Researchers identify Bit2Watt attack capable of destabilizing electrical power grids

The record

Written from the 1 report below. Nothing here is unsourced.

Researchers at Zhejiang University have described a theoretical attack called Bit2Watt, in which a cloud tenant could use ordinary GPU workloads to rapidly modulate a data center's power draw without any exploit, malware, or stolen credentials.
They measured the power modulation on real GPUs and used simulations to model how synchronized modulation from many GPUs could destabilize a grid, with a worst-case simulated 1 MW grid showing current distortion above recommended guidelines and an unstable damping ratio.
No production systems were attacked and no flaw in a specific commercial product was disclosed, and the paper itself concedes that synchronizing power transitions across a real fleet of cloud GPUs remains an open problem.
The finding matters because AI training loads are growing fast, and industry and grid regulators have already warned that large synchronized loads can affect power infrastructure.

Any response from the Zhejiang University researchers on how far the attack scales in a real cloud environment, which The Hacker News has requested.
Whether cloud providers develop ways to fingerprint purpose-built modulation kernels or unusual training-job power patterns like those described.
Outcomes from NERC's Large Loads Task Force on how growing data-center and AI-training loads are managed on the grid.

1 report

English national1

All filed from India

Named China · A100 · Nvidia · RTX 4090 · Tesla V100 · CHES 2026 · Microsoft · NERC · OpenAI · Zhejiang University

The 1 report is listed beside the record.

Answers cite the reports above, or say they can't.