Muxi XiYun C600: China's First Fully Domestic GPU Outperforms Nvidia H200

Chinese company Muxi Technology has won top industry recognition for its advanced XiYun C600 accelerator, designed specifically for artificial intelligence training and heavy-duty computing. Positioned as the first high-performance GPU built entirely within a domestic supply chain, the device relies completely on local architecture, IP cores, manufacturing, and software solutions.

• Muxi XiYun C600 is equipped with 144 GB of HBM3e memory, exceeding the 141 GB capacity of the Nvidia H200.
• The hardware features fully indigenous architecture, instruction sets, packaging, and manufacturing.
• The MXMACA software platform supports over 6,000 CUDA applications for seamless transition.
• The GPU has passed safety certifications and entered mass production for regional data centers.

Technical Specifications and Architecture

The new accelerator supports the FP8 compute format alongside specialized instructions for tensor transposition, which are vital for efficient AI workload processing. By offering 144 GB of high-speed HBM3e memory, Muxi successfully outpaces Nvidia's H200 model, which features 141 GB. This high capacity allows handling massive datasets and larger neural network models locally.

To connect multiple units into powerful computing superclusters, Muxi developed its proprietary MetaXLink technology. Offering performance comparable to Nvidia's interconnect solutions, MetaXLink enables linking up to 64 cards in a single system and scaling clusters to tens of thousands of GPUs for demanding computational tasks.

Software Ecosystem and Market Deployment

To ease adoption and migration from foreign hardware, Muxi built the MXMACA software platform, which ensures compatibility with more than 6,000 CUDA-based applications. This significantly lowers entry barriers for enterprises looking to integrate domestic accelerators into existing computational infrastructures.

Following successful safety and reliability certifications in May 2026, the MXC600 entered full mass production. The accelerators are already deployed in regional computing centers, financial institutions, healthcare, and large-scale simulation projects. Muxi reported that by the end of 2025, it had shipped over 55,000 GPUs to multiple national computing clusters.

Strategic Value and Industry Impact

The rise of the Muxi XiYun C600 marks a significant milestone in achieving technological self-reliance for China's high-performance computing sector. By eliminating reliance on foreign supply chains, local developers can continue expanding AI infrastructure securely and independently.