in AI training, germanium demand is projected to increase by 37% and gallium demand by 85% by 2033. However, the expectation is that growth in supply will be constrained, potentially leading to shortfalls and higher prices. Consider that China, which produces 68% of the world’s refined germanium, is projected to encounter shortages by 2040.
Conclusion – the road to resource resilience
The rapid expansion of data centres globally is often framed as a challenge of electricity supply, land use, or cooling efficiency. This analysis shows it is equally a mining and minerals challenge.
As the growth in global data centres is forecast to more than double between now and 2030, the need to build these facilities to meet increasing computing needs will weigh on upstream minerals mining. The high levels of import dependence and mineral intensity of data centre infrastructure have prompted the need to reconfigure supply chains, and fast-track development of manufacturing hubs.
Even as China has come to embody the definition of concentration risk due to its established dominance in midstream metals refining, it is also vulnerable to shortages in germanium – a critical semiconductor for chip manufacture.
The cloud may appear virtual, but its growth is ultimately governed by the realities of geology, capital intensity, and time. In that sense, the future of digital services will be shaped as much by mining economics as by algorithms.
(Kaase Gbakon, BIG Media Ltd., 2026)
