Current datacenter GPUs are optimized for LLM compute, not for real-time rendering. The economics for running such beefy GPUs just for game streaming won't add up.