Telegram founder and CEO Pavel Durov has announced the launch of Cocoon, a decentralized AI compute marketplace built on TON and aimed at users and developers who prioritize confidentiality. According to Durov, the network is already live and serving its first user requests, with additional computing capacity expected to come online over the next several weeks.
Cocoon enters the market as a privacy-centered alternative to traditional cloud-based AI infrastructure. In Durov’s framing, centralized providers such as Amazon and Microsoft function as costly intermediaries that raise prices while offering weaker privacy guarantees. Cocoon’s pitch is that a decentralized marketplace can address both of those problems at once: reduce intermediation and preserve confidentiality for participants on both sides of the network.
A marketplace for AI apps and GPU supply
At its core, Cocoon is designed to broker transactions between applications that need compute and providers willing to supply GPU resources. The network uses TON as its reward and payment token, creating an economic layer for matching demand and supply without relying on a centralized cloud operator. App developers can contract GPU processing power through the network, while compute providers can monetize idle or available resources privately.
This structure is intended to give developers more direct access to processing power while opening a new channel for GPU owners to participate in the AI economy. Rather than routing activity through a single corporate platform, Cocoon proposes a distributed model in which compute is sourced across a decentralized network. For Telegram and TON, that also extends the use case of their ecosystem deeper into AI infrastructure, an area that has attracted increasing attention across the crypto industry.
Positioning against centralized AI infrastructure
Durov’s comments make clear that Cocoon is not being presented merely as another compute product, but as a challenge to the economics of centralized AI services. His argument is that legacy providers add expense and reduce confidentiality, making them poorly suited for developers or users who want greater control over how AI workloads are processed.
That message lands at a time when the broader AI sector is facing infrastructure constraints. Demand for GPUs has remained elevated, while data center buildouts continue to run into limits tied to power, cooling, water, and physical infrastructure. The source material notes that Microsoft CEO Satya Nadella recently said that even with GPUs available, the company still lacked the infrastructure necessary to power them. That remark has helped underscore a wider industry problem: the bottleneck in AI is no longer only about chip availability, but also about where and how those chips can actually be deployed.
Advocates of decentralized compute networks argue that distributed architectures may help relieve some of these pressures. Instead of concentrating resources in a small number of massive data centers, decentralized markets can aggregate compute from multiple providers and locations. In theory, that could make it easier to organize available GPU capacity without requiring every participant to solve hyperscale infrastructure challenges on their own.
Privacy as a defining feature
One of Cocoon’s clearest points of differentiation is its emphasis on confidentiality. Durov said the first user requests are already being served confidentially, and he framed that as central to the platform’s identity. In a market increasingly concerned with data handling, model interactions, and sensitive enterprise workloads, privacy has become a meaningful selling point.
Telegram is using that concern as a strategic angle. The company’s messaging platform has long been associated with privacy-oriented users, and Cocoon appears designed to extend that brand positioning into AI services. Durov said Telegram users should expect new AI-related features built on “100% confidentiality”, signaling that Cocoon may serve not only external developers and GPU providers, but eventually Telegram-native AI products as well.
While the announcement did not provide technical implementation details beyond the decentralized market structure and TON-based payments, the broader message is clear: Telegram wants to present Cocoon as an infrastructure layer where privacy is not an afterthought, but part of the product itself.
Why the launch matters
The launch is relevant for several reasons. First, it reflects a growing push within crypto and Web3 circles to build decentralized alternatives to AI infrastructure, not just decentralized applications that sit on top of existing cloud providers. Second, it aligns with a wider market thesis that GPU owners should be able to monetize compute more directly, particularly during a period of strong AI demand. Third, it highlights the role of tokenized networks like TON in coordinating payments and incentives for real-world computational resources.
For developers, Cocoon could represent a new procurement path for AI compute, especially if it succeeds in onboarding meaningful GPU supply. For compute providers, it offers a way to sell capacity without going through the standard centralized cloud stack. And for Telegram, it may become another strategic bridge between its large user base and infrastructure services built within the TON ecosystem.
Still, the announcement is at an early stage. The network is live, but its long-term significance will depend on whether it can scale supply, attract sustained developer demand, and deliver on the confidentiality and cost claims made at launch. In markets dominated by entrenched cloud providers, distribution, reliability, and resource quality matter as much as ideology.
What comes next
Durov said Cocoon will continue scaling in the coming weeks by onboarding more GPU supply and bringing more developer demand onto the network. That next phase will be crucial. A decentralized compute marketplace only works if both sides of the market grow together: developers need sufficient capacity and predictable performance, while providers need enough utilization and economic incentive to remain active participants.
For now, the announcement positions Cocoon as a fresh attempt to merge decentralized infrastructure, AI demand, privacy guarantees, and TON-based payments into a single marketplace. Whether it can seriously challenge traditional AI cloud providers remains uncertain, but Telegram has made its direction clear. With Cocoon, it is betting that the future of AI compute can be more distributed, more private, and less dependent on centralized intermediaries.