Extending Meta to the Public Cloud | Dinesh Govindasamy and Sargun Dhillon from Meta

Meta transitioned from a strictly on-prem infrastructure model to a hybrid approach by integrating public cloud capacity to meet the surging demand for AI and LLM workloads. This shift required overcoming significant security and reliability hurdles, leading to the development of the off-net services stack and Transparent TLS (TTLS) to maintain Meta’s rigorous security standards in untrusted cloud environments. To ensure operational consistency, the team implemented the Public Cloud Manager (PCM), which abstracts cloud-specific workflows and integrates them into Meta’s existing fleet health and maintenance systems. By engaging in early-stage co-design with cloud vendors, Meta now treats cloud capacity as a seamless extension of its own infrastructure, enabling rapid onboarding of new regions and hardware SKUs while maintaining the same performance and reliability expectations as their internal data centers.