Resource Intelligence · June 2026
The Silent Wars Powering AI
Water, energy, and critical minerals are becoming geopolitical battlegrounds. Here is what the data shows — and what every organisation relying on AI infrastructure needs to understand before the conflict reaches their supply chain.
TechConsult Editorial Team18 June 2026Updated with Q2 2026 UN & IEA data
In February 2026, more than fifty countries gathered at the U.S. State Department for the first-ever Critical Minerals Ministerial. The official reason was trade coordination. The real reason was panic. Governments on every continent had arrived at the same conclusion simultaneously: the raw materials powering artificial intelligence are dangerously concentrated in a handful of hands, and the scramble to secure them is already reshaping alliances, fuelling armed conflict, and drying up rivers.
AI is commonly discussed as a digital phenomenon — invisible, weightless, cloud-based. That framing obscures a harder physical reality. Every large language model trained, every inference query answered, every data centre humming at full load is a transaction in three scarce commodities: water, electricity, and the minerals that make the hardware run. As AI scales, so does its appetite for all three. And as the appetite grows, so does the potential for conflict — local, national, and geopolitical — over who gets access and who is left without.
This article examines all three resource categories, traces the real-world friction points already emerging, and draws out the strategic implications for businesses and governments that are still treating AI sustainability as a PR exercise rather than a supply-chain and security priority.
Resource I
Water: The Invisible Cost of Every Query
Data centres have two water problems. The first is direct: evaporative cooling systems consume millions of gallons of freshwater daily to keep servers from overheating. A large modern data centre can consume up to 5 million gallons per day — roughly equivalent to the daily domestic water use of a town of 50,000 people. The second problem is indirect: the thermoelectric power plants that feed those data centres with electricity use water too, typically at a rate 3 to 4 times higher than on-site cooling. Every kilowatt-hour carries a hidden water cost that almost never appears in corporate sustainability reports.
1.3BThe number of people in Sub-Saharan Africa whose entire annual domestic water needs could be met by the water AI data centres are projected to consume by 2030.Source: UN University Institute for Water, Environment and Health (UNU-INWEH), June 2026
These are not theoretical projections. The footprint is already visible in court records, municipal water reports, and community protests across three continents.
Loudoun County, Virginia
Loudoun County is home to the world’s highest concentration of data centres — approximately 200 operational facilities. Between 2019 and 2023, data centre potable water use in the county grew by more than 250%, reaching 1.6 billion gallons in 2023 and approaching 10% of all county water consumption. The county’s water authority has been forced to plan entirely new reclaimed-water infrastructure simply to serve data centre cooling demand, yet as of 2026, most facilities continue drawing from treated drinking water supplies shared with residential users.
Querétaro, Mexico
Plans for fast-tracked data centre development in Querétaro, one of Mexico’s most water-stressed states, have placed the region’s agricultural economy under acute pressure. Prolonged drought has already stretched local aquifers; the prospect of industrial water extraction at data-centre scale has generated organised opposition from farming communities and triggered regulatory intervention by the Mexican government. The situation illustrates a pattern seen across the Global South: technology infrastructure built to serve global AI demand is sited in regions that bear the environmental cost while receiving little of the economic benefit.
Castilla-La Mancha, Spain
Meta’s proposed €1 billion data centre in the region of Castilla-La Mancha, expected to consume approximately 665 million litres of water annually, sparked direct confrontations with local farmers already managing under water rationing. The conflict became a national political issue in Spain and contributed to a broader European debate about siting regulation for AI infrastructure.
Real-World Indicator
The Oregon Transparency Battle
Google operates a major data centre cluster in The Dalles, Oregon, drawing from Columbia River basin water. When The Oregonian newspaper requested disclosure of Google’s water consumption figures, the City of The Dalles initially refused, citing trade-secret protections. Multi-year litigation was required before the data was released — and it showed water use had grown substantially over the period of secrecy. The case prompted Oregon’s 2024 legislation mandating public disclosure for data centres above a defined size threshold. It was one of the first instances in which a data centre’s environmental impact became a matter of contested public law — a precedent that will repeat itself globally.
