Muxi
Muxi (沐曦股份/MetaX, STAR Market: 688802) is a Chinese fabless GPU design company that builds full-stack proprietary GPGPU hardware across inference, training, graphics, and scientific computing lines, paired with its MXMACA software platform. It serves AI data centers, cloud providers, research institutes, and vertical industries across China.
- Company typePublic
- Founded2020
- HeadquartersShanghai, China
- Headcount501–1,000
- GTM typeB2B
- OfferingHardware or Manufacturing
What Muxi does
Muxi Integrated Circuits (Shanghai) Co., Ltd. (沐曦股份, MetaX, SSE STAR Market: 688802) is a Chinese fabless GPU design company founded in September 2020 in Shanghai's Pudong New District. The company designs full-stack general-purpose GPU (GPGPU) products built on a proprietary instruction set and architecture, avoiding dependence on foreign GPU IP. Its product portfolio spans four sub-brands: the 曦思 N-series (N100, N260, N300) for AI inference with integrated video codec; the 曦云 C-series (C500, C550, C500X with optical interconnect, C588, C600) for training and inference integrated workloads, with the C600 engineered on domestic advanced process technology and achieving full supply chain closure; the 曦彩 G-series (G100) for graphics rendering; and the 曦索 X-series (X206, X301, X302) for scientific computing and AI4S. These are packaged as accelerator cards, GPU servers, workstations, and three supernode topologies (C550 Shanghai Cube liquid-cooled full cabinet, C500X optical interconnect supernode, C550 3D Mesh supernode). The MXMACA heterogeneous computing platform provides the software stack, including the MacaRT inference engine, MoE optimization suite, compute-communication parallelism tools, and Auto Search engine, with full PyTorch operator support and Megatron-LM compatibility.
Muxi monetizes primarily through hardware sales of accelerator cards, servers, workstations, and supernodes, supplemented by software stack licensing and developer ecosystem services. The go-to-market is a direct enterprise field sales motion targeting AI/data center/cloud providers as the primary segment, with secondary verticals across finance, healthcare, energy, transportation, education/research, and entertainment. Distribution occurs through direct enterprise sales to research institutions and data centers (with multi-thousand-card cluster deployments in Shanghai, Beijing, Hangzhou, and Hong Kong), OEM/ODM system integration partners, and a developer/ISV ecosystem anchored by the developer.metax-tech.com community and open-source contributions on GitHub, Gitee, and CSDN. Pricing is quote-based B2B with no public list prices.
The company is led by Founder/Chairman/CEO Chen Weiliang, Co-Founder/CTO/Chief Hardware Architect Peng Li (formerly AMD's first Chinese female Fellow), and Co-Founder/CTO/Chief Software Architect Yang Jian, with a core team averaging approximately 20 years of GPU experience. Muxi raised tens of billions of RMB across Angel through Series C rounds from Sequoia Capital, ZhenFund, Matrix Partners, Lightspeed, CITIC Goldstone, CICC Capital, and state-backed funds, culminating in its December 2025 STAR Market listing at a market capitalization exceeding 300 billion yuan/HKD. Recognition includes Hurun Global Unicorn List 2023, National Little Giant Enterprise status (July 2024), and National High-tech Enterprise status (October 2024).
Muxi firmographics
Firmographics- Name
- Muxi
- Legal name
- 沐曦集成电路(上海)股份有限公司
- Website
- https://metax-tech.com
- Company type
- Public
- Founded year
- 2020
- Operating status
- Operating
- Headcount range
- 501–1,000 employees
- Short description
- Muxi (沐曦股份/MetaX, STAR Market: 688802) is a Chinese fabless GPU design company that builds full-stack proprietary GPGPU hardware across inference, training, graphics, and scientific computing lines, paired with its MXMACA software platform. It serves AI data centers, cloud providers, research institutes, and vertical industries across China.
