Gonka network supports over 9000 GPUs with high memory and computing power, including H100/H200 models, and already supports nearly 40 GPU hardware models.

According to the latest data from gonkascan.com, Gonka's total network computing power has now exceeded 12,000 H100 equivalent GPUs, approaching the level of traditional large-scale AI computing centers. Over 9,000 high-memory GPUs (H100/H200, etc.) are used, with their proportion continuing to increase, dominating the network's effective computing power weight and enabling the network to stably support high-concurrency inference tasks for models with billions of parameters. The widespread use of mid-to-high-end GPUs has also significantly lowered the barrier to entry for nodes; as long as the minimum memory requirements are met, nodes can participate in MLNode operation. Data shows that the Gonka network supports nearly 40 GPU models, covering data center-level computing power such as H200, H100, and A100, as well as professional and consumer-grade graphics cards such as RTX 6000 Ada, A6000, A40, and RTX 4090. This forms a multi-tiered, composable global computing power supply structure, demonstrating Gonka's highly compatible and efficient decentralized computing power path, providing a more open and flexible infrastructure support for AI inference and training. Meanwhile, the daily number of users of Gonka's five major AI inference models has exceeded 3,000, with inference demand maintaining high growth. Models represented by Qwen3-235B-A22B-Instruct-FP8 have formed continuous, high-frequency, and stable calls in the Gonka network, and their usage pattern has clearly moved beyond the testing phase, approaching the early actual production workload of mainstream AI APIs. Gonka aims to efficiently integrate global GPU computing power in a decentralized manner, providing a permissionless infrastructure for AI inference and training. Incubated by US AI developer Product Science Inc., the founding team consists of the Libermans siblings, former core product leaders at Snap Inc. The project has raised over $69 million in total funding from investors including Coatue (an investor in OpenAI), Slow Ventures (an investor in Solana), Bitfury, K5, Insight, and a partner at Benchmark.