Artificial intelligence (AI) infrastructure has quietly become one of the most consequential geopolitical battlegrounds of our time. What was once a technical concern has transformed into a matter of national security, economic resilience, and strategic power.
A new report from the World Economic Forum and Bain & Company states that the infrastructure decisions governments make today will shape their economic competitiveness for decades to come.
Cumulative investment in AI-dedicated infrastructure exceeded $600 billion between 2010 and 2024 and is projected to surpass $400 billion annually by 2030. Yet the conversation around who controls that infrastructure and on what terms remains surprisingly underdeveloped in most capitals.
The Building Blocks
The WEF paper breaks AI infrastructure into three interlocking components- compute, connectivity and data storage. Compute refers to the processing power required to train and run AI models. Connectivity refers to the networks that link data centres, edge systems, and end users. Data storage refers to the systems that securely store information at scale.
Each depends on prerequisites that are far from guaranteed. Energy, water, land, hardware, and cybersecurity sit on the technical side. Policy frameworks, skilled talent, and access to capital sit on the institutional side.
Global data centre electricity consumption reached roughly 415 terawatt-hours in 2024 and could rise to 1,200 terawatt-hours by 2035, according to the International Energy Agency.
An AI training facility typically requires at least 200 acres of land. The largest sites can use up to five million gallons of water daily. The hardware supply chain is highly concentrated: a single American company leads GPU design, a Dutch firm supplies all advanced chip-making equipment, and over 90% of high-end GPUs are manufactured by a single Taiwanese producer.
Krupesh Bhat, Founder and CEO of Melento, formerly SignDesk, says governments are still underestimating GPUs because they treat them as industrial assets rather than as instruments of geopolitical leverage. “In the next decade, access to compute could matter as much as access to oil did in the twentieth century," he says.
The Sovereignty Spectrum
The WEF paper describes AI sovereignty as an economy's capacity to shape, deploy, and govern AI ecosystems in line with its values while retaining strategic control. Importantly, the authors oppose equating sovereignty with self-sufficiency. For most countries, complete domestic control is unfeasible, with only the United States and China coming close.
Most countries must instead navigate what the paper calls an AI sovereignty spectrum. At one end is the strategy of trusted international partnerships, where sovereignty is achieved through shared infrastructure and enforceable governance arrangements. At the other end is extensive domestic ownership. “In practice, most countries pursue hybrid strategies,” says Sridhar Pinnapureddy, Founder and CEO of CtrlS Datacenters.
Singapore offers an instructive example, combining extensive use of trusted commercial cloud providers, with over 70% of eligible government systems already on its Government on Commercial Cloud platform, with a domestically anchored National Supercomputing Centre that received SGD 270 million in 2024.
But Bhat cautions against the assumption that governments have genuinely internalised this logic. "Most governments still speak the language of sovereignty while operating in the reality of dependency," he tells AIM.
He, however, opines that the real geopolitical divide will not be between sovereign and non-sovereign nations, but between countries that become indispensable nodes in global AI supply chains and those that remain replaceable consumers.
“Countries that chase complete self-sufficiency risk building politically symbolic infrastructure that is economically inefficient, technologically outdated, and strategically isolated within a decade," Bhat says.
That said, Pinnapureddy notes that sovereign AI is not just about owning data centres but about building the full ecosystem, talent, policy, energy, and capital, simultaneously. Without that alignment, infrastructure investments risk becoming stranded assets rather than strategic ones.
Digital Embassies
Perhaps the most innovative aspect of the WEF paper is its approach to digital embassies—secure digital environments in a foreign country where a guest nation's data and workloads remain governed by its own laws.
Pinnapureddy recalls instances of digital embassies. Following a series of cyberattacks in 2007, Estonia signed a deal with the Luxembourg government to host its data in secure facilities. Similarly, Monaco reached an agreement as well in 2021. Recently, Saudi Arabia and Bahrain passed laws to host data of foreign parties.
The WEF paper positions digital embassies not as a niche continuity measure but as a credible option for extending sovereign AI infrastructure beyond national borders.
"The biggest obstacle is not technical capability or even legal frameworks. It is political psychology," Bhat said. "Governments are comfortable outsourcing data storage, but deeply uncomfortable outsourcing symbolic control. Digital embassies challenge a centuries-old assumption that sovereignty must remain geographically anchored."
He added that smaller nations may prove more adaptable precisely because they already understand distributed resilience. "Larger powers still think territorial control guarantees digital control. But in an AI-driven world, cyberattacks, climate disruption, and infrastructure sabotage make geographic concentration increasingly dangerous," he says.
The Resource Reckoning
Water, land, and energy are increasingly determining what can be constructed and where. These are not issues for the future — they are current challenges, already affecting site choices, permitting processes, and national energy policies.
Bhat points out that the industry's focus on AI scaling as a software story is unjustifiable and that training larger models does not necessarily divert resources from agriculture, urban development, and public utilities.
On LinkedIn, Debjani Ghosh, former President of NASSCOM, posted that India's AI infrastructure ambitions will only be credible if they are grounded in honest assessments of energy and water availability at the state level, rather than national-level projections that mask significant regional disparities.
The Cost of Delay
For mid-sized economies that have yet to define a clear AI infrastructure strategy, the window for meaningful strategic choice is narrowing. Bhat argues that the coming years will represent how countries can position themselves within emerging AI supply chains, standards alliances, and regional compute ecosystems.
“After that, infrastructure concentration may become entrenched around a handful of dominant hubs," says Bhat.
The WEF paper recommends that policy-makers segment workloads by sensitivity, treat energy and land as first-order design constraints, and build genuine exit portability into any cross-border arrangement to avoid new forms of lock-in.
Its Global Framework for Innovative and Trusted Digital Embassies, launched alongside the report, offers a principled foundation for bilateral negotiations without replacing them.
The era in which AI infrastructure could be treated as a procurement decision has passed. As Bhat put it, the mistake many governments make is assuming AI infrastructure is an IT investment. It is actually a long-term industrial policy decision that will shape productivity, geopolitical influence, and economic leverage for decades.
Your reaction
Discussion
What's your take on this story?
No comments yet — be the first to share your view.