Moonshot AI and Tsinghua propose PrfaaS, a cross-datacenter LLM architecture separating prefill and decode across GPU cl...

Moonshot AI and Tsinghua propose PrfaaS, a cross-datacenter LLM architecture separating prefill and decode across GPU clusters. By using commodity Ethernet for KVCache transfer instead of RDMA, it achieved 54% higher throughput in tests. https://www.marktechpost.com/2026/04/19/moonshot-ai-and-tsinghua-researchers-propose-prfaas-a-cross-datacenter-kvcache-architecture-that-rethinks-how-llms-are-served-at-scale/ #AIagent #AI #GenAI #AIInfrastructure