Launch of Cocoon: Why Durov is challenging Amazon and Microsoft

​On the last day of autumn, Telegram founder Pavel Durov launched his new project — Cocoon, a decentralized network built on the TON blockchain. The project promises private AI task execution without involving major corporations and has already begun processing its first requests. According to Durov, Cocoon is meant to become an alternative to Big Tech and return control over data to users.

Official launch of Cocoon

The new confidential computing network Cocoon (Confidential Compute Open Network) officially went live on November 30. The project launched by handling its first requests, which are distributed across connected GPU providers. GPU owners receive rewards in Toncoin, while tasks are executed inside encrypted environments that prevent corporations and node operators from accessing user data.

Loading...

The tweet was deleted by the author.

But we saved everything 🙂.

The core idea of the project is straightforward: remove the intermediaries that dominate the AI compute market and effectively control access to computation. Companies like Amazon and Microsoft make AI expensive and opaque, whereas TON’s decentralized model reduces costs and returns data control to users, Durov argues. Cocoon relies on a distributed GPU network where each task “lives” only during execution and leaves no trace in centralized storage.

The team expects user interest to surge in the coming weeks as new providers join and Telegram demand grows. The messenger will become Cocoon’s first major client by integrating private AI features through mini-apps. For the TON ecosystem, this effectively opens a new market segment — private compute — previously dominated by Big Tech.

A test shot

The launch of Cocoon was not unexpected. Pavel Durov first introduced the project a month earlier at the Blockchain Life 2025 conference in Dubai. There, he publicly presented the idea of a private, decentralized network capable of processing AI requests without relying on corporations or centralized cloud services. His speech quickly became one of the most discussed at the forum: the Telegram founder openly claimed that Big Tech is turning artificial intelligence into a tool of control — and decentralization is the only way to prevent that.

Loading...

The tweet was deleted by the author.

But we saved everything 🙂.

At the event, Durov revealed Cocoon’s core technical principles: operation on the TON blockchain, distribution of tasks across GPU providers, confidential execution environments, and the complete absence of centralized data storage. He emphasized that the network is not an experiment but next-generation infrastructure enabling developers to deploy their own AI models without depending on major cloud players.

The Blockchain Life audience responded with clear enthusiasm: the combination of TON, a decentralized GPU marketplace, and Telegram integration looked not only ambitious but genuinely promising. But why did this project emerge right now?

Why Durov needs a decentralized network

The launch of Cocoon coincided with a period of international isolation for Pavel Durov. More than a year ago, he was detained in France and accused of enabling criminal activity through Telegram, including fraud, drug trafficking, and the distribution of illegal content. Since then, he has been under a partial travel restriction, allowed to leave the country only with specific permissions. The situation around Telegram has become part of the broader political and regulatory pressure on digital platforms.

Loading...

The tweet was deleted by the author.

But we saved everything 🙂.

Under these circumstances, the idea of a decentralized network takes on additional meaning for Durov. Cocoon is an infrastructure that cannot be blocked or controlled through traditional pressure mechanisms: it does not rely on data centers, is not tied to any specific jurisdiction, and does not store user data centrally. While Telegram must interact with regulators as a communication platform, Cocoon functions more like a protocol — a distributed system that cannot be “switched off” by administrative order.

Durov has been warning about the risks of a centralized internet for several years, arguing that governments and corporations are gaining increasingly powerful tools for surveillance and control. For him, Cocoon represents a form of digital autonomy.

Does the crypto community need a project like this?

For the crypto community, the emergence of Cocoon looks like a natural continuation of the growing trend toward decentralized artificial intelligence. The industry increasingly points out that AI built on Big Tech infrastructure can no longer be considered a neutral tool: access to compute is restricted, costs are rising, and privacy depends entirely on corporate goodwill. In this context, a project that enables AI inference on a distributed GPU network and allows payments in cryptocurrency addresses several core demands of the ecosystem: privacy, openness, and economic independence.

Beyond its technological appeal, Cocoon aligns with the ideological expectations of the market. The rise of the people-owned AI concept — from QVAC to decentralized compute networks on ICP and Bittensor — shows that users want to control not only their data but the infrastructure AI runs on. In this sense, Durov’s project fits into global trends and may become an entry point to private AI for mainstream audiences, given Telegram’s billion-user base. For developers, Cocoon offers access to affordable compute power without renting servers from AWS or Google Cloud.

However, an immediate market shift should not be expected. Decentralized networks still fall behind cloud providers in stability, speed, and quality control. Questions remain about how well Cocoon can scale and whether GPU providers will have sufficient incentives, especially amid Toncoin’s volatility. Another unresolved issue is who will train large models if the network focuses primarily on inference. Still, for the crypto sector, the launch of such a network is already a significant step: it demonstrates that private AI can exist outside large corporations and evolve as part of the Web3 ecosystem.