Telegram founder Pavel Durov announced the launch of Cocoon, a decentralized confidential compute network designed to process AI workloads with full privacy. The platform is now live, handling its first user requests while enabling GPU owners to earn TON for contributing compute capacity. Durov said the system removes centralized intermediaries like Amazon and Microsoft, aiming to lower costs and strengthen confidentiality for developers and end users.
Cocoon operates as an open, distributed infrastructure, with documentation and source code made publicly available. The rollout marks Telegram’s latest expansion into AI services, with plans to integrate new privacy-focused features directly into its ecosystem.
Over the coming weeks, the network will expand GPU supply and onboard additional developers to scale demand. Durov emphasized that Cocoon is built to return control and data privacy to users, positioning it as an alternative to traditional cloud providers in the rapidly growing AI compute market.