Pavel Durov throws down the gauntlet to @amazon and @Microsoft: Cocoon is Web3’s first real strike against AI hyperscalers (Not that easy though 😬) A decentralized "cocoon" for AI instead of hyperscalers: @telegram's founder has launched Cocoon — a confidential compute network on TON for handling AI requests with full data encryption. For the end user, it looks like the usual access to models (translation, summarization, text generation), but with one key difference: the data is processed in a secure environment, not on Amazon/Microsoft servers. GPU owners plug in their compute and earn TON, while developers get an alternative to @OpenAI / @AnthropicAI / @togethercompute / @replicate — only built on top of a decentralized network. What I find especially interesting is that this is one of the few cases where a "crypto" project is being seriously discussed in an AI context, rather than as yet another token without a product. Here the Web3 component is essentially a transparent compute marketplace: if you have suitable hardware, you can utilize it more efficiently than through a closed, centralized platform, while knowing that the rules are the same for everyone. This is a fair, easy‑to‑understand role for a token: not the "magic of community economics," but a mechanism for distributing revenue among GPU providers. At the same time, many of the media narratives around Cocoon can and should be grounded in reality. To deliver that promised privacy, the network relies on confidential compute and Trusted Execution Environments (TEE), which means it’s not just "any spare GPU from a garage." Most GPUs already sitting in data centers don’t work "out of the box," and specialized TEE‑enabled hardware is scarce and expensive. On top of that, this is by definition a narrow niche: not simply "cheaper than AWS," but "confidential compute protected even from the cloud provider." In such a niche, there’s no obvious reason why prices should be lower than those of traditional hyperscalers. On the other hand, Cocoon has something most Web3 infrastructure projects lack: an anchor customer. If Telegram immediately becomes a major user of the network, and a few serious data centers connect to Cocoon, then both demand and supply appear at once — a minimally viable market. On this foundation, a niche but sustainable confidential AI ecosystem could emerge. I’ll be watching less the "decentralization vs Big Tech" marketing, and more the hard metrics: stability of GPU supply, economics for TON miners, and real‑world performance compared to centralized clouds.