For more than a century, the power of nations has been measured by their ability to control energy resources, trade routes, and industrial infrastructure. Coal accompanied the rise of the first manufacturing powers. Oil reshaped the geopolitical balance of the twentieth century, determined alliances, fuelled conflicts, and gave certain territories a strategic importance far greater than their demographic weight. Another resource is now entering this history: computing power.

On October 13, 2026, the Organisation for Economic Co-operation and Development (OECD), through its work associated with the Global Partnership on Artificial Intelligence, published a report entitled Securing AI Compute: Government Policy Approaches. Behind its administrative language, the document describes a profound transformation in industrial policy. Governments no longer regard advanced computing infrastructure as merely technical equipment. They are beginning to treat it as a strategic asset whose availability, location, and control may determine their economic autonomy.

Of the 46 member countries of the Global Partnership on Artificial Intelligence examined by the OECD, 44 have a national AI strategy. Including the European Union, 21 jurisdictions now identify computing power as a distinct pillar of their strategies, while another ten make explicit commitments. The movement is accelerating: among the 18 strategies adopted or updated in 2025 and 2026, thirteen give computing a central role, compared with eight out of 26 strategies adopted between 2019 and 2024.

This evolution marks a change in the nature of the competition. The first phase of the global race for artificial intelligence focused primarily on algorithms, data, and talent. The next is unfolding in semiconductor factories, data centres, electricity networks, and supply chains. Artificial intelligence is gradually ceasing to appear as an essentially intangible industry. Its physical infrastructure, capital intensity, and territorial dependencies are becoming increasingly visible.

Computing Becomes a Strategic Resource

A graphics processing unit, or GPU, was originally designed to accelerate image processing. Its architecture, capable of executing large numbers of operations simultaneously, has made it a central instrument of machine learning. The most advanced artificial intelligence systems now rely on clusters of specialised processors, connected through extremely high-speed networks and installed in data centres whose technical complexity increasingly resembles that of major industrial facilities.

Yet the strategic value does not reside in the processor alone. It depends on the entire system required to operate it: high-bandwidth memory, interconnections, software, electricity supply, cooling, maintenance, and continuous infrastructure access.

This interdependence transforms computing power into a systemic resource. A country may possess world-class researchers, innovative companies, and extensive datasets without having the physical means to train advanced models. It must then rent that capacity from foreign operators, accept their commercial terms, and depend on their investment decisions.

The question is therefore no longer simply who develops the best models. It is who can train them, deploy them at scale, adapt them to national requirements, and guarantee their operation when economic or political relations deteriorate.

The comparison with oil has historical significance, but also an important limitation. Oil is a natural resource concentrated in particular territories. Computing power is an industrial capability that must be manufactured, renewed, and continuously upgraded as technological progress renders previous generations less competitive. In theory, it can be developed almost anywhere, but in practice only where capital, equipment, expertise, and energy converge.

Its scarcity is not geological. It is technological, financial, and organisational.

Sovereignty Built on Dependencies

The global geography of computing has one defining characteristic: no major power independently controls the entire value chain.

The United States occupies a dominant position in the design of advanced accelerators, artificial intelligence software, and global cloud computing platforms. Nvidia, AMD, and the major American digital service providers play a structuring role in this ecosystem. Yet the physical manufacturing of the most sophisticated semiconductors depends heavily on industrial capacity located in Asia, particularly Taiwan.

South Korea holds a strategic position in advanced memory technologies. The Netherlands, through ASML, is home to an indispensable supplier of leading-edge lithography equipment. Japan retains major capabilities in the materials, components, and machinery required for semiconductor manufacturing.

The result is a deeply internationalised production chain in which every segment can become a point of vulnerability.

American restrictions on exports of certain advanced semiconductors to China have demonstrated that this interdependence can be used as an instrument of foreign policy. Technological control provides more than a commercial advantage. It can influence the industrial trajectory of a competitor.

China has responded by increasing investment in domestic semiconductors, homegrown accelerators, computing infrastructure, and software intended to reduce dependence on foreign technologies. This strategy pursues two distinct objectives: expanding available capacity and reducing exposure to external restrictions.

However, producing more components does not guarantee complete autonomy. Manufacturing processes, production equipment, memory technologies, software ecosystems, and industrial yields impose constraints that cannot be eliminated through political decisions alone.

Computing sovereignty is therefore being constructed within a fundamental contradiction: governments seek autonomy through infrastructure whose production requires extensive international cooperation.

