Telegram Founder Launches Cocoon: A Decentralized Network Challenging Big Tech's AI Monopoly

Telegram Founder Launches Cocoon: A Decentralized Network Challenging Big Tech’s AI Monopoly

Telegram Founder Launches Cocoon: A Decentralized Network Challenging Big Tech’s AI Monopoly

Pavel Durov, the founder of Telegram, has unveiled Cocoon—a decentralized network designed to disrupt the dominance of cloud computing giants like Amazon and Microsoft in the artificial intelligence sector.

According to Durov, the platform promises to deliver AI computations with complete confidentiality, zero tracking, and prices below current market rates.

XChat Security Analysis: Safe as “Bitcoin-style” peer-to-peer encryption?

Three-Layer Architecture Built on PrivacyAccording to technical documentation on the project’s official website, Cocoon operates through a sophisticated three-layer structure built on the TON blockchain:

Worker Layer: This foundational tier runs AI models within Intel TDX (Trust Domain Extensions) trusted execution environments. GPU owners can participate simply by installing images, configuring model names, and setting up their TON wallet addresses to begin processing AI inference requests.

Proxy Layer: Serving as the network’s traffic controller, this layer routes requests by evaluating factors such as model type, current load, and participant reputation. While currently operated by the Cocoon team, the platform plans to open this layer to any participant in the future.

Client Layer: This interface enables services to send inference requests. Telegram’s backend will run multiple client instances to handle user requests, making the messaging platform Cocoon’s first major customer.

Should Governments Ban VoIP to Stop International Phone Scams?

How It Works: A Four-Step Process

The system executes AI queries through a streamlined workflow:

The client establishes an RA-TLS (Remote Attestation Transport Layer Security) connection with a proxy, verifying the proxy’s TEE attestation
The proxy connects to a selected Worker via RA-TLS, confirming the Worker’s TEE credentials
The client submits a prepaid inference request, which the proxy forwards to the Worker for processing within the secure enclave
The Worker returns the response, receives payment through smart contracts, and the proxy delivers results to the client

All communications utilize RA-TLS encryption, ensuring that only the client can access prompts and responses. GPU owners earn Toncoin for processing queries while maintaining complete data confidentiality.

End-to-End Encryption in VoIP: Understanding SIP Protocol and E2EE Support

Market Context and Competition

The network officially launched on November 30, 2024, and has already processed its first user requests, with GPU providers beginning to earn revenue. However, Cocoon enters a market where similar services already exist. Platforms like Vast.ai and RunPod.io have established decentralized GPU rental services, while projects like Akash Network and Render Network offer distributed computing resources.

Critics have raised concerns about latency issues that could impact real-time applications. However, proponents argue that many use cases—including machine learning training, 3D rendering, and scientific simulations—are not latency-sensitive and would benefit from Cocoon’s cost advantages and privacy guarantees.

The network’s topology reflects practical design choices: multiple clients connect to a modest number of proxies (10-100), which coordinate with a large pool of Workers (1,000+). Proxies monitor response times and success rates, building an on-chain reputation system to address performance and reliability concerns.

Why IP Video Phones Don’t Support H.265 Video Encoding?

Privacy at the Core

Durov framed Cocoon as a solution to how centralized providers drive up prices while compromising privacy. The confidential computing approach addresses growing concerns about how major AI companies handle user data. When users interact with services from companies like OpenAI or Google, those providers can access all prompts, responses, and usage patterns. Cocoon’s architecture encrypts this information throughout the entire process, protecting it even from the GPU owners performing the computations.

Governance and Payment Systems

Currently, Cocoon operates under centralized management, with the development team controlling the root smart contract that stores approved image hashes, model hashes, proxy addresses, and other network configurations. The project’s roadmap includes transitioning to decentralized autonomous organization (DAO) governance.

The payment infrastructure resembles payment channels, utilizing smart contracts on the TON blockchain to handle transactions between clients and proxies. This creates a marketplace where pricing emerges from supply and demand dynamics, potentially offering more competitive rates than centralized cloud providers.

Why Satellite Companies Haven’t Encrypted Most Communications?

Looking Ahead

Durov announced plans to expand by attracting more GPU providers and developers, with Telegram users expected to see new AI features ensuring complete privacy. The project’s code repository is available on GitHub, and detailed documentation for developers and GPU owners can be found on the official website.

AlphaTON Capital has announced substantial investment plans to deploy next-generation, high-memory GPU models across strategic data centers, signaling institutional confidence in the platform’s potential.

Whether Cocoon can truly challenge the dominance of established cloud computing giants remains to be seen. However, with Telegram’s massive user base and the growing demand for privacy-preserving AI solutions, the platform has positioned itself as a significant experiment in democratizing AI infrastructure—one that could reshape how billions of users interact with artificial intelligence in their daily digital lives.

Telegram Founder Launches Cocoon: A Decentralized Network Challenging Big Tech's AI Monopoly