Telegram has officially introduced Cocoon, a new decentralized infrastructure designed to reshape the way artificial intelligence workloads are processed and monetized. Unveiled by Telegram CEO Pavel Durov during the Blockchain Life 2025 conference in Dubai, Cocoon—short for Confidential Compute Open Network—is a blockchain-based AI compute network that rewards GPU owners with cryptocurrency for contributing their hardware to run private AI computations.
Built on The Open Network (TON) blockchain, Cocoon aims to create a distributed marketplace where GPU owners can offer their unused computing power. In return, they receive Toncoin (TON), the native cryptocurrency of the TON ecosystem. This decentralized model offers a compelling alternative to centralized cloud providers, giving developers access to scalable, cost-efficient AI infrastructure while preserving privacy and autonomy.
The network is scheduled to go live in November, with Telegram already confirmed as its first major client. Following the official announcement, Cocoon immediately opened applications for both GPU contributors and developers looking to utilize the network’s decentralized AI compute capabilities.
Cocoon’s core innovation lies in its dual-sided marketplace model. On one side, individuals with powerful GPUs—commonly used in gaming, mining, and AI training—can lease their hardware to the network. On the other side, AI developers, startups, and researchers can tap into this distributed compute layer to run inference tasks without relying on centralized cloud vendors like AWS or Google Cloud.
According to Durov, the initiative addresses two pressing challenges in the AI industry: the scarcity of affordable computing resources and concerns over data privacy. By decentralizing the infrastructure layer and encrypting AI inference processes, Cocoon ensures that data remains secure and untraceable, aligning with Telegram’s long-standing emphasis on user privacy.
Furthermore, the platform is expected to lower barriers to entry for AI developers by dramatically reducing the cost of accessing high-performance computing. This democratization of AI infrastructure could accelerate innovation across fields such as natural language processing, computer vision, and generative AI.
From a technical standpoint, Cocoon utilizes the TON blockchain not only for payment distribution but also for task coordination and verification. Smart contracts ensure that compute tasks are completed accurately, and contributors are paid fairly and transparently in Toncoin.
The timing of Cocoon’s launch is strategic. With growing demand for generative AI applications and a global shortage of GPU availability, the network positions itself as a decentralized alternative to the current cloud computing oligopoly. By harnessing underutilized consumer and enterprise-grade GPUs, Cocoon could dramatically expand the available compute pool for AI workloads.
In addition to its use case in AI inference, Cocoon could also support decentralized training of AI models in the future, although the initial focus remains on inference due to its lower computational demands and real-time application potential.
For Telegram, this move solidifies its deeper integration with the TON ecosystem, which has gained momentum in recent years as a scalable, developer-friendly blockchain. By serving as Cocoon’s first enterprise customer, Telegram not only validates the platform’s utility but also signals its commitment to decentralized technologies beyond messaging.
This development also reflects a broader trend in the tech sector—where major platforms are increasingly seeking ways to decentralize infrastructure and reduce reliance on traditional cloud services. Cocoon fits squarely within this movement, offering a blueprint for how decentralized networks can power next-generation AI applications while rewarding everyday users for their hardware contributions.
Looking forward, the success of Cocoon will hinge on its ability to attract a critical mass of both GPU providers and developers. To incentivize early adoption, the network may introduce bonus rewards, staking mechanisms, or governance features that allow users to influence the platform’s future direction.
Security will also be a focal point. Running sensitive AI computations in a distributed environment introduces new risks, which Cocoon aims to mitigate through advanced encryption techniques, zero-knowledge proofs, and verifiable computation layers.
Another crucial aspect will be interoperability. For developers already building on other AI platforms or blockchains, seamless integration with Cocoon’s APIs and SDKs will be key to its adoption. The team behind Cocoon is expected to release comprehensive developer tools ahead of the full network launch.
Cocoon could also open the door for new business models in AI. For instance, startups could build decentralized applications (dApps) that provide AI services—such as chatbots, image recognition, or voice synthesis—directly on the Cocoon network, without intermediaries or platform fees.
As more users become aware of the economic potential of leasing unused GPU power, Cocoon could encourage a new wave of participation in the crypto economy. Similar to how Helium incentivized users to share bandwidth, Cocoon could inspire a community-driven approach to AI infrastructure.
In summary, Cocoon represents a bold step toward a more decentralized and privacy-focused AI future. By combining the power of blockchain, distributed hardware, and economic incentives, Telegram is not only expanding the TON ecosystem but also challenging the status quo of AI computing. As the network approaches its launch later this year, it will be closely watched by both the crypto and AI communities for its potential to transform how machine learning workloads are deployed and monetized globally.