The United States and the Logic of Concentration

America's position rests on a combination that is difficult to reproduce: major technology companies, deep capital markets, research capabilities, component suppliers, and enormous commercial demand.

The leading cloud and artificial intelligence operators can mobilise tens of billions of dollars in investment, develop proprietary accelerators, negotiate long-term energy contracts, and build data centres across multiple regions.

This industrial strength creates a cumulative advantage. Companies possessing the best infrastructure attract researchers, developers, and customers. The resulting revenue and access to capital then finance the next generation of capacity. Access to computing power becomes a driver of economic concentration.

Yet this model also contains significant vulnerabilities. Capital expenditure often precedes expected revenue by several years. Equipment depreciates rapidly, while requirements for electricity, land, and grid connections continue to grow.

Profitability therefore depends not only on future demand for artificial intelligence, but also on infrastructure utilisation rates, electricity prices, accelerator lifespans, and the ability of companies to convert investment into commercially viable services.

An announced data centre is not a completed data centre. Installed capacity is not necessarily utilised capacity. And utilised capacity does not guarantee an adequate financial return.

This distinction is becoming essential in an industry where project announcements can reach extraordinary financial proportions long before operational viability has been demonstrated.

Europe Confronts the Problem of Scale

The European Union approaches this competition from a different position. It possesses substantial scientific, industrial, and regulatory capabilities, but its digital ecosystem remains more fragmented than that of the United States.

European high-performance computing initiatives, particularly through EuroHPC, seek to pool investment and make advanced infrastructure accessible to researchers, businesses, and public institutions.

This approach reflects a structural constraint: most European countries individually lack the financial depth, domestic demand, or industrial resources necessary to reproduce the entire American model.

Pooling resources therefore becomes an instrument of power, but it also introduces questions of governance. Who determines the allocation of computing capacity? Which projects receive priority access? How should investment be distributed among territories? And how far can European sovereignty extend when essential equipment comes from foreign suppliers?

Europe's challenge cannot be reduced to an investment shortfall. It concerns the conversion of public and private expenditure into industrial capacity that is genuinely available.

Financing a supercomputer, attracting a cloud operator, and constructing a data centre do not necessarily produce equivalent effects on sovereignty. Equipment ownership, software control, data location, and access conditions must be examined separately.

Europe can host substantial infrastructure while remaining dependent on foreign actors for its operation or technological development.

Conversely, infrastructure developed through international cooperation can strengthen European autonomy if it guarantees durable, secure, and controllable access.

Sovereignty is therefore neither synonymous with geographic location nor with legal ownership. It depends on effective decision-making power.

Canada and the United Kingdom Explore Another Path

The OECD report identifies Canada and the United Kingdom as the first two members of the Global Partnership on Artificial Intelligence to have adopted national strategies specifically dedicated to computing power, separate from their broader AI strategies.

In December 2024, Canada introduced a strategy backed by CAD 2 billion. It provides for up to CAD 1 billion for sovereign public computing infrastructure, CAD 700 million to encourage the construction of commercial data centres, and CAD 300 million to improve access to computing resources for businesses and researchers.

The United Kingdom has established a roadmap involving £2 billion in public commitments through 2030, including the expansion of national research infrastructure and the development of a new supercomputer in Edinburgh.

These policies reveal a more pragmatic understanding of sovereignty. The objective is not necessarily to manufacture every component or compete directly with American technology giants, but to ensure that critical capacity remains accessible to public institutions, researchers, and domestic businesses.

This distinction is fundamental. An economy can remain technologically open without becoming entirely dependent, provided it possesses credible mechanisms to ensure the continuity of essential functions.

However, announced budget allocations must not be confused with delivered infrastructure. The OECD report primarily documents policy orientations and government instruments, rather than providing a comprehensive inventory of comparable operational capabilities.

It records the emergence of an industrial policy. It does not yet establish its success.

The Gulf and the Conversion of Energy Capital

For the Gulf states, the rise of computing power presents a particular opportunity: converting part of their energy and financial advantage into technological influence.

Saudi Arabia and the United Arab Emirates have spent several years developing advanced digital ecosystems, attracting international partners, and investing in the infrastructure required for artificial intelligence.

Their strategy follows a logic of economic diversification, but also a historical continuity. Energy resources have enabled the accumulation of substantial capital. That capital can now finance the infrastructure of an industry whose operation itself requires abundant and reliable energy.