Major operators are responding with genuine efficiency gains. Amazon disclosed for the first time in June 2026 that its global data centres consumed 2.5 billion gallons of water in 2025, claiming a water usage effectiveness of 0.12 litres per kilowatt-hour — a figure it characterises as seven times more efficient than the industry average. Microsoft achieved a 39% improvement in its water usage effectiveness between 2021 and 2025, and deployed zero-evaporation cooling technology at scale. Google reported that it replenished 64% of its freshwater consumption in 2024, up from 18% in 2023, as part of its water-positive commitment for 2030.
These are meaningful improvements. They are also almost certainly insufficient to offset the sheer volume of new capacity being built. Community opposition to data centre projects between March and June 2025 alone blocked or delayed an estimated $98 billion in development — a figure that reveals both the scale of planned construction and the depth of local resistance it is generating.
“What surprised us most is how often the choices that look greenest from a carbon perspective end up worse for water or for land.”— Dr. Miriam Aczel, Lead Author, UNU-INWEH Report on Environmental Cost of AI’s Energy Use, June 2026
The strategic implication: “low-carbon” AI infrastructure is not automatically “low-water” AI infrastructure. Switching data centre power sources from coal to bioenergy can, on average, cut the carbon footprint of electricity by 70% while simultaneously increasing its water footprint more than thirty-fold. Organisations benchmarking their AI sustainability against carbon metrics alone are measuring the wrong variable in the wrong unit.
Resource II
Energy: When AI Becomes a National Grid Problem
Electricity is the metabolic fuel of AI. Training a single frontier model now requires computational runs of a scale that would have been the entire output of a national research computing network a decade ago. Inference — the ongoing cost of answering queries — is, in aggregate, more significant than training and is growing continuously as AI is embedded into more applications and workflows.
448 TWhElectricity consumed by global data centres in 2025 — more than Saudi Arabia’s entire national consumption
945 TWhProjected data centre electricity demand by 2030, nearly triple the combined use of Pakistan, Bangladesh and Nigeria
+88%Projected increase in US data centre energy demand between 2025 and 2028 — equivalent to adding a second Spain
At the national level, Ireland offers the clearest early warning. Data centres accounted for 21% of all metered electricity in the country in 2023, exceeding the electricity use of every urban household combined. Ireland’s grid operator has since paused new data centre connections around Dublin until 2028. A small country’s grid hit its ceiling before the AI buildout had even reached full speed. The same dynamic, at different timescales, applies to any nation that becomes a preferred location for data centre investment without parallel investment in generation capacity.
The Grid-Priority Question
When electricity supply is constrained, allocation becomes political. Industrial users — including data centres, which typically hold long-term reservation agreements — are often structurally protected from rationing in ways that residential users are not. This asymmetry will generate friction in democratic societies where citizens increasingly understand that AI infrastructure is competing with their household power supply, hospital grid reliability, and manufacturing competitiveness.
The United States saw a preview of this conflict in 2025 and early 2026, when a coalition of tech executives signed a White House-sponsored pledge aimed at limiting the effect of data centre growth on residential electricity bills — an acknowledgement, implicit in the pledge itself, that the conflict was real enough to require a political response.
Strategic Scenario
The Grid Arbitrage Problem for Emerging Markets
As Western markets impose tighter siting restrictions and grid-capacity constraints on new data centres, operators will increasingly look to emerging markets — Southeast Asia, East Africa, the Gulf — where land is cheaper, regulations are looser, and political incentives for large foreign investment remain strong. The risk is a form of infrastructure colonialism: host nations accept the jobs and FDI headline, but absorb the grid stress, water consumption, and environmental liability while the economic value of AI computation accrues elsewhere. Consultants advising governments on data centre policy in these regions should model the full utility burden — not just the investment figure — before advising their clients to sign.
Resource III
Critical Minerals: The New Oil, With All the Same Politics
If water is AI’s hidden consumption cost and energy is its operating cost, critical minerals are its capital cost — the physical substrate without which none of the hardware exists. Graphical processing units and the semiconductor architecture beneath them depend on cobalt, nickel, copper, lithium, tungsten, tantalum, rare earth elements, gallium, germanium, and graphite. The concentration of these resources in the ground, and the concentration of processing capacity above it, has created a geopolitical structure that analysts increasingly compare to the petroleum economy of the twentieth century.
