The AI Power Play: Questions of AI Sovereignty cannot be Divorced from Energy and Geopolitics

AI Sovereignty: Navigating the Complex Intersection of Innovation and Public Policy

AI sovereignty is a hard problem to solve. Some constraints are unique to the sector, such as access to semiconductor supplies and frontier models, or inherent compute ceilings. But others are more structural – these include data centres’ energy needs and their inherent vulnerabilities to real-world geopolitics. States should start thinking more about these.

By Ankita Dhawan and Pranav Butalia

Understanding AI Sovereignty

“AI sovereignty” is the latest buzzword in policy circles. Everyone wants it, but no one can define it. That is because it is hard to define and harder to achieve. A working definition, for the sake of this article, is the ability of a state actor to maintain autonomous, enforceable, and continuous control over strategically significant AI infrastructure, from brick-and-mortar to data and compute, free from the threat of external coercive actions.

Policymakers are currently over-indexing on the temporary and visible problems affecting AI sovereignty – chip embargoes, GPU shortages, and export controls on models. In our view, the history of innovation is a testament to the fact that temporary variables tend to solve themselves. Scarcity creates incentive; incentive produces innovation; innovation restores access.

Countries that will eventually attain AI sovereignty are the ones currently asking more basic questions. “Can we power, cool, and protect our infrastructure? Continuously, and under pressure, without depending on other nations?” We see these answers as key structural levers in the AI race, and ones that do not necessarily resolve through market dynamics alone.

Energy Sufficiency: The Real Power Play

At its heart, the AI race is a resource war masquerading as a software problem. Critical minerals are required to build servers that process data. Data processing generates heat, and heat must be removed continuously lest the entire system fail. All these processes require access to continuous and reliable power. Consequently, the countries best positioned to win the AI race are those where the problem of energy sufficiency has already been solved.

The classic example is the UAE. Here, a legacy hydrocarbon economy and the abundance of sovereign capital have allowed alignment between energy needs and AI sovereignty. Water scarcity, a downstream vulnerability, has been tackled through industrial desalination plants and the creation of dedicated liquid cooling facilities. As a result, sovereign champions like MGX and G42 have been able to rapidly build out domestic AI infrastructure on a national scale. Falcon has emerged as a homegrown frontier model, domestic chip access is expanding, and private capital from Microsoft and Stargate signals external confidence. In fact, the UAE is becoming an exporter of “sovereignty as a service,” a model that we discuss in greater detail under the next heading.

Global Energy Trends

What about the rest of the world? Nuclear energy is staging a comeback, for starters. In the United States, for instance, Microsoft has signed a twenty-year agreement to restart dormant nuclear plants. Meanwhile, nuclear bets made by France decades before datacentres existed are producing exactly what AI infrastructure requires – cheap, stable power supplies insulated from price shocks and seasonal variations. Mistral’s emergence and SoftBank’s recent investments in France validate our thesis: a solid energy floor accelerates the development of sovereign AI capacity, which in turn attracts private capital. Such private capital, we believe, is not antithetical to sovereignty – if deployed correctly, it can be an enabler instead.

At the other end of the spectrum, energy sufficiency is not just about generating power – it is equally about managing it. Grid adequacy, or the ability to evacuate generated power through transmission corridors, is increasingly becoming a ceiling in many markets. Nigeria, for example, has raw engineering talent as well as significant hyperscaler interest. However, the country frequently experiences severe grid instability, forcing datacentres to build their own fossil-fuel power plants in a direct clash with the state’s broader energy and climate commitments.

This was also the case in India in the early 2000s. The India of today presents an entirely different kind of challenge, with an annual seasonal energy surplus that typically gets wasted. Better-managed grids would allow datacentres to act as energy sinks, absorbing any excess power in the system. Companies have begun to act in this direction. Google recently obtained a licence allowing it to procure and manage electricity directly at the grid level to power its upcoming one-gigawatt data centre located in one of India’s largest solar corridors.