The parallel with the great industrial transformations of the twentieth century becomes particularly visible.

Nevertheless, access to energy and capital is insufficient to create an autonomous technological ecosystem. The acquisition of the most advanced accelerators remains subject to commercial and geopolitical considerations. Specialist expertise, software, and maintenance capabilities create additional dependencies.

Climatic conditions introduce another constraint. In hot regions, data centre cooling may require more expensive or technically complex solutions, depending on the technologies selected and the availability of local water resources.

Success will therefore depend on the ability of these countries to transform physical investment into durable industrial expertise, domestic businesses, and productive economic applications.

Without such a transformation, the infrastructure risks becoming little more than a territorial extension of technological ecosystems controlled elsewhere.

Electricity: Artificial Intelligence's New Frontier

The competition for computing power is encountering a limitation that digital strategies have long treated as secondary: electricity.

In its 2025 report Energy and AI, the International Energy Agency estimated that data centres consumed approximately 415 terawatt-hours of electricity in 2024, representing around 1.5% of global electricity consumption.

Its central scenario projected consumption of approximately 945 terawatt-hours by 2030, more than twice the 2024 level. Data centres would then account for slightly less than 3% of global electricity demand.

These figures concern all data centres, not artificial intelligence alone. AI nevertheless represents one of the principal drivers of the anticipated increase.

The challenge extends beyond the total quantity of electricity generated. Advanced computing infrastructure requires stable power, adequately dimensioned networks, and appropriate cooling capacity. It must be located in territories where constraints relating to grid connections, land availability, and energy supply can be overcome.

Industrial timelines, however, are not synchronised.

A data centre may be developed within a few years, while new transmission lines, generating capacity, or grid equipment can require considerably longer lead times.

The availability of transformers, turbines, electrical equipment, and specialised personnel may become as decisive as the availability of processors.

Technological competition is consequently extending into energy infrastructure. Countries capable of generating abundant, competitively priced, reliable electricity and delivering it rapidly to new computing facilities possess a potential advantage.

Energy sources are not interchangeable, however. Solar and wind power can provide competitive electricity, but their variability requires storage, flexibility, or complementary supply. Natural gas offers dispatchable generation, but creates exposure to fuel markets and emissions. Nuclear power offers continuous low-carbon electricity, although construction timelines and initial capital costs may limit its immediate contribution.

Computing is therefore bringing traditional energy and industrial trade-offs back into the centre of the digital economy.

The country with the best algorithms will not necessarily be the one capable of operating them at the lowest cost.

A Geography That Could Benefit New Territories

The energy intensity of artificial intelligence could alter the map of digital investment.

Countries with competitive electricity, available land, international connectivity, and institutional stability may attract infrastructure previously concentrated in a small number of technological centres.

The Nordic countries, certain Gulf economies, and several nations with substantial renewable energy resources offer, to varying degrees, favourable conditions.

But this redistribution is neither automatic nor uniform.

Proximity to users remains essential for applications requiring low latency. Large training facilities, by contrast, may benefit from locations farther from final markets, provided communications networks, security, and energy supply are sufficient.

This distinction could encourage territorial specialisation between model-training centres, inference infrastructure, data storage, and digital services.

It also raises a question for emerging economies. Should they seek to construct expensive national infrastructure, or prioritise regional partnerships providing access to shared capacity?

For many countries, the second approach may prove more economically rational.

A sovereignty policy should not necessarily aim for technological self-sufficiency. It should identify the functions for which external dependence creates a strategic risk and those for which contractual cooperation provides adequate protection.

Computing could therefore become a new field of regional integration, comparable in certain respects to energy interconnections or telecommunications infrastructure.

The Hidden Cost of Sovereignty

National strategies often emphasise initial investment. Yet the real cost of computing infrastructure extends far beyond purchasing accelerators and constructing buildings.

It includes energy expenditure, cooling, networks, software licences, cybersecurity, specialised personnel, maintenance, and equipment replacement.

The last of these is particularly important. Unlike transport infrastructure or certain energy installations whose useful lives may extend over several decades, advanced computing equipment evolves rapidly.

A substantial investment may lose much of its economic competitiveness well before the end of its physical operating life.

Governments must therefore choose between public ownership, industrial partnerships, subsidised access, and long-term contracts.

National ownership of a computing facility may guarantee certain sovereign prerogatives, but it also exposes public finances to technological obsolescence. Renting private capacity can offer greater flexibility while creating contractual dependence.

