New Bit2Watt Attack Could Let Cloud Tenants Disrupt Power Grids Without an Exploit

A cloud tenant using nothing but ordinary GPU access can push a data center's power draw up and down fast enough to threaten the grid it runs on, with no exploit and no break-in. That is the claim behind Bit2Watt, described by three Zhejiang University researchers in a paper accepted to CHES 2026, the IACR's hardware-security conference, and the evidence splits in two: they measured the power

The Bit2Watt research describes a purely theoretical, yet plausible, cyber‑physical attack where a malicious but legitimate cloud tenant modulates ordinary GPU workloads to create high‑frequency power draw fluctuations that can destabilize local, renewable‑heavy power grids, without exploiting any software vulnerability or breaking into infrastructure[3][5][6]. The paper’s proof‑of‑concept suggests that coordinating around 1,000 GPUs on a 1 MW grid can significantly increase harmonic distortion and heat, potentially degrading damping and risking cascading failures or blackouts[4][5][8][10]. From a RealGround perspective, this expands the AI supply chain risk surface: AI and GPU workload patterns themselves become a grid‑scale threat vector, requiring cross‑layer defenses that integrate workload scheduling, power‑quality monitoring, and coordination between cloud providers and grid operators[4][5][6][9]. Practically, organizations operating AI data centers should treat GPU scheduling and tenant controls as critical cyber‑physical controls, subject them to continuous red teaming for malicious load patterns, and fold these scenarios into AI supply chain and SBOM-style risk assessm