AI development often comes at the expense of data collected en masse by centralized operators. Telegram founder Pavel Durov aims to address this issue with a decentralized solution called Cocoon, built on the TON blockchain.
Pavel Durov presents the Cocoon project
For many years now, the Telegram messaging app has been actively involved in the development of the cryptocurrency sector, particularly due to its founding principles of privacy and decentralization, even though its first venture with the GRAM token ended in a bitter regulatory failure in 2020.
Following this setback, the TON (The Open Network) blockchain took off independently, while remaining closely tied to the ambitions of Telegram founder Pavel Durov, who was determined not to be silenced in this way. This provided an opportunity to develop an entire ecosystem of decentralized mini-apps (dApps) linked to the messaging platform, generating over $1 billion over the past year.
It is against this backdrop that Pavel Durov has just made another major announcement at the Blockchain Life 2025 event held in Dubai in recent days. On the agenda: the development of a decentralized AI network called Cocoon, for Confidential Compute Open.
Why is it important to do things this way rather than in a centralized manner, which is sometimes more convenient? It’s important because the world has taken a strange turn. Over the past twenty years, we have been gradually losing our digital freedoms.
Pavel Durov
A network powered by graphics processing units (GPUs)
From a technical standpoint, the deployment of this decentralized AI network will, of course, take place on the TON blockchain, with active support from the Telegram messaging app.
Telegram will be the first major client to use Cocoon for confidential AI queries—and will invest heavily in promoting the network across its global ecosystem
Cocoon
Interestingly, it will be possible to make the computing power of graphics processing units (GPUs) available to the network in exchange for rewards paid in Toncoin (TON).
This is because the primary goal of the Cocoon project is to offer a decentralized AI service using “private and confidential computing.” In this case, this approach relies on a DePin-type infrastructure, already widely used by the sector’s flagship project, Bittensor (TAO).
The real challenge for this decentralized AI network will lie in its ability to generate sufficient computing power to ensure smooth operation while maintaining competitive usage costs. Its public launch is scheduled for next November.
