OpenAI launched its first model on non-Nvidia hardware in February, slashing AI coding response times from seconds to milliseconds — and in less than five months, that experiment has produced a ...