Image Courtesy: @I_Leak_VN via X.com
A blockchain network marketed as transforming cryptocurrency mining into useful artificial intelligence work is facing scrutiny after a new study claimed its massive computing operation generates no meaningful AI output despite consuming enormous amounts of power.
The research focuses on Pearl, a Layer-1 blockchain that promotes its mining system as performing AI-related computation instead of traditional crypto hashing. Researchers estimate the network operates at roughly 24 exahashes per second, equivalent to around 320,000 Nvidia RTX 3090-class GPUs consuming approximately 112 megawatts of electricity. The findings come from a preprint study examining the platform’s so-called Proof-of-Useful-Work system.
The study argues that Pearl’s miners are not performing genuine AI training or inference tasks. Instead, they reportedly process random matrix multiplications that resemble the mathematical operations used in machine learning but serve no practical purpose. While the network verifies that these calculations are completed correctly, researchers claim it cannot determine whether the computations originated from a real AI workload.
To test the system, researcher Abhinaba Basu created a mining setup that submitted randomly generated matrices without any associated AI task. The network reportedly accepted the work and rewarded it in the same way it would legitimate AI-related computations, suggesting miners have little incentive to perform useful processing.
The report also examined more than 8,000 mining workers and found no clear evidence that they were running machine-learning frameworks. Runtime analysis indicated workloads consistent with pure mathematical calculations rather than the memory-intensive behavior typically associated with modern AI models.
Beyond the technical concerns, the study suggests Pearl’s popularity has affected GPU availability. Researchers estimate that rental prices for lower-cost GPUs on the marketplace Vast.ai rose by about 38% after Pearl’s mining software became publicly available in May. Utilization rates reportedly climbed from 57% to 94%, increasing costs for researchers and developers seeking affordable computing resources.
The study does not argue that Proof-of-Useful-Work systems are fundamentally impossible. Instead, it concludes that Pearl’s current design lacks a reliable mechanism to ensure miners perform genuinely useful AI tasks. Pearl and its partners have promoted the network as a way to combine AI and cryptocurrency mining, but researchers contend the evidence currently points to large-scale computation that looks like AI work without delivering meaningful AI results.
