Cocoon: Durov’s New Decentralized AI Platform on TON

Pavel Durov, the founder of Telegram, has launched a new project named Cocoon — short for Confidential Compute Open Network — a decentralized AI-compute network built on TON (The Open Network). The platform aims to offer privacy-preserving AI services by combining blockchain, distributed GPU computing, and confidential execution environments.

What Is Cocoon

Cocoon is a network that allows owners of graphics processing units (GPUs) — from individual users to data centers — to offer their computing power to process AI tasks. In return, they are rewarded in TON cryptocurrency.
The platform is built to ensure data privacy: AI tasks run inside trusted execution environments (TEEs), meaning that even nodes processing the data cannot view user inputs or outputs. This is intended to guarantee confidentiality and protection against misuse of personal or sensitive data.
Developers and applications needing AI compute can pay with TON and access the network’s distributed capacity, benefiting from a decentralized alternative to centralized cloud-AI services.

Why Cocoon Is Being Built

According to Durov, the growing popularity of AI models has led many users to surrender increasing amounts of personal data to centralized AI providers — often more than necessary for the task. Cocoon is designed to counter this: by decentralizing AI compute and building privacy protections from the ground up, it aims to give users an AI option that doesn’t compromise their digital freedom.

Moreover, Cocoon offers a new economic model: individuals with spare GPU capacity can monetize their hardware by joining the network and earning TON tokens. This helps decentralize AI infrastructure and reduce reliance on large cloud providers seen by Durov as “expensive intermediaries.”

Launch and Early Deployment

The project was first announced at the Blockchain Life 2025 conference in Dubai on October 29, 2025.
Cocoon officially went live on November 30, 2025. According to Durov, the “first AI requests from users are now being processed” with full confidentiality, and GPU owners have already started earning TON in exchange for their computing power.
The platform is now open for GPU owners and developers who want to contribute computing power, run AI models, or integrate their apps with Cocoon.

What Cocoon Could Mean for the Future of AI & Privacy

Decentralized AI alternative: By leveraging distributed GPU resources and blockchain infrastructure, Cocoon could reduce dependence on major cloud-AI providers, potentially lowering costs and democratizing access to AI compute.
Enhanced privacy for users: Confidential compute and encryption mean user data — prompts, inputs, even outputs — remain hidden from anyone except the user. This may ease privacy concerns that are increasingly common with AI tools.
New income streams for individuals/hobbyists: GPU owners who are not part of large data centers could monetize their idle hardware, turning personal computers into contributors to the global AI infrastructure.
Integration with existing ecosystems: Because Cocoon is backed by Telegram, it may reach a large user base quickly. Telegram itself is expected to be the network’s first major client, integrating Cocoon-powered AI features (e.g. bots, assistants, smart replies) into its platform.

Challenges and Open Questions

While Cocoon is promising, several challenges and uncertainties remain:

Performance vs. central cloud: It is unclear whether a decentralized network of distributed GPUs can match the performance, speed, and reliability of centralized cloud-AI providers — especially for large models and latency-sensitive tasks.
Node hardware and security: For confidentiality promises to hold, participating nodes must support trusted execution environments (TEEs) or equivalent secure hardware — not all GPUs or home setups may qualify.
Adoption and scale: The concept relies on large numbers of GPU providers and AI-tasks demand. Without broad adoption, the network may struggle to maintain compute capacity, pricing competitiveness, or privacy guarantees.
Trust and transparency: While blockchain and TEEs help, users will need assurance that the system truly keeps data private and that on-chain or off-chain policies are enforced.

Conclusion

Cocoon represents one of the first ambitious attempts to build a privacy-first, decentralized AI infrastructure rooted in blockchain and distributed compute. If successful, it could reshape how AI services are delivered and consumed — giving users more control over data, offering GPU owners new earning opportunities, and challenging centralized AI-cloud dominance.

Whether Cocoon can scale, maintain performance and privacy, and attract enough users remains to be seen — but its launch marks a bold experiment at the intersection of Web3, AI, and digital privacy.