“Much like oil defined the global economy of the twentieth century, critical minerals are becoming the foundation of economic growth, technological innovation and national security in the twenty-first.”— Mining South East Europe, June 2026
The China Dominance Factor
China understood the strategic importance of critical mineral processing decades before its rivals did, and built a structural advantage that remains very difficult to dislodge. As of 2026, China is projected to supply over 60% of refined lithium and cobalt, around 80% of battery-grade graphite and rare earth elements, and approximately 70% of battery-grade manganese through to 2035. In December 2024, Beijing deployed this leverage directly, restricting exports of gallium, germanium, and antimony — key minerals for semiconductor production — to the United States, and extending graphite export controls that had been introduced the year before. Operators seeking permits to ship these materials now face Chinese approval processes of indefinite duration.
The International Energy Agency noted in February 2025 that a wide range of export control measures had been announced in recent months, with China announcing further controls on tungsten, tellurium, bismuth, indium, and molybdenum. These are not trade spats. They are the opening moves of a resource competition that the United States and European Union are only beginning to respond to at scale.
The DRC: Where AI’s Supply Chain Meets Armed Conflict
The Democratic Republic of Congo holds roughly 70% of the world’s cobalt reserves, alongside vast deposits of coltan — the ore used to produce tantalum for semiconductors and aerospace components. A war is actively being fought in eastern Congo over who controls those reserves. In June 2025, a diplomatic agreement between the United States, Rwanda, and the DRC produced a peace framework — but its primary economic provision was securing preferential U.S. corporate access to Congolese tantalum, gold, cobalt, copper, and lithium. Critics noted that mechanisms for protecting affected communities remained weak, and that the deal structurally reproduced the same extractive dynamic that has defined foreign engagement with Congo for a century.
Live Conflict Indicator
Greenland and the Resource Geography of AI Supremacy
The U.S. government’s interest in Greenland — which escalated into active geopolitical pressure in 2025 and early 2026 — was substantially driven by the territory’s vast deposits of rare earth elements and other minerals critical to AI hardware and defence technology. The episode, which concluded with the United States extracting a deal granting American companies exclusive access to significant portions of Greenland’s mineral resources, represents the first clear case in which AI supply-chain logic directly shaped the territorial ambitions of a major power. It will not be the last.
Globally, demand for critical minerals is forecast to nearly double between 2025 and 2030, with demand potentially reaching four times current levels by 2040. The challenge is not that the minerals do not exist in sufficient quantity in the ground. The challenge is that the refining, processing, and manufacturing capacity that converts raw ore into usable material is overwhelmingly concentrated in a single country — one that has demonstrated a willingness to use that concentration as a coercive instrument.
Strategic Outlook
What This Means for Organisations and Policymakers
The three resource vectors — water, energy, minerals — are not independent. They interact. Shifting AI infrastructure to lower-carbon power sources may dramatically increase water consumption at the power generation stage. Sourcing hardware from more politically stable mineral suppliers may increase energy intensity in processing. Optimising for one variable shifts the burden onto another, often in a different geography — and usually onto populations with less political power to resist it.
The UN University’s June 2026 report was explicit: evaluating AI sustainability through any single metric hides trade-offs and redistributes environmental burdens onto regions already facing stress. Any organisation, government, or consultant that is benchmarking AI sustainability against carbon alone is not conducting sustainability analysis. It is conducting carbon analysis and calling it something else.
For technology consultants advising on AI strategy, infrastructure, or procurement, the practical implications are direct. Client organisations need to understand the full resource profile of their AI infrastructure — not just the energy efficiency ratings of their cloud providers, but the water stress index of the regions where their compute runs, the mineral provenance of the hardware in their supply chains, and the regulatory trajectory in the jurisdictions where their data centres sit. These are no longer environmental niceties. They are operational and reputational risks with measurable financial exposure.
