AlphaTON Capital has announced agreements to acquire and deploy a high-performance compute cluster comprising over 1,000 Nvidia B200 graphics processing units (GPUs), valued at approximately $68 million. The initiative establishes AlphaTON as a core compute provider for Telegram’s Cocoon AI, which it describes as the world’s largest decentralized, privacy-preserving artificial intelligence (AI) network. The company also aims to generate revenue through high-performance compute and GPU rental services.
According to AlphaTON, the project’s $82.5 million total capitalization includes $30 million in equity, $52.5 million in debt financing facilitated by Vertical Data, and allocation for working capital and deployment costs. The Nvidia B200 GPU assets will be recorded on AlphaTON’s balance sheet as revenue-generating infrastructure. The debt is structured for full amortization over 36 months, with a requirement to maintain a debt service coverage ratio above 1.0x throughout the period.
AlphaTON has partnered with several infrastructure specialists for deployment and operations. CUDO Compute, a European Nvidia Cloud Partner, will handle installation, orchestration, workload management, and operations, and will commercialize the GPUs via its marketplace and AI services platform. SNET Energy UK contributes expertise in sustainable-power data center development and AI infrastructure optimization. Vertical Data both arranged the project’s debt facility and is supplying the cluster hardware and deployment services.
The new cluster features over 1,000 Nvidia B200-class GPUs and is built for modularity and upgradability to keep pace with evolving AI hardware and model requirements. AlphaTON targets 90 percent utilization on the cluster, with priorities including dedicated compute for Cocoon AI workloads on Telegram and additional external GPU rentals facilitated by CUDO Compute. For revenue, AlphaTON cites two main channels: reserved capacity for Cocoon AI on a privacy-preserving network, and broader commercialization through bare metal GPU rental and AI services.
Operational and financial projections are robust: AlphaTON estimates an internal rate of return (IRR) of 59.7 percent, net present value (NPV) of $59.6 million, and a 5x equity multiple. The forecast includes a 90 percent utilization rate, margins from 64 to 73 percent, and Year 1 gross profit margins of 32 percent, assuming annual price compression and operating cost inflation. A successful pilot using Cocoon AI was completed on November 25, 2025. “Our strategic partnership with Telegram and Cocoon provided early access to the technology, and we successfully launched a Cocoon pilot on November 25, 2025,” said Logan Golema, CTO of AlphaTON Capital. “Following this validation, we are now going to deploy the technology into our secure and decentralized AI hosting infrastructure to further enhance our network’s scalability and performance.”
Deployment milestones include cluster launch and Cocoon AI testing in Q4 2025; hardware delivery, workload validation, and benchmarking in Q1 2026; and expansion of orchestration capabilities in Q2 2026, with up to four additional clusters planned depending on demand and performance. This expansion comes as market analysts project rapid growth of data center AI workload capacity between 2025 and 2030.
Source: AlphaTON Capital
