The next great strategic liability in world politics isn’t oil dependence or even semiconductor dependence. It is the risk of AI coercion, stemming from reliance on foreign-controlled computing infrastructure, software, and energy systems for functions that are fast becoming essential to national growth, governance, and security. AI is becoming the operating system of modern economies: embedded in power grids, financial systems, healthcare, logistics, and national defense. AI coercion is the use or withholding of AI systems, networks, and hardware to influence another nation. Imagine if a foreign country could withhold the AI your country needs to run these sectors. Most governments have not yet treated transboundary AI dependence and coercion as a strategic threat. They should. The window for action is closing. Globally, AI infrastructure is consolidating faster than governments are moving and governments need to affirmatively build resilience through diplomatic initiatives and AI partnerships.
The Threat
Why should AI dependence worry us more than past dependence on foreign oil — a dependence that, for most countries, has been manageable until now? Maybe oil dependence was manageable but that’s no longer true. For decades, American military power and multilateral institutions like the Law of the Sea Convention guaranteed the free flow of goods across oceans, making commodity dependence a calculated risk. Today, in contrast, Washington is no longer the world’s free-trade enforcer — it is increasingly a selective gatekeeper, actively blockading Iran, Cuba, and Venezuela and deploying tariffs as geopolitical weapons. And multilateral institutions are on the ropes.
International AI coercion risk also remains categorically different for a structural reason: the global commodity trade is for the most part highly diffuse, and distributed with relatively few chokepoints, while the AI supply chain is the opposite. A handful of companies — and in some cases, a single individual — can extend or withdraw access at will. Elon Musk’s ability to shape the battlefield in Ukraine simply by toggling Starlink coverage is not an edge case. It is a preview of the leverage that AI dependency creates for private companies. This risk of private coercion is magnified by the all-too-foreseeable risk that AI companies will be coerced by the countries in which they are based to become tools of official foreign policy.
The tools of AI coercion are already visible. Advanced AI capacity is concentrated predominantly in the United States and China, and both have demonstrated a willingness to use technological dominance as an instrument of statecraft. Washington has deployed sweeping export controls restricting access to the advanced semiconductors on which frontier AI runs. Beijing controls roughly 90 percent of refining capacity for rare earth minerals, which are essential for AI semiconductors and data centers. China has already imposed export restrictions on rare earth minerals, signaling that critical AI supply chains run through chokepoints Beijing is prepared to exploit. The mere fact of dependence creates leverage.
Strategies for Countering AI Dependence
Countering AI dependence — which rests on the ability of a country to operate the AI systems on which its economy and government depend with agency and autonomy — is now essential for all countries seeking true sovereignty. Control of AI rests on four things: compute (the chips and data centers that process information), intelligence (the data and models that generate insights), applications (the software delivering AI to hospitals, schools, and militaries), and the electricity that powers all three.
Today, control over those four sovereignty elements is extraordinarily concentrated. The United States alone attracted $159 billion in private AI investment in 2025, while U.S. institutions produced 40 notable AI models compared to Europe’s three. Most countries are not in this race. A handful are trying to be. France’s Ministry of Defense has contracted with Mistral AI to develop a French “AI brain” and keep military intelligence on French servers. South Korea built HyperCLOVA X, a large language model trained on Korean language and culture, rather than outsource its cognitive infrastructure abroad. Most countries are not responding with comparable seriousness. Canada committed $2 billion to sovereign AI infrastructure, yet key facilities remain operated by U.S. cloud providers subject to American law and government influence. Across Latin America, Africa, parts of Europe, and most of Asia, governments are largely ordering takeout — welcoming foreign data centers, providing the land and the electricity, while foreign firms keep the models, the chips, and the intelligence. While giving lip service to the idea of building local AI capacity, in reality, for the next decade or more these countries will merely rent out low-value infrastructure to house someone else’s high-value intellectual property.
Governments facing this reality have three options. The first is autarky: complete self-sufficiency, keeping the full AI stack at home. Only a handful of states could plausibly achieve this absolutist version of AI sovereignty and even for them it may be a myth. The second option is accepting dependence passively, which risks digital colonialism, with dominant powers supplying essential systems to vassal states on terms the latter cannot control. Any nation that slides far enough down that path risks becoming controlled by the algorithm, sovereign in name but dependent in practice. The third option, and the only workable one for most countries, is to reduce risks as much as possible through networked interdependence: preserving strategic control of their “AI brain” through trusted partnerships and reciprocal legal commitments.
History shows that structured interdependence can substitute for autarky when the right agreements exist. After the 1973 oil crisis sent economies into recession, industrial nations created the International Energy Agency (IEA), requiring members to hold emergency reserves and commit to collective action during supply disruptions. The result was meaningfully improved energy security without impossible self-sufficiency. AI risk mitigation needs a similar architecture: formal agreements among like-minded states that guarantee access, reduce cutoff risk, and prevent the weaponization of critical AI, energy, and financial linkages. Those agreements do not yet exist, and as discussed in more detail below, building them is a central diplomatic task of the AI age.
Of the elements above, electricity is already a critical constraint. Chips are useless without power. The IEA projects that global data center electricity consumption will double by 2030, rising from 415 terawatt-hours today to 945 terawatt-hours, roughly equal to Japan’s entire current national power consumption. In the United States, data centers will account for nearly half of all electricity demand growth this decade. Elon Musk has already warned that by the end of 2026, the U.S. may already be producing more chips than it can power. The IEA calls this the Age of Electricity, as the efficient expansion of electricity capacity and grids turns into a critical challenge.
