AI Is Now Within Cannon Range

Strike and outage

It has been reported that on March 1 Iranian missiles and drones struck targets across the Gulf, and that one munition landed on an Amazon data center in the United Arab Emirates. The timing was notable: Claude, the large AI assistant from Anthropic, which reportedly runs significant workloads on Amazon Web Services, experienced a global outage the same day. Social posts and a new Polymarket market quickly lit up with users asking one blunt question: when will it go down again?

Military use and irony

Two days earlier it has been reported that a U.S.–Israel airstrike killed Iran’s supreme leader and senior officials — a strike for which U.S. forces reportedly used AI tools, including Claude, integrated into military workflows via partnerships such as the one between defense actors and the analytics firm Palantir. Irony followed fast. It has been reported that the U.S. administration had just ordered restrictions on Anthropic over military access; but bans or not, the model was reportedly embedded in targeting chains before any policy could take full effect.

Geography, chips and legal grey zones

Cloud companies have been racing to place capacity in the Gulf — Amazon has facilities in the UAE and Bahrain and planned investment in Saudi Arabia; Microsoft runs nodes in the UAE and Qatar; and it has been reported that OpenAI, Nvidia and SoftBank are partnering on a more-than-$30 billion AI campus in the Emirates. The U.S. recently signed a so-called “Pax Silica” deal with the UAE and Qatar to ease export of high-end chips, and it has been reported that regional players reshuffled supplier links to comply. But physical facts don’t change: code runs on chips, chips sit in racks, racks are in buildings with coordinates. There are roughly 1,300 hyperscale data centers worldwide and about 770 more under construction — and in a region where missiles fly, those buildings are no longer neutral.

The new question of protection

When AI is used as part of national intelligence or targeting, adversaries have little incentive to treat the hosting facility as purely civilian. Existing laws on “dual-use” infrastructure were written for factories and bridges, not data centers that hold citizens’ bank records, medical files and, potentially, battlefield targeting models. Who should protect that infrastructure — cloud providers, the U.S. Department of Defense, host-state air defenses? If AI has become national infrastructure, its security is no longer only a corporate problem.