The Telegram's Cocoon AI: Not Yet Mature

In October 2025, Pavel Durov announced Telegram's Cocoon AI initiative. The approach was different from that taken by Meta, OpenAI, and some other social media firms. Telegram would not pump billions into infrastructure. Instead Telegram would use existing AI data centers' excess capacity. The benefit? No massive capital investments and no delay waiting for construction, GPUs, and infrastructure.

Telegram, in a way similar to Apple, has been focusing on other issues. Apple had a management change and rolled out the disruptive Neo. AI is a "to be" innovation. But Apple is positioning itself as "smart" for not jumping on the frontier model circus train.

Pavel Durov, who embodies Telegram, has been juggling some burning red road flares. The criminal trial in France is hanging over his head. Will he pay a fine or will he spend years in a cheerful French prison like Fleury? Telegram reported a $200 million loss in fiscal 2025. Have revenues rebounded in 2026? Russia has frozen hundreds of millions in Telegram cash. Plus, Durov's company has bond obligations to service. A couple of weeks ago, Durov pulled his brother's blockchain technology and the TONcoin crypto currency back into Telegram. Will the Swiss regulators take action because the TON Foundation of today no longer matches the TON Foundation described in 2022? Telegram linked start ups — Alpha Compute (né AlphaTON Capital) and TON Strategy Company have been battered in the US capital market. Alpha Compute's original business model was to sell AI processing to Telegram has changed. Now Alpha Compute is selling AI compute to frontier artificial intelligence developers. TON Strategy Company, a firm emulating Strategy Co.'s Bitcoin asset business model, is struggling because the value of TONcoin has dropped significantly in the last six months.

Not surprisingly, Telegram's Cocoon AI has not as yet revolutionized how Telegram operates. From my vantage point, Telegram's progress has been slow, and it is incremental. Telegram implemented a "smart" orchestrator. A bot developer can combine several bots in order to accomplish a series of tasks. Telegram allows AI to generate summaries of posts. Bots can talk to other bots. An AI editor allows a bot to translate a post or fix its grammar. There is an AI-assisted emoji and sticker search. AI bots can generate news summaries, provide customer service, and interact with bots inside chats. Click here for the source document

The showcase example of what a developer can do with these AI functions is the @LobsterClawBot. It handles pulling crypto prices and supports a conversational agent. If a solo crypto trader wants automatic price information, this bot might do the trick. With the new features, a small business operator who wants to provide a customer support bot, the features make that a possibility without writing code.

Social media commentators suggest that almost 200 AI bots are available on Telegram. These bots handle search, writing and editing of posts, crypto trading, and similar utility functions.

The publication Startup Fortune was enthusiastic about the AI innovations from Telegram. The publication's assessment links the alleged one billion Telegram users to a latent desire to code bots for Telegram. I agree that the new features can eliminate writing code for a bot; however, the suggestion that one needs zero programming "knowledge" may not match the experience for those developing their first bot. (One does have to know what must be accomplished and have some of the Introduction to Programming information at hand.) Click here for the source document

These new AI features complement existing support for third-party AI services in bots and mini apps. Bots can connect to frontier AI services to conduct research and present the results in Telegram Messenger. The orchestration capability is a significant technical step for Telegram. But these AI enhancements are not the Cocoon service described in 2025 and early 2026.

I have been following Telegram's use of AI for more than a year. A buzz rippled around the company's partner and developer community about Grok. That rumor fizzled out. For more than a year, a bot developer could pay for an AI application programming interface or use a free one. The bot could then provide its user with a way to handle customer support or correct the grammar in a Telegram Messenger post to a public or private Group or Channel. With the addition of the orchestration service, bots become more flexible. However, there is no AI support within Telegram to reduce the mountain a developer has to climb to create a complex, crypto centric distributed application of dApp. Maybe at some point in the future? But today. No.

To me, Telegram's AI tactics are defensive measures to stay in the AI game. Cocoon was originally a privacy-first, decentralized compute story to tell. The TON Foundation would amplify that message in its meet ups, lectures, and videos. However, the actual AI features are useful to some developers, not revolutionary. Automation, bot-to-bot coordination, and developer-accessible AI wrappers are features. But AI innovations like these may be more immediately useful to phishers, eGame operators, gray market crypto services, spam networks, and similar Telegram Messenger users. For someone just sending a message, Cocoon is a "to be" service.

At this point in time (late May 2026), Durov commands an organization with no public succession plan. His early signatory for deals (Vladimir Vedeneev, Global Network Monitoring) could step in if Durov goes to prison. Nikolai Durov, a mathematician with two Ph.Ds is a long shot. Possibly Max Crown, now the neutered TON Foundation president, could fill Durov's shoes until he is released. But this short list is pure speculation.

After the announcement in October 2025 about Cocoon, Telegram has to produce some substantive, impressive AI actions. Every day, innovations flow from Google and other high-profile AI firms. Telegram has no frontier model, preferring it seems to use Chinese open source technology. A rich service stack to support distributed applications is not mentioned in the available public Telegram content.

Based on my notes, Durov's Cocoon will at some point (one hopes) distributed AI services snapped into the TONcoin crypto. Alpha Compute’s pivot from Telegram / TON centricity to a single mysterious AI-lab customer says to me, "The Telegram ecosystem is talking about AI, not delivering operating substance and revenue. The asset company (TON Strategy) continues to wrestle with the value loss of its principal asset: TONcoin crypto.

Stephen E Arnold, June 2, 2026