AI’s bottleneck is physical—competition for power, land, and metals risks sidelining US military needs as commercial AI outpaces national security planning.

The rapidly rising energy demands of artificial intelligence (AI) compute are not a new problem. In 2022, Dominion Energy first announced it could no longer meet projected power demands in “Data Center Alley.” This dilemma echoes a 2024 warning from OpenAI CEO Sam Altman on how future AI progress depends on a massive “energy breakthrough.” 

AI is often depicted as intangible software, but this digital ability is an incredibly material-intensive technology. Its foundation is hardware: data centers filled with servers, cooled by immense volumes of water, wired into copper-dense grids, and built with components reliant on specialized gases, rare earths, and vast amounts of energy. This physical reality has profound implications for US national security, as the commercial AI sector and the military overlap in their sharing of the same, increasingly strained, industrial base.

This shared dependency has forged a new AI-industrial complex, and its primary strategic risk is not foreign dependence, but domestic allocation. Because of intensifying competition for AI’s physical resources (e.g., power, land, and metals), this ‘physical juncture’ for digital capabilities is creating chokepoints that directly impact the industrial base and the warfighting abilities of the US military and its allies.

Power, Land, and Metals: The Physical Chokepoints of AI

Hyperscale data centers function like industrial plants, demanding immense, reliable power. The International Energy Agency estimates that global electricity demand from data centers will approach 1,000 terawatt-hours this year—on par with the entire energy consumption of Japan. Such demands translate into an immense need for material resources, with copper being the key.

While many analyses focus on the energy or compute cost of a ChatGPT search, much less exploration exists around the material costs for that search. 

Copper is the foundational metal of electrification, and the AI boom adds a powerful new layer of demand on top of the needs of expanding power systems. Data center racks in the AI era require two to four times more copper than traditional racks. As a result, global copper demand from data centers is expected to more than double by 2030, with projections suggesting the sector will consume a cumulative 4.3 million tons by 2035.

This demand surge is colliding with a deepening supply crisis. Analysts warn of a looming structural deficit that could reach 10 million tons by 2040. The supply side is inelastic; bringing a new copper mine into production can take more than a decade globally and up to 29 years in the United States. Compounding the problem, refining capacity is heavily concentrated outside of North America, and current markets and pricing are not up to the task of properly incentivizing new infrastructure build.  

This resource competition collides with supply chains for military systems. Communication equipment, electronic warfare systems, advanced radar, and hardened facilities are all copper-intensive. When AI firms outbid utilities for power, acquire land near critical substations, or secure long-term supply agreements, they are fundamentally reshaping the industrial base and thus making the production of military systems a lower priority.

Structural Advantages: Why the Private Sector Outpaces the Military

The structure of the AI-industrial complex inherently favors the private sector. Commercial firms operate with a speed, financial agility, and a risk tolerance that the Pentagon cannot match, even with policy mandates like Executive Order 14265 aimed at accelerating acquisition. Despite pledges by Secretary of Defense Pete Hegseth to improve Pentagon approaches, they are still constrained by slow budget cycles and arcane procurement rules. 

So, how to balance the needs of national security? Market logic allocates resources based on price and expected return; it does not optimize for deterrence, surge capacity, or resilience in wartime. By the time defense requirements are asserted, the key bottlenecks are likely controlled by commercial interests.

Historically, this was addressed through wartime mobilization authorities, but these tools are ill-suited to digital age requirements. Passage of the FY 2026 National Defense Authorization Act (NDAA) did not resolve the governance gap. The legislation correctly identified a threat, but it opted for a phased approach, passing Section 849 to prohibit the procurement of additive manufacturing machines from adversaries. However, it omitted the Senate’s proposed Section 849A, which would have provided the investment authorities (from the “FoRGED Act” framework) to fund domestic alternatives. This created a policy contradiction: a ban on foreign technology without a dedicated mechanism to build a domestic replacement. While the final law includes important exceptions for intelligence and cyber warfare and a one-year window before the ban takes full effect, the core problem remains.

