AI has evolved from a technology experiment into critical infrastructure. The strategic center of gravity has shifted from capability to control over the conditions that enable scale, including compute, energy, trusted data regimes, governance frameworks, and diplomatic leverage to shape international standards. Nations that treat AI as a cross-government infrastructure agenda — rather than a digital ministry initiative — will be better positioned to scale safely and negotiate from strength.
Five years ago, as described in an Arthur D. Little (ADL) PRISM piece on the same topic, we anticipated four forces shaping AI: (1) venture-led growth, (2) a global talent race, (3) private sector adoption, and (4) rapid increase in AI knowledge skills and capacity. All four proved real; they also proved incomplete as a picture of what determines national advantage.
What markets alone cannot secure, but nations cannot afford to be without, now defines the competition: compute capacity, affordable energy, sovereign data regimes, trusted governance, and the diplomatic leverage to shape standards and access. The countries pulling ahead in 2026 have recognized that durable AI advantage is built on these foundations, not on model benchmarks alone.
Today’s global AI race revolves around securing and governing the full technology stack, from hardware and data to models, energy, and regulatory frameworks. Four pillars define sovereign AI capability in practice.
Nations increasingly balance national AI ambitions with strategic alliances. Many have aligned with technological blocs for compute access, semiconductor supply, and model development while maintaining selective independence in sectors considered strategically sensitive. The choice is rarely binary. Countries are building hedged positions, seeking preferential access without full dependency.
The key implication is that the rules governing compute access and AI standards are becoming inseparable from trade and security policy. Nations that treat AI as a purely domestic technology agenda risk finding themselves locked into dependencies on terms they did not negotiate. Figure 1 shows how leading governments are combining investment, compute, safety, and industrial policy to strengthen their position.
Define your alliance posture deliberately. Passive alignment defaults to dependency; active positioning creates room for sovereign choice.
Countries are increasingly aspiring to govern AI end to end across chips, cloud infrastructure, models, and datasets. The strategic concern is not just technical. Nations that rely entirely on foreign AI infrastructure are not just adopting foreign technology; they are embedding foreign values, legal jurisdictions, and strategic priorities into their own critical systems.
Open source models have broadened participation — a genuinely positive development. However, open access alone does not eliminate dependencies on the chips required to run these models, the cloud infrastructure needed to deploy them at scale, the curated data required for fine-tuning, or the reliable power necessary to sustain them. Open models lower the barrier to entry, but they don’t deliver sovereignty. Nations need to distinguish between AI access and AI independence.
AI is being mobilized to address core national priorities, including productivity, healthcare, environmental monitoring, energy resilience, and defense. That moves AI from the domain of technology ministries into the domain of cabinet-level infrastructure planning.
National strategies increasingly define specific sectors and mission areas where AI deployment is a strategic priority. Energy stands out as both a critical application domain and a binding constraint on scale, as described further below. The most powerful national AI strategies are built around concrete national problems, not generic capability targets.
Societal acceptance has become a binding constraint on where and how fast AI can scale. The EU’s public resistance to AI in high-stakes decisions (e.g., hiring and credit scoring) and the global backlash against opaque generative AI systems illustrate the same dynamic: capability without legitimacy does not lead to sustainable scaling.
Nations that embed transparency, accountability, and genuine public oversight into their AI ecosystems gain the legitimacy to move faster, not slower. Technical talent still matters, but public legitimacy determines whether a nation will deploy AI at the scale required for national advantage or be repeatedly constrained by the courts, the press, and civil society. Trust is not a constraint on AI ambition; it is a prerequisite for scaling it. Nations that get governance right will outpace those that treat it as a compliance burden.
These pillars play out differently depending on where a nation sits strategically. As we illustrated in Figure 1, leading governments are combining these levers with various postures and implications for others navigating the same landscape.
Across leading government AI initiatives, one constraint is universal: infrastructure. Within infrastructure, energy has emerged as the most immediate practical limit on scale, the bottleneck that will determine which nations can credibly deploy compute-intensive AI and which cannot.
The scale of the demand shock is striking. The International Energy Agency (IEA) projects global data center electricity consumption will more than double by 2030, rising from approximately 460 TWh in 2024 to over 1,000 TWh, equivalent to adding Japan’s entire current electricity demand to the global system within six years. A single large-scale AI training cluster can consume as much power as a small city. This is no longer an operational planning detail; it is a geopolitical variable.
The challenge is compounded by an uncomfortable paradox at the heart of AI strategy: energy is where AI delivers some of its most valuable benefits and imposes its highest costs.
AI is materially improving the performance of power systems. Machine learning (ML) models support demand forecasting with accuracy that was previously unachievable, enabling grid operators to balance supply and load in real time. Predictive maintenance algorithms are extending the life of generation assets and reducing unplanned outages. Renewable integration, long constrained by intermittency, is increasingly manageable using AI-driven storage dispatch and grid balancing.
Similarly, cybersecurity systems protecting critical energy infrastructure are increasingly AI-augmented.