- Ownership category
- akta.pro rank
Muxi industry classification
Industry- Product category
- GPU Chips and AI Compute Hardware
- NAICS
- Computer and Peripheral Equipment Manufacturing (33411)
- SIC
- Electronic Computers (3571)
- akta.pro industry
- Compute & Acceleration ICs (CPU/GPU/AI/FPGA/ASIC)Graphics & Accelerator Cards (GPUs, compute accelerators)Hardware, Technology & Telecom/Computing Hardware & Peripherals/PC Components & Upgrades (Motherboards, GPUs, RAM, PSUs, Cooling)AI Compiler, Runtime & Kernel Optimization Software (CUDA/ROCm/XLA, graph compilers)Hardware, Technology & Telecom/Artificial Intelligence & Machine Learning/AI Infrastructure & Compute (GPUs/ASICs, accelerators)AI/Neural Processing Units (NPUs) & ML AcceleratorsHardware, Technology & Telecom/Semiconductors & Electronic Components/Digital Logic & Processors (MCU/MPU/SoC)
Keywords
Where Muxi is headquartered
LocationHeadquarters
- HQ city
- Shanghai
- HQ country
- China
- HQ region
- Asia
Offices10 records
Markets served
Muxi business model
Business model- GTM type
- B2B
- Offering type
- Hardware or Manufacturing
- Cost components
- Technology or R&D, Supply Chain, Personnel, Operations, Marketing or Sales, Infrastructure
Revenue model
- GPU Hardware Sales (Accelerator Cards, Servers, Workstations, Supernodes): Muxi generates primary revenue through sale of its GPU product lines including accelerator cards (N/C/G/X series), GPU servers, workstations, and supercomputing clusters. The company operates a 'in-production一代, in-development一代, pre-research一代' product strategy, with N-series and C-series already in mass production and shipping.
- Software and Ecosystem Licensing: Revenue generated through MXMACA software platform, developer community services, and software stack licensing. The software stack includes MacaRT inference engine, Model Zoo, and compatibility with mainstream ecosystems (PyTorch, Megatron-LM, etc.) enabling customers to deploy existing AI workloads on Muxi hardware.
Go-to-market motion3 records
Distribution channels3 records
Marketing channels7 records
Muxi product offering
Product offeringCore offering
Muxi designs, manufactures, and sells full-stack GPU chips and integrated compute systems for AI training and inference, graphics rendering, and scientific computing. Its portfolio includes four GPU accelerator card lines (曦思 N-series inference, 曦云 C-series training-inference, 曦彩 G-series graphics, 曦索 X-series scientific AI), packaged into GPU servers, workstations, and supercomputing cabinets, and supported by the proprietary MXMACA heterogeneous computing software platform. Customers are primarily Chinese data centers, cloud providers, AI research institutes, and enterprise verticals.
Product overview
沐曦股份 (MetaX, Stock Code: 688802.SH) is a full-stack GPU chip and solution company founded in September 2020 and listed on the STAR Market in December 2025. The product portfolio is organized into four GPU accelerator card sub-brands: the 曦思N系列 (N-Series) for AI inference (including 曦思N100, N260, N300), the 曦云C系列 (C-Series) for training and inference (including C500, C550, C500X光互连, C588, C600), the 曦彩G系列 (G-Series) for graphics rendering (曦彩G100), and the 曦索X系列 (X-Series) for scientific AI (曦索X206, X301, X302). These GPUs are packaged into GPU servers (N260/N300/C500/C550/C588/C600/X206/X302服务器), GPU workstations (曦思N260工作站), and GPU supernodes (曦云C550 Shanghai Cube液冷整机柜, 曦云C500X光链超节点, 曦云C550 3D Mesh超节点). The MXMACA异构计算平台 provides the unified software stack enabling hardware-software co-design, with full PyTorch operator support, Megatron-LM compatibility, and optimization for large model training and inference.
Differentiator
Problem solved
Functional benefit
Brands
- 曦思 (Xi Si) N-Series: AI inference GPU product series including N100, N260, and N300 for AI inference applications
- 曦云 (Xi Yun) C-Series
- 曦彩 (Xi Cai) G-Series
- 曦索 (Xi Suo) X-Series
- MXMACA
Products and services
- 曦思 N100 (MetaX N100) 曦思 N100 is an AI inference GPU with GPGPU+DLA heterogeneous architecture for fast computation and high-bandwidth data movement, integrated with hardware video codec supporting 96-channel video decoding and 128-channel video encoding for AI video processing tasks. Targets AI inference customers in data centers, smart city, intelligent transportation, and industrial vision.
- 曦思 N260 (MetaX N260) 曦思 N260 is a large model inference GPU designed for high-density inference deployment in AI workloads, with high-bandwidth memory.
- 曦思 N300 (MetaX N300) 曦思 N300 is an AI inference GPU offering enhanced performance for large model inference tasks.
- 曦云 C500 (MetaX C500) 曦云 C500 is a general-purpose GPU based on proprietary GPGPU architecture and domestic advanced process technology, supporting training and inference workloads.