Geopolitical Viability: Here and Now

While energy sufficiency determines whether a country can build data centre capacity, geopolitical viability determines whether it should. For decades, data crossed borders freely. Localisation was not the norm, and the question of where a data centre sat was answered not by energy need, but by regulatory feasibility and geopolitical stability.

Singapore became one of the world’s most important data hubs because it sat at the intersection of the intercontinental subsea cable network while offering rule of law, political stability, and access to capital. This, despite the fact that it is perpetually warm and exists as a small island with no natural energy or water advantage. Ireland too built one of Europe’s most significant data ecosystems on the back of an English-speaking workforce, predictable common law rulings, and a low-tax regime offering a gateway into the unified European market.

Today, geopolitical viability has become even more consequential, though its meaning has also evolved with technological advancements. The sheer volume of data, both an input and an output, changes everything. AI workloads require exponentially more compute than anything that preceded them, and infrastructure has struggled to keep pace. Where entire regions were once serviced by a small handful of centralised datacentres, today’s tiered architecture spans large hubs, smaller regional facilities, and thousands of edge nodes sitting next to the devices and users actually generating the data. AWS alone utilises this model in over 600 locations worldwide. A natural consequence of more widely dispersed infrastructure is more jurisdictions, more laws, more exposure, and more points of potential failure.

This makes data sovereignty increasingly hard to solve for. In today’s age, data sovereignty has also become interchangeable with AI sovereignty, with most governments reaching for localisation as an instrument to achieve both simultaneously. The result is a wave of domestic capacity mandates that require the building of local infrastructure and grant national governments ultimate supervision and control. Brazil has crossed one gigawatt of data centre capacity on the back of its localisation rules – more than the next four Latin American markets combined. Malaysia’s pipeline has surged to over five gigawatts, with Johor alone accounting for nearly 80% of national capacity.

But, from a geopolitical lens, localisation does not guarantee sovereignty – a distinction that cyber warfare has made abundantly clear. Local data infrastructure has become a vulnerable attack vector. Estonia learned this the hard way in 2007, after which it resorted to hosting government data abroad in “digital embassies” which preserve domestic legal jurisdiction. Ukraine hedges its geopolitical risk by moving strategic government data out of the country – keeping data inside its borders can mean losing it entirely. Ultimately, while data sovereignty is about control and continuity, data localisation is about geography. In peacetime, the two can look identical. In conflict, they diverge fast.

Sovereignty as a Service

For most countries, data sovereignty and AI sovereignty remain constrained by capital. They can neither afford to build infrastructure at scale nor afford to cede control. A potential answer is sovereignty as a service – the digital embassy model pioneered by Estonia, which has since been formalised into a diplomatic and commercial offering. The UAE, through G42 and its network of internationally distributed facilities, has taken this the furthest through government-to-government arrangements. Here, legal jurisdiction over data travels with the data itself, not with the servers. Host states provide the infrastructure while client states retain jurisdiction and control. Encryption algorithms ensure that data moves to the most suitable servers (which the host can neither access nor read) with the client flag accompanying. In such scenarios, hosts are physically present but legally blind.

Space: The Final Frontier

Ironically, energy sufficiency and geopolitical viability are not independent variables. Strength (or weakness) in one feeds into the other. Another place where both constraints seem simultaneously lifted is space. Here, energy and cooling are solved problems – infinite solar capacity exists without thermal limits.

By contrast, geopolitics in space starts from a different position entirely. The Outer Space Treaty formally declares space to be the common heritage of all mankind. This effectively means that no single state can claim sovereignty over space or exclusivity over the installation of orbital infrastructure. The same encryption architecture that allows a foreign flag to reside alongside its data in an Abu Dhabi facility can be deployed in orbit. The data does not care where the server is.

Space might be the only frontier where sovereignty does not have to be immediately fought for. Only time will tell whether it remains a neutral ground or becomes the next contested territory.

Ankita Dhawan is Founder and Principal Policy Strategist at Consilium Advisors, a regulatory and policy advisory focusing on AI. Pranav is an energy and infrastructure lawyer who currently works as a PPP Consultant at the World Bank Group. Views are, and remain, the authors’ own.