In oil & gas specifically, AI is accelerating subsurface interpretation, reducing exploration risk, optimizing production in aging fields, and enabling predictive maintenance of offshore and pipeline infrastructure at a cost and speed not previously achievable. The sector that was once the slowest to adopt digital tools is now a consequential proving ground for industrial AI.
AI infrastructure is a structurally significant source of electricity consumption. The shift from retrieval-based search to AI-generated responses alone is estimated to require roughly 10 times the energy per query, according to IEA. Multiply that across billions of daily interactions, and the aggregate demand becomes a planning problem for grid operators and a cost problem for developers. Hyperscale data centers running large model inference are power hungry by design and necessity.
Energy availability is no longer just an operational concern for data center site selection. It is now a strategic variable that shapes where AI infrastructure gets built, which countries attract and retain large-scale compute investment, and, ultimately, which nations can credibly participate at the frontier of AI deployment.
The energy-AI intersection raises three policy questions that a single ministry cannot answer:
Energy is not peripheral to the AI race. It is a core test of whether a nation can align digital ambition with physical infrastructure and industrial policy.
Once AI is understood as infrastructure rather than a software race, the policy question shifts from whether to act to how to act in a time of deep uncertainty. Geopolitical, technological, and regulatory conditions will remain fluid, so policy architecture must be robust across multiple futures rather than optimized for a single trajectory.
Four scenarios frame this uncertainty (see Figure 2). In the Bipolar Dominance scenario, the US and China shape standards, requiring others to secure access while building sovereign fallback capacity. In an Open AI Multiverse, open ecosystems expand participation, shifting focus to domestic capability, talent, and interoperability. In a Fragmented AI World, divergent rules create silos, leading to an emphasis on AI diplomacy and trusted data channels. Finally, in a Regulation Reset, strict rules slow deployment but reward nations that lead in safe, compliant AI. No scenario will emerge in pure form, but each highlights vulnerabilities that a resilient strategy must address. Across these pathways, nations need a policy architecture that remains robust as market structure, alliances, and regulatory intensity change.
Three durable moves can help sequence investment, reduce dependency, and preserve room for strategic choice across all four scenarios:
These are long-term strategies that require steady investment, strong institutions, and global engagement. Each nation’s vision, regulation, and diplomacy will define its place in an AI-driven world.
Countries are refining AI strategies, funding initiatives, and forming partnerships, but without a structured diagnostic, these efforts risk misallocating resources toward visible activities like talent programs, research grants, and regulatory commissions, while leaving critical gaps in the underlying foundations of scale. ADL’s CRAFT framework translates strategic intent into a practical tool to identify gaps, sequence investment, and set national priorities.
Regulation remains a consistent driver of competitiveness, shaping market access, standards, and trust across all futures. Compute becomes dominant in fragmented or bloc-driven environments, while algorithms and talent are more decisive in open ecosystems where infrastructure constraints are lower.
CRAFT is best as a diagnostic tool before it becomes a strategy. It forces a focus on capability gaps rather than aspirational priorities, enabling more disciplined sequencing of national investment.
AI sovereignty is not a theoretical concept. Rather, it is a practical capacity to govern, build, procure, and deploy AI in line with national law and strategic priorities, even under external pressure. This requires five operational measures working in tandem:
Without these disciplines, even advanced AI adoption can quietly deepen external dependency. A nation that deploys state-of-the-art AI across its public services but relies entirely on foreign chips, clouds, and models has not built capability — it has built a sophisticated dependency. The test of AI sovereignty is not whether you have a sovereignty policy; it is whether you could maintain AI operations if your primary foreign technology provider withdrew access tomorrow.
Domestic capability is insufficient for nations that want to shape rather than simply respond to the global AI environment. Countries need a voice in how AI norms are set across safety, interoperability, transparency, and energy-aware scaling. Last year’s AI Action Summit in Paris reinforced this point: coordination now matters not only for ethics and human rights, but also for competitiveness, inclusion, and the ability to access global AI infrastructure on acceptable terms. With more than 100 governments participating and substantive agreements reached on frontier AI safety, open source governance, and compute access, nonparticipation effectively means accepting outcomes negotiated by others.
Nations that do not engage in AI rule-setting may still adopt AI, but they will do so on terms set by the US, China, or the EU — terms that reflect those actors’ interests rather than their own. Those that shape standards will find it significantly easier to build compliant, trusted, and interoperable systems. Those that inherit standards designed elsewhere will experience permanent structural disadvantages. Diplomatic presence in AI governance forums is not a soft-power nicety but a hard-capability investment. Nations absent from rule-setting will pay a compliance and interoperability premium for decades.
What began as a race to build better models has become a contest over the conditions under which AI can scale, be governed, and generate national advantage. The advantage will accrue not to those who innovate fastest in isolation, but to those who can align technology, infrastructure, and policy into a coherent system of control. To win, nations must understand and act on four items:
By Trung Ghi, Abhishek Srivastava, Ir. Syed Fazal, Julian Yeo
cloud technology axon
Headquartered in France with an established footprint across the Unite...
