Study Says Pearl Burns 112 MW Across 320,000 GPUs for Zero Useful AI Work
Updated
Updated · Tom's Hardware · Jun 14
Study Says Pearl Burns 112 MW Across 320,000 GPUs for Zero Useful AI Work
2 articles · Updated · Tom's Hardware · Jun 14
Summary
A new preprint estimates Pearl’s network is running at about 24 EH/s—the equivalent of 320,000 RTX 3090-class GPUs—while producing no useful AI computation despite its Proof-of-Useful-Work pitch.
Researcher Abhinaba Basu says Pearl’s cuPOW protocol verifies only that matrix multiplications were done correctly, not that they came from real training or inference, allowing random inputs to earn rewards.
Basu reports 44 pool-accepted shares from a miner feeding random matrices and says analysis of 8,012 workers in one pool found no identifiable machine-learning framework code, with runtime behavior matching pure compute-heavy math rather than transformer inference.
The paper links Pearl’s May mining rollout to a 38% jump in budget GPU rental prices on vast.ai and utilization rising from 57% to 94%, adding an estimated $600,000 a year in costs for independent researchers.
That finding challenges Pearl’s claim that mining and AI can be the same job, even as partner Together AI markets subsidized inference endpoints funded by PRL mining proceeds rather than by miners performing the inference themselves.
As Pearl consumes 112 MW for no useful AI, who is profiting from the displaced research and soaring GPU costs?
If random data can fool an AI blockchain, is the entire 'Proof-of-Useful-Work' concept fundamentally flawed?
320,000 GPUs, Zero Useful AI: Cornell Study Exposes Pearl Blockchain’s Proof-of-Useful-Work Failure
Overview
In June 2026, Cornell University researchers led by Abhinaba Basu announced a study that debunked the Pearl blockchain project's core claims. Despite Pearl's heavy promotion of transforming crypto mining into valuable AI computation, the study found that the network produced zero useful AI work. This revelation directly contradicted Pearl's marketing and sparked widespread debate in the crypto community. The findings highlighted that Pearl's protocol only checked mathematical correctness, not actual AI usefulness, leading experts to question the project's legitimacy and raising concerns about wasted resources and misleading claims in the Proof-of-Useful-Work space.