Study Claims AI Cryptomining Network Uses 320,000 RTX 3090 GPUs for 'Random Matrix Math'
A recent study has brought attention to what researchers describe as an AI cryptomining network operating with approximately 320,000 RTX 3090-class graphics cards. According to the findings, these GPUs allegedly consume 112 megawatts of power while performing what the study characterizes as "random matrix math"—calculations that reportedly produce zero useful AI computation.
The network, attributed to a company called Pearl, allegedly rents out GPU capacity that operates on mathematical operations that do not contribute to any meaningful AI training or inference tasks. The study suggests this represents a significant drain on available GPU resources, which has contributed to a notable increase in GPU rental prices.
The findings indicate that GPU rental costs have risen by approximately 38%, potentially driven in part by the demand from such operations. The study raises questions about the efficiency and purpose of large-scale GPU deployments that do not contribute to productive AI work.