On the evening of March 1, a drone strike on an Amazon Web Services facility in the United Arab Emirates cascaded into a global outage that read like a parable of modern dependency. The attack—widely reported as targeting AWS servers in the UAE’s mec1-az2 availability zone—knocked Claude, the AI assistant that relies heavily on AWS, offline. Users swarmed alternative services; surge traffic overwhelmed Google’s Gemini infrastructure, producing further disruption and a short-lived scramble for capacity.
The immediate drama masked a deeper reality: large-scale AI is now inseparable from physical infrastructure and energy flows. Cloud-hosted models are sustained by data centres, power plants and fibre routes that can be damaged or disrupted by conventional military action. The Middle East, with its cheap electricity and strategic geography, has been positioning itself as the next global hub for that infrastructure. Governments and sovereign-backed firms are promoting data-centre builds as a way to convert hydrocarbons into a new export—compute and data—rather than crude oil.
That conversion is intentional and explicit. Saudi Arabia’s finance minister has repeatedly framed the country’s ambition not to export more barrels but to “export data.” Private and public investment has followed: consultancy tallies show dozens of operators building scores of facilities across the Gulf, with major projects clustered in Dubai, Abu Dhabi, Riyadh and Dammam. Forecasts from professional services firms foresee threefold growth in the region’s data-centre capacity within a few years, driven by demand for AI training and cloud services.
Cheap power is the obvious lure. Electricity in parts of the Gulf can cost a fraction of North American and European rates, making it economical to run large GPU farms that burn kilowatts by the second. But low operating costs come with geopolitical costs. When an adversary can reach a data centre by drone, missile or sabotage, the calculus changes: what looks like a cost advantage becomes a security liability. The March outage forced a reckoning over where to site critical compute and who should run it.
The episode also exposed tensions inside the tech ecosystem. OpenAI’s decision to take contracts with defence establishments stirred opposition from users and developers who then sought alternatives, shifting traffic in ways that stressed other platforms. That political reaction illustrates how governance, commercial ties and public trust can reshuffle technical loads overnight—magnifying single points of failure into systemic risk.
For policymakers and corporate strategists the choice is stark. Dependence on foreign-owned, geographically concentrated cloud infrastructure risks national productivity and strategic resilience. Some governments will conclude that paying higher domestic energy costs to keep compute onshore is an acceptable insurance premium. Others will double down on distributed, localised models—miniature on-premise systems that act like the ‘‘backup generators’’ of a digital economy.
The economics complicate any simple retreat. Many AI firms already operate on razor-thin margins, subsidising access while burning capital on compute. If supply chains and site selection become dominated by security considerations, the price of tokens and API calls will rise. The result could be less free access to high-quality models, a reinvigoration of paid tiers, and a bifurcation between resource-rich organisations that can afford sovereign compute and the rest of the market.
This is not merely an industry problem. The attack reframes AI as a strategic asset akin to power grids, satellites or refineries—systems that states protect and contest. As energy-rich states seek to monetise their comparative advantage by selling compute rather than crude, they will also attract geopolitically sensitive workloads and, therefore, strategic attack vectors. For the global AI ecosystem, resilience will mean both architectural decentralisation and new diplomatic bargains about the sanctity of computing infrastructure in conflict zones.