A cheap $800 computer can run a 120 billion parameter AI model, GPT-OSS-120B, locally at over 10 tokens per second, utilizing a decent CPU and fast DDR5 memory.