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 operates an entire datacenter as a single chip. Published as a review article in Nature, this architecture, based on CXL technology, aims to address the challenges of scaling AI models to trillion-parameter scales by connecting numerous accelerators and memory quickly and efficiently. The proposal also outlines methods to prevent unintended data mixing or exposure across multiple concurrent AI workloads.