Cheap Power, New Risks: What the Hormuz War Means for the Gulf's AI Ambitions

For three years the Gulf states have pitched themselves as the natural home for the next generation of AI infrastructure. The argument rested on three pillars: abundant and cheap energy, deep sovereign capital, and political alignment with Washington. The war that closed the Strait of Hormuz at the start of March has tested all three, and it has added a fourth consideration that investors had largely ignored: physical security. On 1 March, drones struck three Amazon Web Services data centre facilities, two in the United Arab Emirates and one in Bahrain.

According to AGBI, the strikes disrupted services for customers for at least 48 hours. Before the war, security was not a major consideration for Gulf data centre investors, according to Arthur Atkinson of the digital infrastructure adviser Teledata, quoted by AGBI. The Guardian quoted the observation that defending data centres now means missile defence. For governments that have promoted AI as the successor to oil, that is an awkward message, because it puts digital infrastructure in the same category of strategic target as refineries and export terminals.

Rest of World points out that about 17 submarine cables pass through the Red Sea, carrying the vast majority of data traffic between Europe, Asia and Africa, and that additional cables run through Hormuz to serve Iran, Iraq, Kuwait, Bahrain and Qatar. The UAE has more cable diversity than most of its neighbours, but The Guardian notes that many landings are concentrated on the east coast at Fujairah, a partial geographic chokepoint. AGBI's reporting suggests that the near-term effect will be delay rather than cancellation.

The paradox is that the war strengthens the Gulf's core selling point even as it weakens others. The key signals over the coming months are whether hyperscalers proceed with announced Gulf regions, whether Stargate UAE and Saudi Arabia's Humain projects keep their schedules, and whether cable repairs and shipping normalise.