The $98 billion in data centre projects blocked or delayed by community opposition between March and June 2025 alone suggests the cost of ignoring local resource dynamics is not theoretical. Major operators including Alphabet, Amazon, Meta, Microsoft, Alibaba, and Tencent have all faced allegations involving community opposition over water and power use, according to MSCI Controversies data as of November 2025. The reputational and regulatory cost of these conflicts will only increase as public awareness of the resource footprint of AI grows.
The geopolitical layer is more difficult to hedge but equally important to model. Organisations whose AI hardware supply chains run through China-dominated processing capacity are exposed to the same coercive leverage that Beijing has already demonstrated against the United States. Diversification of supply is expensive and slow — but the cost of supply disruption at scale is higher. The same logic that drove petroleum supply-chain diversification after the 1973 oil shock applies to critical minerals in 2026. The organisations that begin that work now will be materially better positioned than those that begin it after the next export restriction announcement.
Conclusion
The Resource Bill Is Coming Due
AI is not a weightless technology. It runs on water drawn from rivers and aquifers. It runs on electricity generated by burning fuel or moving water through turbines. It runs on minerals extracted from soil in countries whose populations bear the environmental cost while wealthy nations capture the economic value. These are not edge cases or negative externalities to be managed around the margins. They are the physical foundation on which every model, every application, and every competitive advantage built on AI ultimately rests.
The organisations and governments that treat resource security as integral to AI strategy — rather than as an afterthought — will be better positioned for the decade ahead. Those that do not will find that the resource wars powering AI are not a problem for someone else’s continent. They are a problem for every procurement decision, every infrastructure investment, and every policy choice made about where and how AI runs.
The warning signs are already in the court records, the water authority reports, the conflict zones, and the export restriction announcements. The data is public. The question is whether the organisations with the most to lose are reading it.
References & Sources
[1]
UN University Institute for Water, Environment and Health (UNU-INWEH). Environmental Cost of AI’s Energy Use: Carbon, Water and Land Footprints. June 3, 2026.
unu.edu/inweh
[2]
TIME Magazine. “AI Could Use as Much Water as 1.3 Billion People by 2030, U.N. Report Warns.” June 2026.
time.com
[3]
Axis Intelligence. “AI Data Center Water Usage Statistics 2026: 264 Billion Gallons and Rising.” June 2026.
axis-intelligence.com
[4]
WaterVerge. “Data Centers Are Drinking Your Water: How AI Cooling Strains Local Drinking Supplies.” May 2026.
waterverge.com
[5]
MSCI Sustainability & Climate. “When AI Meets Water Scarcity: Data Centers in a Thirsty World.” November 2025.
msci.com
[6]
Balsillie Papers. “Artificial Intelligence, the Environment and Resource Conflict: Emerging Challenges in Global Governance.” September 2025.
balsilliepapers.ca
[7]
Consumer Reports. “AI Data Centers: Big Tech’s Impact on Electric Bills, Water, and More.” March 2026.
consumerreports.org
[8]
International Energy Agency. “Growing Geopolitical Tensions Underscore the Need for Stronger Action on Critical Minerals Security.” February 2025.
iea.org
[9]
ODI Think Change. “Critical Minerals Geopolitics in 2026: Risks, Supply Chains and Global Power Shifts.” 2026.
odi.org
[10]
Science (AAAS). “Global Majority Countries Must Embed Critical Minerals into AI Governance.” March 2026.
science.org
[11]
Foreign Policy Analytics. “Artificial Intelligence and the Critical Minerals Crunch.” October 2025.
fpanalytics.foreignpolicy.com
[12]
OpenDemocracy. “Who Really Wins from the US Critical Minerals Drive?” February 2026.
opendemocracy.net
[13]
Mining South East Europe. “The Great Critical Minerals Shift: How Energy, Defence and AI Are Reshaping the Global Resource Strategy.” June 2026.
miningsee.eu
[14]
Council on Foreign Relations. “The U.S. Critical Minerals Dilemma: What to Know.” July 2025.
cfr.org
[15]
Northeast Bylines. “How AI Will Create an Even More Violent and Unstable World.” June 2026.
northeastbylines.co.uk
[16]
World Economic Forum. “Global Risks Report 2026.” January 2026.
weforum.org © 2026 TechConsult Insight · Strategy & Emerging Risk · All Rights Reserved