Pavel Durov Unveils ‘Cocoon’: A Confidential Compute Network on TON and Telegram

• Pavel Durov presented Cocoon, a privacy-focused AI compute layer built on TON, expected to launch in November.
• Durov described the initiative as a way to restore privacy and digital autonomy, as it challenges AWS and Azure in cloud computing.

At Blockchain Life 2025 in Dubai, Telegram founder Pavel Durov took the stage to unveil Cocoon, short for the Confidential Compute Open Network. This decentralized infrastructure is designed to bring private and secure AI computation into the blockchain space, merging artificial intelligence and decentralized technology.

Built on the TON (The Open Network) blockchain, Cocoon allows users and developers to process AI tasks confidentially. The public beta launch of Cocoon is expected in November, with registration already open for GPU providers and developers who want to contribute or build on the network.

Pavel explained:

Pavel’s Plan for Restoring Privacy

Instead of relying on traditional cloud providers like AWS or Azure, Cocoon taps into a worldwide pool of GPU owners to power AI computation. Other decentralized compute projects, including Akash and Render, have explored this model. Still, Cocoon’s connection to Telegram’s enormous user base gives it a built-in community that could accelerate the adoption of decentralized AI at scale.

This announcement followed Durov’s earlier remarks during a visit to Kazakhstan in early October, where he introduced Telegram’s new AI laboratory. There, he revealed that Telegram had been quietly developing Cocoon’s technology for several months.

The lab’s mission, he explained, is to explore confidential AI models that align with Telegram’s principles of user autonomy and data protection. He shared that third-party developers have generated over $1 billion in revenue through Telegram Mini Apps over the past year.

Cocoon is built to harness TON’s scalability, high speed, and deep integration with Telegram’s global user base, a network capable of handling over 100,000 transactions per second, outperforming Solana’s 65,000 and vastly exceeding Ethereum’s 30 transactions per second.

The project introduces Confidential AI Compute, a system that allows artificial intelligence workloads, such as model inference and data queries, to run securely without exposing private information. All data processed through Cocoon remains encrypted, meaning even the nodes performing the tasks cannot view the underlying content.

The network also creates a new earn-and-use economy around computing power: individuals or organizations with spare GPU hardware can contribute their resources to Cocoon and earn TON tokens in return. Developers and applications that require AI capabilities can pay in TON to access this decentralized compute layer, routing their tasks through Cocoon’s network of contributors.

Notably, Telegram will be Cocoon’s first client. The app will use the network to power confidential AI features such as message summarization and smart reply drafting directly within its platform.

This will help bootstrap early demand and also demonstrate how Cocoon can operate securely at scale. The project will be deeply woven into Telegram’s Mini Apps and bot ecosystem.