Pearl Integrates AI with GPU Mining via Proof-of-Useful-Work

Pearl has proposed a 'Proof-of-Useful-Work' (PoUW) structure that integrates AI computations into GPU mining. The focus is on connecting calculations performed during mining with customer-facing AI training and inference. Pearl Research Labs describes Pearl as a Layer 1 blockchain utilizing matrix multiplication for mining, a core operation repeatedly used in neural network learning and inference.
Unlike traditional Proof-of-Work (PoW), which performs hash calculations for blockchain security, PoUW is designed so that GPUs perform matrix multiplications while handling both network security and AI-related computations. Pearl refers to this as a '2-for-1' structure, claiming additional computational overhead remains at the '1+o(1)' level.
The open-source code includes full nodes, wallets, zero-knowledge proof systems, and GPU mining infrastructure. The network uses a UTXO-based structure similar to Bitcoin, leveraging matrix multiplication and verifiable proofs during mining. Together AI announced collaboration with Pearl Research Labs on May 15, introducing the 'Gemma-4-31B-it-pearl' inference endpoint at prices over 25% lower than standard rates.
The discount is offset by the value of PRL tokens issued during AI inference processes. Together AI positions this as the first product integrated with Pearl Network. Pearl Research Labs reported PoUW kernel overhead for Llama 70B and DeepSeek V3.2 at 5.08% and 3.9%, respectively, based on their own measurements.
The key debate centers on defining and verifying 'useful computations.' Independent researcher Abhinaba Basu published a June 3 analysis of Pearl's verification method, examining the network's computational capabilities and AI inference structure. His study is a preprint, highlighting that distinguishing between mining structures capable of performing AI computations and those forcing miners to execute paid AI tasks remains critical.
Together AI’s discounted service demonstrates potential ecosystem integration but does not prove the entire network performs customer inference. The broader issue in AI infrastructure lies in verifying computation sources and methods rather than just volume. Pearl’s PoUW structure will be judged by whether customers are compelled to use submitted matrices, whether work sources can be verified, and if paid inference demand persists.
Pearl has publicly disclosed its technical architecture through whitepapers and open-source code, but the actual proportion of customer tasks in the network requires further verification. Discrepancies between self-measured performance and external analysis mean reproducibility of validation procedures remains a key issue. Whether Pearl’s PoUW successfully implements GPU mining connected to AI services depends on additional disclosure regarding mandatory customer task usage and verification methods.
Korean Source
This article is an English localization of a Korean-language crypto news report. Original headline: 펄, GPU 채굴에 AI 결합…유용한 연산 검증 쟁점