Iran threatened OpenAI's $30B Stargate facility in Abu Dhabi
A video with satellite imagery targeted the 1GW site. Concentrating a decade of compute into a handful of identifiable campuses creates an exposure no financial structure addresses.
Iran has posted a video containing satellite imagery threatening "complete and utter annihilation" of OpenAI's Stargate data centre in Abu Dhabi, a $30 billion facility described as the company's premier one-gigawatt site, according to Tom's Hardware.
A propaganda video is not a military operation, and the appropriate response is not alarm. What it does is make explicit a category of risk that the AI infrastructure build has not priced: these facilities are large, fixed, publicly identified and increasingly strategic.
Concentration is the design
The economics of frontier training push towards concentration. Training runs need accelerators connected with high-bandwidth, low-latency interconnect, which means physical proximity. Distributing a training cluster across sites imposes costs that defeat the purpose.
So the industry is building a small number of enormous campuses, each publicly announced, each in a known location, each drawing a gigawatt or more.
OpenAI has a partnership to deploy at least 10GW of Nvidia systems, and SB Energy granted it $5.5 billion in warrants tied to a 20-year lease on a planned 10GW Ohio campus. Five gigawatt-scale AI data centres are expected online this year. TCS is committing about $7.4 billion to a one-gigawatt campus in southern India. DeepSeek is planning more than 160,000 Huawei accelerators at a one-gigawatt site in Inner Mongolia. South Korea has committed $919 billion towards 8.4GW by 2029.
Every one of those is a named, mapped, immovable asset.
Why the Gulf specifically
Abu Dhabi is attractive for the reasons that make it exposed. Cheap energy, sovereign capital willing to commit at scale, political stability, and a government that wants the strategic position — MBZUAI's Institute of Foundation Models published six open-weight K2 Horizon models this month with weights, code, training data and methods.
It is also in a region with active state-level hostilities, and a facility that a foreign power considers a legitimate strategic target is a different asset from one in Ohio.
The exposure nobody has priced
The financial structures assembled around AI infrastructure this year are elaborate and all concern counterparty risk rather than physical risk.
Nvidia is investing in the compute providers, leasing the buildings and supplying the chips. SB Energy granted its tenant warrants before its own IPO. Anthropic has roughly $90 billion in compute obligations across Nscale, Lambda and Volta. Big Tech booked more than $160 billion in second-quarter gains on AI stakes.
None of that addresses what happens if a gigawatt campus is destroyed, or made unusable, or simply becomes uninsurable. The insurance market for facilities of this value in regions of this profile is thin, and the concentration means a single loss is not diversifiable.
The adjacent evidence
The physical and digital exposures are converging. Ukraine attributed a fatal autonomous drone strike to a Russian Molniya drone using an Nvidia Jetson Orin module for target selection. TeamT5 attributed a more than doubling in Chinese state-linked attack volume to AI-assisted reconnaissance. Taiwanese prosecutors raided an Nvidia PCB supplier over origin fraud allegations, and indicted nine people over smuggled servers.
AI infrastructure has become strategic infrastructure, and strategic infrastructure attracts state attention of every kind.
OpenAI has not commented on the threat, and has not disclosed what physical security or insurance arrangements the Abu Dhabi facility carries.
Runs the newsroom. Rename this profile in the studio to your own byline.
Related
Every weekday, the AI stories that moved money or shipped code.
No cross-posting, unsubscribe anytime. See all newsletters