Pavel Durov, founder of Telegram, has announced a new initiative called Cocoon. It is a decentralised AI compute network built on the TON (The Open Network) blockchain. It was announced at the Blockchain Life 2025 conference in Dubai.
Cocoon, which stands for Confidential Compute Open Network, is scheduled to launch this month and will count Telegram as its first major customer. The network will function as a marketplace where individuals or data centers contribute GPU computing power and earn the native cryptocurrency of TON, Toncoin, in return.
From the developer side, Cocoon offers access to distributed, cost-effective AI compute resources handled confidentially, meaning data and queries remain encrypted end-to-end so that no party in the chain (including GPU providers) can see the actual content of processed requests. Integration with Telegram’s ecosystem is also planned: Cocoon will support AI-powered utilities inside Telegram’s mini-apps and platform features, such as summarisation and automated drafting.
At a technical level, Cocoon relies on TON’s high-performance architecture to handle millions of transactions per second. Participants joining Cocoon will specify model architectures (for example, DeepSeek, Qwen), daily request capacities, and token sizes for their workload.
Durov positioned Cocoon as a response to what he described as the gradual erosion of digital freedoms and privacy under centralised AI platforms run by major tech companies. He framed decentralised compute as a way to reduce single points of failure, curb data-profiling or censorship risks, and shift power away from a few large corporations.
Furthermore, Telegram users will soon have access to advanced AI features within the app, processed in a decentralised and encrypted environment compared to traditional cloud-based services. For developers and GPU owners, Cocoon presents a way to monetise hardware or access compute without depending on big-tech infrastructure.
Of course, there are practical challenges ahead: ensuring consistent performance in a distributed network, matching the reliability and latency of cloud providers, and building a pricing and governance model that aligns incentives for GPU providers, developers, and users. The success of Cocoon will hinge on its ability to deliver a seamless experience at scale, without compromising privacy or performance.