This legislative gap places a heavier burden on existing, less agile tools such as the Defense Production Act (DPA). While the DPA has been used successfully, for example, to secure domestic rare earth magnet production, it is fundamentally an interagency mechanism shared across seven federal departments. It is not the dedicated, rapid instrument needed to manage the scale, speed, and trajectory of the AI-industrial complex. 

To navigate this challenge, the Trump Administration could use existing authorities, such as Executive Order 14265, to implement a more robust framework. Such an approach must accomplish three objectives. First, it must identify defense-critical inputs with precision, moving beyond abstract mineral categories to specific components and infrastructure nodes. Second, it needs to establish mechanisms to protect access to these inputs during demand spikes, using tools such as pre-negotiated offtake agreements or targeted stockpiling of processed components, not just raw ore—similar to a strategy Japan has already employed. Finally, it must formally redefine midstream processing and power grid infrastructure as defense-enabling assets, worthy of the same focus as rail stations, seaports, or airports. With industrial competition becoming more focused on supply inputs, a copper refinery is as critical to national security as a shipyard. 

Another path to alleviating pressure on terrestrial infrastructure is to rethink the physical location of AI compute power. Concepts such as undersea or space-based computing platforms, often viewed as futuristic, may offer some solutions to the allocation problem. This can expand domain capabilities where the US military already possesses deep operational expertise.

Subsea data centers, for example, cooled by ocean water and located near coastal transmission, can reduce land competition and shift AI infrastructure away from congested terrestrial grids. While more speculative, space-based platforms present unique advantages, including constant access to solar energy and inherent resilience against terrestrial disruptions. This illustrates that strategic advantage comes from diversifying sources of computation, power, and materials. 

Learning from China’s Integrated AI Policy Approach

China approaches this challenge with a unified strategy that merges AI policy, industrial development, and national security. Beijing has made massive investments in power generation, grid expansion, and materials processing, even when short-term market signals were unfavorable. This supply chain dominance provides sway over supply, timing, and scale.  Furthermore, Chinese military requirements are integrated directly into civilian infrastructure planning, ensuring that commercial growth reinforces national power via its Military-Civil Fusion policy.

China’s approach is in stark contrast to the American system. US leadership in AI rests on an industrial foundation increasingly shaped by commercial actors optimizing for profit. An adversary could exploit such a gap in a crisis, potentially by buying up the companies and inputs for building and sustaining AI data center growth.

A viable national security strategy for AI must align technology policy, energy infrastructure, and materials governance toward a single, common goal: resilience under demand shocks. This requires prioritizing defense-critical inputs, securing processing capacity, and accepting the hard truth that markets, left to their own devices, can hurt military readiness. In this AI-industrial complex era, strategic advantage is forged by copper, kilowatts, and code. The AI revolution will be won by the country that builds, powers, and sustains the industrial systems that enable AI to work in tandem with the economy and the military, efficiently and effectively. 

About the Authors: Macdonald Amoah, Morgan Bazilian, and Jahara Matisek

Macdonald Amoah is an independent researcher with interests across critical mineral supply chains, advanced manufacturing gaps, the industrial base, and the geopolitical risks in the mining sector.

Morgan D. Bazilian is the director of the Payne Institute and professor at the Colorado School of Mines, with over 20 years of experience in global energy policy and investment. A former World Bank lead energy specialist and senior diplomat at the UN, he has held roles at NREL and in the Irish government, and advisory positions with the World Economic Forum and Oxford. A Fulbright fellow, he has published widely on energy security and international affairs.

Lt. Col. Jahara “Franky” Matisek (PhD) is a US Air Force command pilot, nonresident research fellow at the US Naval War College and the Payne Institute for Public Policy, and a visiting scholar at Northwestern University. He is the most published active-duty officer currently serving, with over 150 articles on industrial base issues, strategy, and warfare.

DOD Disclaimer: Views of Lt Col Matisek are his own.

Image: Audio und werbung/shutterstock

The post Why the US Military Could Lose the Contest for Materials Crucial to AI appeared first on The National Interest.