Durov Unveils Cocoon: Decentralized AI Compute Network Built on TON
Telegram founder Pavel Durov has announced the launch of Cocoon, a decentralized AI compute network built on The Open Network (TON). Introduced during the Blockchain Life 2025 conference in Dubai on October 29, 2025, Cocoon aims to make artificial intelligence more accessible, private, and efficient by leveraging the principles of blockchain technology.
Cocoon, short for Confidential Compute Open Network, enables GPU owners to contribute their computing resources to a decentralized network in platform for TON tokens, the native cryptocurrency of the TON blockchain. At the identical time, AI developers can access affordable, privacy-focused computational power to train and deploy models. The platform integrates encryption at every stage, ensuring data confidentiality during AI processing.
A new model for privacy-preserving AI
Cocoon’s architecture introduces a novel confidential compute layer, allowing AI workloads to be processed securely without exposing the underlying data. This feature directly addresses one of the most pressing issues in AI—data privacy. While traditional AI cloud services often rely on centralized systems that can access user data, Cocoon’s design prevents operators from viewing or extracting the content of AI requests. The result is a fully encrypted environment for AI computation.
Durov emphasized that the goal of Cocoon is to merge the privacy guarantees of blockchain with the computational capabilities of distributed GPUs. The project’s slogan, “Private AI on TON,” reflects its mission to give developers and enterprises control over their data while reducing dependency on centralized cloud providers like Google, Amazon, or OpenAI.
Integration and ecosystem adoption
Telegram, which has more than 900 million global users, will serve as Cocoon’s first major client. The messaging platform plans to integrate Cocoon’s compute layer for selected AI-driven features, demonstrating the scalability and security of the system in a live production environment. According to Durov, this integration will assist validate Cocoon’s potential to support large-scale applications that rely on AI processing.
The public beta of Cocoon is scheduled for November 2025. Several GPU infrastructure providers and TON ecosystem partners have already joined the initiative, committing hardware and network capacity to the project. Participants will be rewarded with TON tokens based on verified compute contributions and network reliability.
Cocoon’s introduction signals a growing convergence between decentralized computing and artificial intelligence. By tokenizing compute power, TON creates a marketplace where AI demand meets distributed supply, potentially lowering costs and promoting transparency. This approach could reshape how AI models are trained, tested, and deployed across industries.
Industry observers view Cocoon as a strategic expansion of TON’s ecosystem, positioning it as a competitive layer for AI infrastructure. The combination of decentralized incentives, encrypted computation, and blockchain accountability presents an alternative to centralized AI infrastructure dominated by major tech firms.
As Cocoon prepares for its public beta, the project represents Durov’s latest effort to extend Telegram’s influence beyond messaging and into the emerging landscape of decentralized AI and blockchain-powered compute networks. With its focus on privacy, scalability, and open participation, Cocoon could become a foundational pillar for the next generation of AI applications built on TON.
