Thread Guy

GoodAlexander × David Choi

Chinese open-source models reportedly approach Opus or Sonnet performance on roughly $10,000 GPUs, while Qwen 3.6-class compute costs about $2 per hour.That 40-80x gap supports local inference and crypto-financed compute, but USD.AI’s $1 billion GPU-loan target, reference rate, and CHIP model remain unproven.