Cocoon operates as a decentralized AI inference marketplace built on The Open Network (TON) blockchain. The network connects GPU providers (Workers) who supply computational resources with developers and applications requiring private AI inference capabilities. Transactions settle in Toncoin (TON) , with payment records and settlement finalized on the TON blockchain.
The network implements a three-layer architecture comprising Clients, Proxies, and Work Nodes. Clients submit inference requests through Proxies, which route workloads to available Workers operating Trusted Execution Environment (TEE) -enabled GPU servers. This design abstracts hardware complexity while maintaining verifiable execution guarantees.
Economic Model and Monetization Mechanism
GPU owners earn Toncoin rewards for processing AI inference requests. The compensation structure operates on a supply-demand pricing model where market dynamics determine per-task rates. Developers pay in TON for network access, creating a closed-loop economic system within the TON ecosystem.
AlphaTON Capital Corp. (Nasdaq: ATON) serves as a major infrastructure contributor. The company reported revenue generation commencing December 2025 following deployment of its initial GPU fleet to support Cocoon AI inference workloads. AlphaTON has since expanded its footprint with additional high-performance chips and continues scaling toward a planned deployment of more than 570 NVIDIA B300 GPUs scheduled for delivery in March 2026.
The company projects a 27% internal rate of return (IRR) and $11 million in net present value (NPV) from the rollout. AlphaTON raised net $44 million in capital, including a $15 million registered direct offering at $1.00 per share, with proceeds directed primarily toward AI infrastructure expansion and working capital.
Privacy Infrastructure: Trusted Execution Environments
Cocoon distinguishes itself from centralized AI providers through its privacy-first architecture. The platform utilizes Trusted Execution Environments (TEEs), including Intel TDX (Trust Domain Extensions) and AMD SEV-SNP, to ensure confidential execution of AI models.
TEEs create hardware-isolated environments where sensitive data undergoes processing without exposure to external threats or unauthorized access. The system generates cryptographic attestation evidence proving that code executes within a genuine Intel TDX environment. This attestation mechanism, integrated with Intel DCAP (Data Center Attestation Primitives), provides verifiable assurance to clients that their data remains encrypted throughout the computation pipeline.
Data remains encrypted even during active processing, preventing GPU owners from accessing the content of queries or models they execute. This architecture addresses fundamental privacy concerns associated with centralized AI platforms operated by companies like OpenAI, Google, Amazon AWS, and Microsoft Azure.
AlphaTON executed a $46 million compute infrastructure investment on January 5, 2026, for 576 NVIDIA B300 chips. The transaction structure includes $4 million in cash, $32.7 million in non-recourse debt financing, and $9.3 million in equity across installments paid by full deployment in March 2026.
The company secured a five-year enterprise colocation agreement with Nordic data center operator atNorth, providing 2.2 megawatts of renewable-powered capacity for high-performance computing workloads in Sweden. AlphaTON deployed initial B200 GPUs on November 30, 2025, coinciding with Telegram’s official Cocoon network launch, followed by H200 GPUs on December 8, 2025.
As of December 2025, the Cocoon network recorded Total Value Locked (TVL) of approximately 4,487 TON. Network composition includes 30 work nodes, 18 proxies, and 12 clients. These metrics indicate early-stage development with validated operational proof.
Telegram serves as Cocoon’s first major customer. The messaging platform’s AI features, including message translation, already receive partial support from the network. Telegram’s approximately 1 billion monthly active users represent a potential adoption base for the decentralized compute network.
AlphaTON established a definitive agreement with Midnight Foundation in December 2025, designating the company as a Founding Federated Node Architecture Provider for privacy-preserving AI. This agreement generates immediate monthly recurring revenue and includes reimbursement for documented network growth costs.
The company launched AlphaTON Claude Connector on January 21, 2026, an open-source platform integrating Anthropic’s Claude AI models with TON blockchain technology via Telegram. The solution enables users to manage digital assets through natural-language conversations. AlphaTON is developing an Agentic trading and monitoring application that will utilize agents to seek best execution across multiple asset classes including tokenized securities, cryptocurrencies, equities, private shares, and real-world assets.
Technical Implementation
The Cocoon codebase resides in the public repository at github.com/TelegramMessenger/cocoon. The repository contains tools and documentation to serve and access models via the network. Workers run the Cocoon protocol stack on TEE-supported GPU servers.
The system supports both Intel TDX and AMD SEV-SNP for confidential virtual machines, with a unified abstraction layer through the TeeInterface. Deployment requires enabling TEE and preparing the GPU for confidential computing, including enabling TDX in BIOS.
Cocoon positions as a privacy-preserving decentralized AI alternative to large technology platforms such as Google Gemini and OpenAI ChatGPT. The network challenges the monopoly held by Big Tech corporations in AI infrastructure provision.
Projects like Akash Network and Render Network already enable distributed computing resource rental, though Cocoon’s integration with Telegram’s user base distinguishes it from competitors. The TON blockchain’s multi-chain architecture processes millions of transactions per second through its sharded design, providing the underlying settlement layer.
Financial Performance Indicators
The stock remains down roughly 90% from early October levels, when it traded near $7 before the October 10 crypto market liquidation that eliminated half a trillion dollars from total market capitalization. Toncoin (TON) declined approximately 38% during the same period.
The company exited SEC “baby-shelf” restrictions on December 3, 2025, filing a $420.69 million shelf registration statement declared effective December 11, 2025. This provides enhanced financing flexibility for strategic acquisitions and infrastructure deployment.
FAQ
What hardware requirements exist for GPU owners to participate in Cocoon?
GPU servers must support Trusted Execution Environments (TEE) , specifically Intel TDX or AMD SEV-SNP. Operators enable TEE in BIOS and prepare the GPU for confidential computing. The Cocoon protocol stack runs on TEE-supported GPU servers. The repository provides deployment documentation.
How does Cocoon ensure data privacy during AI inference processing?
Cocoon utilizes hardware-based Trusted Execution Environments that create isolated computational zones. Data remains encrypted throughout the processing pipeline, including during active computation. The system generates cryptographic attestation proving code execution within a genuine TEE. Neither GPU owners, Telegram, nor developers can access query contents or model parameters.
What is AlphaTON Capital’s role in the Cocoon network?
AlphaTON Capital Corp. (Nasdaq: ATON) functions as a major infrastructure provider. The company deployed initial GPU fleets in November-December 2025 and began generating revenue from AI inference processing in December. AlphaTON committed $46 million for 576 NVIDIA B300 chips and secured 2.2 MW of data center capacity. The company also established partnerships with Midnight Foundation and launched the Claude Connector integration.
What applications currently use Cocoon’s computational resources?
Telegram serves as Cocoon’s first major customer. The messaging platform’s AI features, including message translation, already receive partial support from the network. Developers building Telegram Mini Apps and bots can access the network for private AI inference. The Claude Connector enables natural-language asset management through Anthropic’s AI models integrated with TON.
What distinguishes Cocoon from existing decentralized compute networks like Akash or Render?
Cocoon’s primary differentiator is its integration with Telegram’s ecosystem and approximately 1 billion monthly active users. The network emphasizes confidential computing through TEE hardware, whereas many existing decentralized compute platforms do not enforce hardware-level privacy guarantees. Cocoon also operates specifically on the TON blockchain with native Toncoin settlement.
Isai Alexei is a journalist and financial analyst covering cryptocurrency markets and traditional securities for Blockchaindose. He has spent ten years analyzing digital assets, trading activity, and market structure.
