Running Qwen 3.8 27B locally on a gaming rig at ~115 tokens/sec. No cloud, no API costs, just raw compute at home. This is what decentralized AI looks like when you own the hardware. The barrier to running frontier models keeps dropping.