Panmnesia and Meta Propose CXL-Based AI Datacenter Architecture in Nature Journal

AI infrastructure startup Panmnesia and global hyperscaler Meta have jointly proposed a next-generation AI datacenter architecture that aims to operate an entire datacenter as a single chip. Published as a review article in Nature, this proposal addresses the challenges of scaling AI models to trillion-parameter scales, focusing on efficient connection of numerous accelerators and memory, and exploring solutions like CXL-over-Optics to extend stable signal reach across an entire datacenter.