Telegram’s Durov Launches Cocoon: Decentralized AI Network Goes Live on TON

Introduction: A New Chapter for Decentralized AI and User Empowerment

In a landmark development for the decentralized technology space, Telegram co-founder Pavel Durov has officially launched Cocoon, a privacy-preserving decentralized AI network built on The Open Network (TON). The platform went live, marking a significant step in the convergence of blockchain and artificial intelligence. Cocoon’s operational model is straightforward yet revolutionary: it enables owners of graphics processing units (GPUs) to rent their computing power to the network. In return for processing user queries and requests, these hardware owners earn Toncoin (TON), the native cryptocurrency of the TON blockchain. Announced by Durov at the Blockchain Life 2025 conference in Dubai, United Arab Emirates (UAE), in October, Cocoon is positioned as a direct response to growing user demand for an AI platform that prioritizes data privacy and challenges the dominance of centralized service providers.

The Architecture of Cocoon: Decentralized Computing on TON

At its core, Cocoon is a distributed computing platform that leverages the existing infrastructure of The Open Network. By building on TON, an independent layer-1 blockchain historically associated with the Telegram messaging application, Cocoon inherits the blockchain’s scalability and integration potential. The fundamental economic mechanism involves a two-sided marketplace. On one side are users who submit AI-related queries and tasks. On the other side are GPU owners who provide the raw computational power required to process these tasks.

The incentive is clear and built directly into the protocol: compute providers are compensated in TON tokens. This model effectively creates a decentralized network of AI computation, where the supply of computing power is not controlled by a single corporate entity but is crowdsourced from participants globally. Durov explicitly contrasted this model with traditional providers, stating, “Centralized compute providers such as Amazon and Microsoft act as expensive intermediaries that drive up prices and reduce privacy. Cocoon solves both the economic and confidentiality issues associated with legacy AI compute providers.”

Pavel Durov’s Vision: Privacy and Economics as Driving Forces

The launch of Cocoon is a direct manifestation of Pavel Durov’s long-standing advocacy for user privacy and decentralized systems. His announcement at the Blockchain Life 2025 conference framed Cocoon not merely as a new product but as a philosophical stand against the current trajectory of AI development. Durov identified two primary issues with the status quo: cost and confidentiality. Centralized providers, by virtue of their market position, can dictate pricing and have full visibility into user data and queries.

By architecting Cocoon as a decentralized network, Durov aims to dismantle this intermediary role. The economic issue is addressed by creating a more direct and competitive market for computing power, potentially lowering costs for end-users. The privacy issue is mitigated by distributing tasks across a network of independent nodes, preventing any single entity from accumulating vast datasets of user activity. This vision aligns with the initial principles of blockchain technology—disintermediation, user sovereignty, and censorship resistance—now applied to the burgeoning field of artificial intelligence.

The Broader Context: The Push for Decentralized AI

The launch of Cocoon occurs within a significant industry-wide movement advocating for decentralized AI. The blockchain community, privacy advocates, and cypherpunks have long warned about the societal risks of centralized AI systems. These concerns are not new but have gained urgency with the rapid adoption of powerful AI models.

David Holtzman, chief strategy officer of the Naoris decentralized security protocol, articulated these threats to Cointelegraph, noting that centralized AI gives “governments and corporations enormous leverage over individuals that can compromise user privacy, threaten traditional cybersecurity safeguards, and lead to social conditioning by organized actors.” He further argued that blockchain technology can serve as an antidote by verifying information sources, ensuring tamper-proof records, and enabling trustless communication between nodes on a distributed network.

This sentiment is supported by data. A poll conducted by the Digital Currency Group (DCG) in May revealed that 77% of 2,036 respondents believed that decentralized AI would benefit society more than centralized systems. Furthermore, in 2024, AI researchers from the Dfinity Foundation—the non-profit steering development of the Internet Computer Protocol (ICP)—along with executives from decentralized AI developer Onicai, outlined seven rules for ethical AI. A key tenet was running AI on permissionless blockchain networks to ensure transparency and data integrity.

Comparing Decentralized AI Initiatives: A Landscape View

While Cocoon is a prominent new entrant, it joins a growing ecosystem of projects aiming to decentralize artificial intelligence. It is useful to contextualize its approach against other initiatives without speculating on their relative success.

• SingularityNET: One of the earliest projects in this space, as highlighted in a related magazine feature, focuses on creating a decentralized marketplace for Artificial General Intelligence (AGI) services. Its scope is broad, aiming to be a hub for various AI algorithms and agents that can interact with each other.
• Internet Computer Protocol (ICP): As referenced in the ethical AI rules proposed by the Dfinity Foundation, ICP provides a platform for hosting smart contracts and entire applications in a decentralized manner. Its application to AI involves creating tamper-proof and transparent AI models that run on-chain.
• Cocoon on TON: Cocoon’s specific niche is in decentralized compute. Rather than focusing solely on the AI models or algorithms themselves, it provides the foundational processing layer in a distributed way. Its integration with the TON blockchain and its association with Telegram’s massive user base are distinctive factors that differentiate its potential scale and market role.

Each project tackles decentralization from a different angle—be it algorithmic marketplaces, on-chain execution, or distributed hardware resources—contributing to a more resilient and user-centric AI infrastructure.

Initial Operations and Future Implications

According to Pavel Durov, the Cocoon network has already processed its first user requests, and GPU owners have begun earning TON for their contributions. This initial activity demonstrates a functioning proof-of-concept for its core economic model. The immediate implication is the creation of a new utility stream for TON tokens, potentially increasing demand for the cryptocurrency within its own ecosystem as it becomes the required medium of exchange for accessing decentralized AI compute power.

For the broader market, Cocoon’s launch represents a significant validation of the decentralized AI thesis by a major tech figure. It brings mainstream attention to the technical and philosophical debates surrounding AI governance and data control. By providing a working alternative, it challenges other projects in both the blockchain and traditional tech sectors to address concerns over cost and privacy more effectively.

Conclusion: A Strategic Step Towards a Decentralized Digital Future

The launch of Cocoon on the TON blockchain is more than just another product release; it is a strategic maneuver in the ongoing battle for the soul of the internet’s next generation. Pavel Durov has positioned Cocoon at the intersection of three critical technological trends: messaging platforms becoming super-apps, the rise of decentralized physical infrastructure networks (DePIN), and the global demand for private, ethical AI.

By leveraging TON’s blockchain for secure settlements and incentivizing a global network of hardware providers, Cocoon offers a tangible solution to the problems of centralized control in AI compute. Its success will depend on its ability to scale reliably, attract a critical mass of both users and compute providers, and demonstrate clear advantages in cost and performance over established incumbents.

For readers and industry observers, key metrics to watch will be the growth in network computing power, the volume of processed queries, and developer adoption building applications on top of Cocoon’s infrastructure. As centralized AI continues to face scrutiny over data usage and bias, decentralized alternatives like Cocoon are poised to become increasingly relevant, potentially reshaping how society interacts with and benefits from artificial intelligence.