Telegram makes TON a hub for decentralized AI inference
Telegram will become the first major client for its own Cocoon network. This move pledges the platform’s immense user base and query volume to bootstrap a new economy for private AI inference on the TON blockchain.
- Telegram unveiled Cocoon, a decentralized AI inference network built on TON.
- Developers can access GPU power and pay with TON tokens; GPU owners earn by contributing compute.
- AlphaTON Capital backed the rollout with global GPU infrastructure to power private, onchain AI computation.
On Oct. 29, Telegram CEO Pavel Durov took the stage at the Blockchain Life conference in Dubai to unveil Cocoon, a decentralized network built to process artificial intelligence queries privately on the TON blockchain.
Durov said the system will connect developers seeking compute power with GPU owners willing to lend it, rewarding them in Toncoin (TON) tokens for executing inference tasks. Telegram, which plans to be Cocoon’s first large-scale user, will route portions of its own AI workloads through the network and promote the platform to its global audience of more than a billion users.
Telegram builds a decentralized bridge between AI and blockchain
The Dubai announcement was not the first glimpse of Telegram’s AI ambitions. Durov initially teased the project just weeks earlier during an October trip to Kazakhstan, where he introduced a dedicated AI laboratory.
At that time, he revealed Telegram had spent months developing technology at the intersection of blockchain and artificial intelligence, framing it as a necessary step to offer confidential and transparent AI to the platform’s billion-plus users.
For the network to function, Cocoon requires a ready pool of participants. According to its official channel, the onboarding process is now live. Application developers seeking to use the network can directly message the channel, specifying the AI model architecture they plan to run, such as DeepSeek or Qwen. They are also required to provide their expected daily query volume and average token size for inputs and outputs.
On the supply side, GPU owners looking to earn TON tokens for their computational power must detail their hardware capabilities. The channel requests specifics on the number of GPUs available, their model type, VRAM capacity, and expected uptime.
AlphaTON Capital backs Cocoon network
Shortly after Durov’s announcement, AlphaTON Capital confirmed its strategic investment in high-performance GPU infrastructure to support Cocoon’s rollout. The Nasdaq-listed firm, known for its deep involvement in the TON ecosystem, said it would deploy advanced GPU fleets across global data centers to provide the computational backbone for the network.
CEO Brittany Kaiser called the launch a “pivotal moment in the convergence of blockchain technology, data protection and artificial intelligence,” committing the firm as a “cornerstone contributor” to the ecosystem. Executive Chairman Enzo Villani emphasized that “privacy and security are paramount in the AI era,” stating that Cocoon’s architecture meets a critical market need.