In brief
- Telegram launched Cocoon, a decentralized AI compute network on TON that pays GPU owners in crypto for powering private AI tasks.
- Telegram will use Cocoon internally, integrating it into Mini Apps and AI tools for its billion-plus users.
- The project challenges Big Tech’s dominance, positioning TON as an open, user-controlled alternative to AWS and Azure.
Telegram CEO Pavel Durov took to the stage at Blockchain Life 2025 in Dubai on Wednesday to announce Cocoon, a decentralized AI compute network built on The Open Network (TON) blockchain that will pay GPU owners in Toncoin for powering private AI inference.
The network, formally called the Confidential Compute Open Network, will launch in November with Telegram as its first major customer. Applications for GPU providers and developers opened immediately following the announcement.
Cocoon creates a marketplace where individuals contribute computing power through their graphics processing units and receive TON cryptocurrency in exchange. On the other side, developers gain access to low-cost AI infrastructure that processes queries without exposing user data to centralized providers.
This decentralized AI infrastructure is billed as a rival against the monopoly that Big Tech corporations like Amazon’s AWS and Microsoft Azure have. In fact, Durov positioned the project as a response to eroding digital freedoms. Both Telegram and the TON blockchain were also conceived with this privacy-first approach in mind.
The timing aligns with growing concerns about centralized AI systems. When companies like OpenAI or Google process your queries, they see everything—your prompts, your data, your patterns, and metadata. Cocoon's confidential computing approach keeps that information encrypted throughout the process, even from the GPU owners doing the actual computation.
Telegram's globally adopted platform provides immediate scale. The messaging app will integrate Cocoon across its ecosystem, powering AI features in its Mini Apps and potentially transforming how users interact with artificial intelligence daily.
Max Crown, CEO of the TON Foundation, said the launch marks a shift toward an open, user-driven compute economy that returns control of AI infrastructure to users—rather than corporations that exploit it.
"Leveraging Telegram's billion-strong user base and TON's high-performance, scalable blockchain technology, Cocoon has the potential to redefine how billions interact with AI in their everyday digital lives,” he said in a press release shared with Decrypt. “Cocoon is the convergence of social networking, AI, and decentralized technology at unprecedented scale.”
AlphaTON Capital, a Nasdaq-listed digital asset infrastructure and TON treasury company, announced its intention to make a “substantial investment” in all the hardware required to support the network. The company plans to deploy next-generation, high-memory GPU models across strategic data centers, supporting advanced model architectures including DeepSeek and Qwen.
The project builds on TON blockchain's multi-chain architecture, which processes millions of transactions per second through its sharded design. The parties said this structure allows Cocoon to handle massive AI workloads while maintaining the transparency and market-driven pricing that blockchain enables.
Decentralized AI networks have gained momentum as an alternative to tech giants. Projects like Akash Network and Render Network already let users rent distributed computing resources, though Cocoon's integration with Telegram's user base appears to be a big selling point.
Like other decentralized cloud/storage networks, the market pricing will emerge from supply and demand dynamics as GPU owners and developers negotiate costs through the network. This creates potential for more competitive pricing than centralized cloud providers typically offer, though it also introduces uncertainty around service reliability and consistency.
"In a world where centralized AI systems harvest data and concentrate power, Cocoon rebalances the equation by giving users control over their computation, privacy, and ownership,” Crown said.

