MetaX
Category: AI Infrastructure
A leading Chinese fabless semiconductor company specializing in high-performance GPGPUs (General-Purpose Graphics Processing Units) designed for AI training, inference, and large-scale heterogenous computing. MetaX was founded in 2020. The company is led by Chen Weiliang. Based in Shanghai, China. Team size: 800-1000. Total funding raised: $960.0M. Latest round: IPO (Shanghai STAR Market, Ticker: 688802). Key investors include HongShan (Sequoia China), Matrix Partners China, CICC Capital, China Structural Reform Fund, Chaos Investment, ZhenFund.
- Founded
- 2020
- Headquarters
- Shanghai, China
- Team size
- 800-1000
- Total funding
- $960.0M
Value proposition
Offers high-efficiency, high-versatility domestic GPU solutions with an independent architecture and instruction set that provides seamless compatibility with mainstream software ecosystems (e.g., CUDA) to facilitate domestic chip substitution.
Products and solutions
MXN Series (XiSi / 曦思): Specialized GPUs for high-performance AI inference and media processing (e.g., N100, N260, N300)., MXC Series (XiYun / 曦云): Flagship training and general-purpose computing GPUs (e.g., C500, C550, C500X, C588, C600 with HBM3e support)., MXG Series (XiCai / 曦彩): High-performance GPUs dedicated to graphics rendering and visual computing (e.g., G100)., MXS Series (XiSuo / 曦索): New X-series for specialized computing (e.g., X206, X301, X302)., MXMACA: A full-stack software platform compatible with mainstream computing ecosystems and programming models., GPU Servers: N260 Server, N300 Server, C500 Server, C550 Server, C588 Server, C600 Server, X206 Server, X302 Server., GPU Supernodes: C550 Shanghai Cube liquid-cooled cabinet, C500X Optical Interconnect Supernode, C550 3D Mesh Supernode.
Unique value
Developed a completely independent GPU architecture and instruction set from the ground up, moving away from reliance on third-party IP or standard licensed architectures.
Target customer
Cloud service providers, data center operators, government research institutions, and enterprises in AI and high-performance computing (HPC).
Industries served
Artificial Intelligence (Large Model Training/Inference), Cloud Computing & Data Centers, Smart City Infrastructure, Autonomous Vehicles, Digital Twins & Metaverse, Scientific Research (HPC)
Technology advantage
Features 'MetaXLink' high-speed interconnect technology for multi-card scaling and a highly optimized hardware-software synergy that allows for rapid migration of AI workloads from international standards like NVIDIA CUDA.
How they differentiate
MetaX differentiates through a completely independent GPGPU architecture and instruction set, moving away from reliance on third-party IP. Its 'MetaXLink' high-speed interconnect technology enables ultra-large-scale '万卡' (10,000-card) clusters for LLM training, and it emphasizes high software compatibility with the CUDA ecosystem to minimize migration costs.
Main competitors
Moore Threads (摩尔线程), Biren Technology (壁仞科技), Enflame Technology (燧原科技)
Key partnerships
Zhipu AI (Beijing Zhipu Huazhang): Strategic collaboration for the adaptation of large language models (e.g., ChatGLM, GLM-5.2 Day 0 adaptation June 2026)., Gitee AI: Partnership to run and optimize open-source models like DeepSeek-V3/R1 on MetaX hardware., Inspur & SuperCloud: Server infrastructure and distribution partnerships., ShanghaiTech University: Joint research and talent cultivation initiatives., Alibaba Cloud / Qwen (千问): Day 0 adaptation of Qwen3.6-35B-A3B on MetaX hardware, collaboration with FlagOS for multi-chip deployment (April 2026)., MiniMax: Day 0 adaptation of MiniMax M3 flagship model on MetaX C-series GPUs (June 2026)., ModelScope (魔搭社区): Co-hosted open-source challenge with Swift framework on MXMACA (June 2026)., SGLang / Alibaba Cloud / Tencent Cloud: Co-hosted technical meetup on AI inference (June 2026)., Shanghai AI Lab: Multi-chip heterogeneous inference collaboration via DeepLink open-source framework (June 2026)., InfiniTensor / InfiniCCL: First-adopter adaptation of InfiniCCL v0.1.0 communication framework (June 2026).
Notable customers
Zhipu AI, Gitee AI, China Telecom, Inspur, SuperCloud, Alibaba Cloud (Qwen), MiniMax, Shanghai AI Lab (BAAI)
Major milestones
Successful launch and mass production of the XiSi N100 (Inference) GPU in 2023, Mass production of the flagship XiYun C500 (Training) GPU in early 2024, Official listing on the Shanghai Stock Exchange STAR Market in December 2025 (Stock Code: 688802.SH), Adaptation of DeepSeek-V3 and ChatGLM large language models on proprietary hardware, Launch of C600 next-gen GPU with HBM3e on domestic advanced process (2025), Announced Hong Kong IPO plan for H-share listing (June 2026), Day 0 adaptation of Alibaba Qwen3.6, MiniMax M3, and Zhipu GLM-5.2 flagship models (2026)
Growth metrics
Achieved 2024 revenue of 740M RMB ($105M), 2025 revenue of 1.644B RMB ($230M, +121% YoY), and Q1 2026 revenue of 562M RMB ($78M, +75% YoY). Market capitalization ~2.78T RMB ($390B) as of June 2026.
Market positioning
Primary domestic challenger to NVIDIA's high-end data center GPUs (A100/H100) in the Chinese AI training and inference market.
Geographic focus
Mainland China (core focus on domestic substitution), expanding toward Asia-Pacific data center markets.
Patents and IP
Over 210 patent applications filed with more than 120 granted as of late 2024, with 97% of granted patents being invention-based.
About Chen Weiliang
Chen Weiliang is a semiconductor industry veteran with over 20 years of experience. Prior to founding MetaX, he spent 15 years at AMD, where he served as a Senior Director and the global head of GPGPU (General-Purpose Graphics Processing Unit) design. During his tenure at AMD, he led the successful design and mass production of 15 generations of high-performance GPU products and was instrumental in establishing AMD's R&D capabilities in China.
Latest news about MetaX
More AI Infrastructure companies
Official website: https://www.metax-tech.com/