When Iran's Islamic Revolutionary Guard Corps released satellite imagery of OpenAI's USD 30 billion Stargate data centre in Abu Dhabi on April 4, accompanied by a threat of complete and utter annihilation, the message was aimed at more than one facility. It was a declaration that AI infrastructure has become a frontline target in modern geopolitical conflict, with consequences that ripple far beyond the Gulf. For Asian policy makers, enterprise AI customers, and data centre operators, the Stargate threat has accelerated a reassessment of physical infrastructure risk that had previously been treated as a secondary consideration behind cyber and data concerns.
The IRGC video, narrated by Brigadier General Ebrahim Zolfaghari, displayed high-resolution imagery of the facility's desert location, which the IRGC pointedly noted is absent from Google Maps. The video named the American corporate partners behind the project: OpenAI, Nvidia, Microsoft, Oracle, and Goldman Sachs. The threat was specific, deliberate, and designed to signal that Iran views the computational backbone of American AI as a legitimate military target. Whether the threat is credible or symbolic, the fact that AI infrastructure is now explicitly in military messaging represents a new phase in how geopolitics and AI interact.
The kinetic precedent that changed the calculation
This was not an isolated provocation. It followed actual kinetic strikes against cloud infrastructure in the region. On March 1, Iranian drones struck three Amazon Web Services facilities across the UAE and Bahrain, knocking them offline and triggering cascading service outages that disrupted banking, payments, ride-hailing, and enterprise software across the Gulf. Abu Dhabi Commercial Bank, Emirates NBD, payments platform Hubpay, and ride-hailing giant Careem all reported significant disruptions. The attacks marked the first known deliberate targeting of data centres in armed conflict.
The significance of the March strikes was not the physical damage alone but what the strikes revealed about the vulnerability of cloud infrastructure to precision kinetic attack. Data centres are typically designed against natural disasters, cyber attack, and infrastructure failure, but resilience against drones, missiles, or other military strike modalities has not been a mainstream design requirement. The attacks forced a rapid reassessment across the industry.
The International Institute for Strategic Studies has published analysis indicating that approximately 40 percent of major data centres globally are in locations considered potentially exposed to kinetic threat scenarios in worst-case geopolitical conflicts. Middle Eastern, Eastern European, East Asian, and certain Latin American locations all appear in the exposed category.
The Stargate project and what is at stake
Stargate is OpenAI's flagship AI supercomputer project, a joint initiative with Oracle, Softbank, and Microsoft that commits USD 500 billion over five years to AI infrastructure development. The Abu Dhabi facility is one of several planned Stargate sites, with additional facilities in Texas, Ohio, and planned locations elsewhere.
For OpenAI, the Abu Dhabi facility is particularly important because of UAE regulatory clarity, favourable energy pricing, and geographic positioning between European and Asian markets. The UAE government has been aggressive in positioning itself as a global AI hub, with partnerships including G42's acquisition of a stake in Cerebras and joint ventures with US and Chinese AI firms.
The IRGC threat complicates this strategy. Insurance premiums for data centre facilities in the Gulf region have reportedly risen by 40 to 60 percent since the March AWS strikes. Some institutional customers have begun requiring data residency outside the Gulf for sensitive workloads, which affects the commercial viability of Gulf AI infrastructure investments.
What this means for Asian AI infrastructure
Asian AI infrastructure operators face similar questions. Singapore's position as a major data centre hub has been built on political stability, strong rule of law, and excellent connectivity. Taiwan's semiconductor and AI infrastructure remains exposed to potential conflict with China. South Korean infrastructure is physically close to potential North Korean kinetic capability. Japanese facilities are generally considered secure but still face potential regional conflict exposure.
Southeast Asian markets vary. Malaysia and Indonesia are generally considered geopolitically stable for infrastructure purposes. Vietnam's location on the South China Sea creates specific exposure to potential regional maritime conflict. The Philippines faces similar concerns. Thailand is generally stable but not insulated from regional dynamics.
Chinese AI infrastructure is essentially inaccessible to US firms because of regulatory restrictions, but for Chinese companies, domestic infrastructure is secure against external threats even as it faces US export controls on advanced chips. The Center for Strategic and International Studies has documented how Chinese firms have responded to compute constraints through efficient training and mixed hardware approaches.
Insurance and risk markets adapt
The insurance industry has begun pricing data centre risk differently. Lloyd's of London and specialist providers including Allianz Global Corporate and Specialty have introduced new policy categories for AI infrastructure that include kinetic risk coverage. Premiums for facilities in higher-risk geographies have risen substantially. Cyber insurance policies are being restructured to account for physical destruction scenarios that were previously treated as uninsurable.
Institutional investors in data centre REITs have begun requiring geographic diversification as a condition of investment. Equinix, Digital Realty Trust, and regional operators including Keppel DC REIT and Digital Core REIT have all discussed risk diversification strategies in recent investor communications.
For corporate customers, the practical response has been increased redundancy, including deploying AI workloads across multiple regions and multiple providers. Single-region deployments, which were common when the primary risk models assumed natural disaster or equipment failure, are now understood as inadequate protection against geopolitical risk.
Policy and strategic implications
National governments have begun treating AI infrastructure as critical infrastructure in formal policy. Singapore's Cybersecurity Agency has expanded its critical infrastructure definitions to explicitly include AI compute facilities. Japan's National Center of Incident Readiness and Strategy for Cybersecurity has issued similar guidance. Australia's Critical Infrastructure Centre has added AI infrastructure to its protected sectors.
The policy implications extend to international coordination. NATO's strategic communications work now includes AI infrastructure as potential conflict target. Asian security partnerships including the Quad (Australia, India, Japan, US) have begun discussing AI infrastructure security as a coordination priority. ASEAN's approach has been more cautious, reflecting the grouping's traditional aversion to explicit security alignment.
US export controls on AI chips to certain countries, which had been controversial on economic grounds, are now more widely accepted as a strategic necessity by policy makers who recognise that AI infrastructure in adversary hands could become a military asset. This shift in framing will likely strengthen rather than weaken export control regimes over the next several years.
What enterprise customers should do
For Asian enterprise customers of AI services, the practical response involves several elements. First, geographic diversification of AI workloads should be considered where feasible. Using multiple cloud regions for the same workload provides insurance against localised disruption, whether from kinetic attack, natural disaster, or other causes.
Second, provider diversification reduces dependency on any single AI provider. Customers who have standardised entirely on OpenAI, Anthropic, or a single Chinese provider should evaluate having credible alternatives ready to deploy if circumstances change suddenly.
Third, sensitive workloads may justify on-premises or sovereign cloud deployment rather than commercial hyperscaler deployment. Banking, government, and critical infrastructure customers have been moving in this direction for several years, and the March AWS strikes have accelerated the shift. Gartner's enterprise AI research has documented the growing move toward sovereign and on-premises AI infrastructure for high-sensitivity workloads.
The Stargate threat matters beyond the specific facility and specific threat actor involved. It marks the formal recognition that AI infrastructure is now an element of national power and a potential target in geopolitical conflict. How Asian governments, enterprises, and AI providers respond to this new reality will shape the regional AI landscape for the remainder of the decade. The assumption that AI infrastructure investment decisions could be made on purely economic grounds, ignoring geopolitical risk, has now ended definitively. Every significant AI infrastructure investment from this point forward will need to account for physical security risk as a core design consideration.
Comments
Sign in to join the conversation. Be civil, be specific, link your sources.