Pearl, an L1 blockchain with 320,000 GPUs, is pricing compute through proof-of-work

• —Pearl inherits Bitcoin's PoW and difficulty adjustment but miners perform matrix multiplications instead of SHA-256 hashes
• —NoisyGEMM injects structured noise into matrices so miners cannot shortcut the work and must compute honestly
• —Difficulty retargets every 194 seconds, reflecting real GPU throughput under current power, cooling and network conditions
• —Together AI launched a Gemma-4-31B-it-Pearl inference endpoint on May 15, 2026, priced over 25% below standard cloud rates