There is no universal solution.

The appropriate choice depends on the nature of the applications, data sensitivity, national security requirements, the domestic industrial base, and the financial capacity of the state.

The decisive question is therefore not how much a country announces it will invest, but what useful capacity it obtains, at what total cost, and with what degree of control.

A New Hierarchy of Powers

The emergence of computing as strategic infrastructure could change the traditional criteria of economic power.

Large economies possess clear advantages in financing, research, and domestic demand. Yet other factors are becoming decisive: access to energy, equipment availability, regulatory stability, international connectivity, and the ability to organise industrial partnerships.

This evolution does not mean that GPUs will replace oil in the global hierarchy of resources. Hydrocarbons remain essential to transport, industry, and numerous energy systems. Computing does not possess the same applications or economic characteristics.

The comparison instead reflects a transformation in the role of critical infrastructure.

In the twentieth century, controlling energy made it possible to power machines, transport systems, and industries. In the twenty-first, mastering computing infrastructure increasingly helps determine who can develop, deploy, and use intelligent systems that will shape a growing share of production, research, defence, and services.

The difference lies in the accumulation of dependencies. Computing power itself requires energy resources, raw materials, precision industries, telecommunications networks, and capital.

It does not replace the material foundations of power. It adds another layer to them.

This accumulation could strengthen countries capable of coordinating policies previously treated separately: industry, energy, research, foreign trade, security, and financing.

Conversely, governments that approach artificial intelligence exclusively as a matter of software or regulation risk underestimating the physical conditions of their autonomy.

From Declared Sovereignty to Measurable Sovereignty

The OECD report reflects a rapidly growing awareness. Within a few years, computing power has moved from being a technical resource to becoming an explicit objective of national policy.

But this development introduces a new requirement: measuring outcomes rather than intentions.

The number of strategies adopted, budgets announced, or data centres proposed is insufficient to establish a country's actual computing strength.

Meaningful comparisons would require information on operational capacity, performance, availability, utilisation rates, energy costs, accessibility to businesses, and the degree of technological dependence.

They would also need to distinguish infrastructure used to train the most advanced models from facilities dedicated to everyday inference, scientific research, or industrial applications.

These functions do not require the same equipment or levels of investment.

An economy can achieve meaningful autonomy in the use of artificial intelligence without possessing the infrastructure necessary to train the most expensive frontier models.

Conversely, hosting a large data centre does not guarantee that domestic businesses will gain competitive access to its resources.

Sovereignty thus becomes a matter of functional control rather than technological prestige.

The true indicator will be a state's ability to maintain essential services, support its productive economy, and develop its own applications of artificial intelligence, including when external conditions become unfavourable.

Industrial history shows that major technological transformations do not immediately redistribute power. They often begin by strengthening already dominant actors before creating new opportunities for territories capable of identifying the resources and expertise that have become strategic.

Artificial intelligence may follow the same trajectory.

The United States currently possesses considerable advantages. China is pursuing industrial substitution. Europe is seeking to pool its resources. Canada and the United Kingdom are experimenting with sovereign capacity policies. The Gulf states are attempting to convert their energy and financial strength into technological influence.

Behind these different strategies lies a common reality: artificial intelligence is no longer simply a competition to produce better systems. It is becoming a competition to control the material conditions of their existence.

For much of modern history, the strategic question was who controlled the resources capable of powering industrial economies. Another question is now being added: who will control the infrastructure capable of powering their artificial intelligence?

The answer will not be found exclusively in laboratories or investment announcements. It will emerge in semiconductor factories, energy contracts, electricity networks, and data centres that actually enter operation.

After oil, the GPU does not mark the disappearance of geography. On the contrary, it reminds us that even the most advanced technologies remain dependent on the territories, resources, and infrastructure that make them possible.

Main Sources

  1. OECD (October 13, 2026) — Securing AI Compute: Government Policy Approaches, OECD Policy Briefs, No. 67. https://doi.org/10.1787/7f96dc99-en
  2. International Energy Agency (2025) — Energy and AI. https://www.iea.org/reports/energy-and-ai
  3. Government of Canada — Canadian Sovereign AI Compute Strategy. https://ised-isde.canada.ca/site/ised/en/canadian-sovereign-ai-compute-strategy
  4. UK Government — Compute Roadmap, Department for Science, Innovation and Technology, July 2025.
  5. OECD (2023) — A Blueprint for Building National Compute Capacity for Artificial Intelligence.