Signal Could Underwater Data Centers Pose a Risk to AI Treaty Verification?

Summary

Proposals for international agreements limiting frontier AI development depend on effective verification, and a central challenge is detecting undeclared compute facilities used to evade restrictions. Underwater data centers have been floated as one such evasion route, since their physical concealment could complicate monitoring and inspection regimes. The paper examines whether underwater data centers at a scale comparable to frontier AI compute are actually feasible. It then assesses whether this scenario could pose a genuine risk to future AI treaty verification efforts. The authors argue this kind of assessment should be factored in from the treaty-design stage.