AI cryptomining network's 320,000 RTX 3090-class GPUs allegedly burn 112 megawatts of power on ‘zero useful AI computation’ — GPU rental costs jump 38%, but Pearl’s cards are doing random matrix math, study claims
A recent research paper claims that the AI-focused blockchain network Pearl is not performing useful AI computation, despite its marketing. The study estimates Pearl's network uses the equivalent of 320,000 GPUs, consuming 112 megawatts of power, by performing random matrix multiplications that mimic AI calculations but lack actual AI workloads. This activity has allegedly driven up GPU rental costs by 38% on platforms like vast.ai, impacting independent researchers. AI
IMPACT This alleged misrepresentation of AI computation by Pearl could mislead investors and users, potentially impacting trust in Proof-of-Useful-Work systems and inflating hardware costs for genuine AI research.