GAIA: GPU Power Attacks for AI Data Centers

GAIA: GPU Microarchitecture-Based Power Attacks and Countermeasures for AI Data Centers

Project Summary

AI data centers rely heavily on GPU clusters for training and inference of large models. This project investigates microarchitecture-level power side-channel attacks targeting GPU hardware in AI data center environments, and develops corresponding countermeasures.

The GAIA project aims to:

  • Characterize power consumption patterns exposed by GPU microarchitectural features during ML workloads
  • Develop practical side-channel attack methodologies that can extract sensitive model weights or intermediate data
  • Design and evaluate hardware and software countermeasures to mitigate these vulnerabilities

Research Team

  • PI: Dr. Arafin, George Mason University
  • Co-PI: Khaled Khasawneh, George Mason University

Funding

This project is supported by the Virginia Innovation Partnership Authority (Award #N-1Q27-004).

← Back to Grants