Among all electricity options, wind and solar offer the strongest foundation for reducing international weaponization of AI dependence. Those energy technologies are now the cheapest source of new power in most of the world: 91 percent of newly commissioned utility-scale renewable projects in 2024 produced electricity more cheaply than the least expensive new fossil-fuel alternative, and battery storage costs have fallen 93 percent since 2010. Renewable energy produced at home or imported from trusted allies is far less vulnerable to being embargoed, sanctioned, or repriced by a foreign government. The distributed nature of solar and wind makes them more resilient to single-point disruptions than centralized fossil energy. Wind and solar do carry supply chain vulnerabilities, as panels, turbines, and batteries depend heavily on Chinese manufacturing, and diversifying those supply chains is itself a diplomatic task. However, once a country installs a panel, it will produce power reliably for 25 to 30 years, powered by sunshine in that country. No other electricity source matches wind and solar on cost, scalability, and strategic independence simultaneously. Some see nuclear power, particularly a new generation of advanced reactors, as providing another source of domestically produced, reliable energy. Today, however, advanced nuclear cannot match solar and wind when it comes to speed of deployment, cost, or safety.
Fossil-dependent AI power, by contrast, is a strategic trap. The U.S.-led conflict with Iran has disrupted traffic through the Strait of Hormuz, the chokepoint through which flows roughly one-fifth of the world’s daily oil and gas supply. The IEA has called it the largest supply disruption in the history of the global oil market. Countries that build AI infrastructure on imported fossil fuels are not building toward independence. As one expert has observed, you cannot weaponize the sun. You can, and adversaries do, weaponize a strait.
Networked AI risk reduction arrangements must be built through deliberate diplomatic effort, going well beyond ordinary commercial and trade agreements. Success requires national investments in AI excellence and their own “AI brain,” trusted partners for critical layers of the AI stack, formal agreements that reduce the risk of cutoff or coercion, and built-in redundancy so that no single point of failure becomes a chokepoint.
To begin, nations in a position to do so (a few major economies and technology-forward emerging economies) should seek to ensure that they are indispensable players in the global AI supply chain. In general, that will mean being the best in the world at some part of the AI stack, such as very specialized aspects of semiconductor manufacturing or owning essential intellectual property. Nations and regions that are best-in-class will be essential to other nations and least likely to face foreign AI coercion, as Jeremy Jurgens argues.
Diplomatic Strategies for AI Interdependence
Domestic action alone will not suffice, however, to build national “AI brains.” Diplomatically, four realistic pathways already have some political momentum, and each works because every participant benefits. The first diplomatic pathway is regional electricity integration, viable where grids are physically connected. Norway generates over 90 percent of its electricity from hydropower and exports clean power to Germany, the Netherlands, and the United Kingdom; Sweden adds further low-carbon exports through hydro and nuclear. Singapore imports hydropower from Laos through Thailand and Malaysia. The diplomatic product could be regional power-pooling agreements that treat AI infrastructure security and clean energy expansion as joint objectives.
The second avenue for AI diplomacy is cross-border hosting of computation. Because data moves more easily than electricity, countries can place certain computational workloads in trusted foreign jurisdictions with abundant power, provided arrangements include legal protections, data privacy, continuity guarantees, and clear access rights. France and the UAE signed a framework in February 2025 covering a one-gigawatt AI data center for the UAE in France, joint investment of up to $50 billion, and virtual data embassies for sovereign cloud infrastructure in both countries. France gains investment and strategic partnership; the UAE gains trusted low-carbon compute capacity it cannot yet generate at home. The model is replicable across many country pairs.
The third diplomatic pathway is strategic renewable procurement and investment. Countries cannot easily import electrons from distant partners, but they can align industrial policy so that AI ambition drives clean power buildout at home. Japan’s latest Strategic Energy Plan requires AI data centers to run on decarbonized electricity and subsidizes up to 50 percent of capital investment for facilities using 100 percent clean power. These are industrial policies that reduce costs and improve AI security, while also shifting to clean energy. This formula provides a template other governments can adopt and potentially implement jointly.
The fourth diplomatic priority is sustainable financing for the AI economy. Governments must create clear and stable investment environments for AI: transparent rules, enforceable contracts, and predictable regulations. These should clarify public expectations on environmental outcomes, technology standards, and AI safeguards, for example, to create the clarity needed to attract private capital at scale. Diplomatic agreements that harmonize investment frameworks across borders and create common standards for AI facilities can unlock private financing far more effectively than public grants alone.
The Climate Connection
While these four international strategies to enhance AI control should be pursued for that reason alone, they also make sense for another, more unexpected rationale. Surprisingly, reducing vulnerability to AI coercion would accelerate global action against climate change.
Although in the near term, electricity-hungry AI data centers will add to global greenhouse gas emissions, the AI revolution stands to reduce climate pollution. First, as argued above, the pursuit of AI sovereignty will result in a massive deployment of renewable energy.
Second, the build-out of the clean energy system would likely pave the way for the rapid adoption of a broader array of clean technologies, including electric vehicles, heat pumps, batteries and digitalization, as well as powering new sectors such as green water treatment and waste processing, green cement, and green steel. Why? Treating renewable energy as baseload – the foundation of the electricity grid – results in an abundance of low-cost clean energy most of the time. This is because avoiding blackouts on cloudy, windless days requires building more renewable plants than are needed on a typical day. This means that most days, consumers and industry have access to very low-cost surplus electricity, which incentivizes switching away from fossil fuels.
Third, AI could make the electricity grid more efficient. The IEA estimates that widespread AI adoption in power, transport, and industry could reduce global emissions by around 1,400 million tonnes of CO2 by 2035, larger than the projected emissions from all global data centers themselves.
The most interesting major climate news of the next few years, therefore, may be the new alignment of interests among energy, environment, and AI ministries and their stakeholders. That new political coalition for the green economy will yield a huge climate dividend, available to any government willing to treat AI sovereignty and clean power as two sides of the same strategic coin.