- 曦云 C550 (MetaX C550) 曦云 C550 is a general-purpose GPU designed for training and inference workloads with high-bandwidth memory.
- 曦云 C500X (MetaX C500X) 曦云 C500X is a GPU with optical interconnect capability designed to enable high-speed inter-GPU communication for large-scale training clusters.
- 曦云 C588 (MetaX C588) 曦云 C588 is a GPU accelerator card for compute workloads, part of the C-series product family.
- 曦云 C600 (MetaX C600) 曦云 C600 is a next-generation general-purpose GPU based on proprietary GPGPU architecture and domestic advanced process technology, achieving full supply chain domestication from design through manufacturing and packaging testing.
- 曦彩 G100 (MetaX G100) 曦彩 G100 is a graphics processing GPU for professional visualization and rendering workloads.
- 曦索 X206 (MetaX X206) 曦索 X206 is a scientific AI GPU for AI for Science workloads including materials simulation, life sciences, climate modeling, and fluid dynamics.
- 曦索 X301 (MetaX X301) 曦索 X301 is a GPU accelerator card in the X-series for scientific AI computing workloads.
- 曦索 X302 (MetaX X302) 曦索 X302 is a GPU accelerator card in the X-series for scientific AI computing workloads.
- 曦思 N260 Server 曦思 N260 Server is a GPU server configured with N260 accelerator cards, designed for AI inference workloads in data center environments.
- 曦思 N300 Server 曦思 N300 Server is a GPU server configured with N300 accelerator cards for enhanced AI inference performance.
- 曦云 C500 Server 曦云 C500 Server is a GPU server configured with C500 accelerator cards, supporting training and inference workloads.
- 曦云 C550 Server 曦云 C550 Server is a GPU server configured with C550 accelerator cards for high-bandwidth compute workloads.
- 曦云 C588 Server 曦云 C588 Server is a GPU server configured with C588 accelerator cards for compute workloads.
- 曦云 C600 Server 曦云 C600 Server is a GPU server configured with C600 next-generation general-purpose GPU cards.
- 曦索 X206 Server 曦索 X206 Server is a GPU server configured with X206 scientific AI GPU modules, suitable for AI for Science workloads.
- 曦索 X302 Server 曦索 X302 Server is a standard dual-socket server with eight X302 OAM modules, delivering multi-precision computing for life sciences, materials science, climate/ocean modeling, and fluid dynamics.
- 曦思 N260 Workstation 曦思 N260 Workstation supports up to 4 N260 actively-cooled PCIe cards, designed for flexible workspaces and high-density inference scenarios outside data center environments.
- 曦云 C550 Shanghai Cube Liquid-Cooled Full Cabinet 曦云 C550 Shanghai Cube is a flagship high-density liquid-cooled server cabinet (47U) supporting up to 128 liquid-cooled C550 cards in a single cabinet with 32-GPU super-node topology, reducing footprint by 70% compared to conventional solutions.
- 曦云 C500X Optical Interconnect Supernode C500X Optical Interconnect Supernode supports DragonFly interconnection topology through photonic-electronic hybrid interconnection, enabling up to 8 servers with 64 optical-interconnected GPUs per supernode. Based on standard PCIe servers compatible with air-cooled low-power cabinets.
- 曦云 C550 3D Mesh Supernode C550 3D Mesh Supernode supports 3D Mesh topology with low-cost, low-latency electrical interconnection for a supernode of up to 8 servers and 64 cards, with open bus technology for hyperscale data center deployment.
- MXMACA Heterogeneous Computing Platform MXMACA (MetaX Advanced Computing Architecture) is Muxi's unified heterogeneous computing platform integrating mainstream algorithm frameworks, computing libraries, toolchains, and development environments to deliver a full-stack software solution covering development, debugging, and performance optimization. It natively adapts over 6,000 international applications and 1,000+ models across AI, HPC, and big data domains, with full PyTorch operator support and Megatron-LM/DeepSpeed compatibility.
Quantifiable outcome
- Training throughput improvement of ~20% vs native Megatron-LM with MXMACA optimization suite; up to 8% improvement from MoE 'hot expert' optimization; up to 10% improvement from DualPipeV; GPU utilization improvement 7-10% vs native Megatron-LM with TP Overlap
- +3 more outcomes
Companies that use Muxi
Customer profileNamed customers3 records
Segments4 records
Ideal customer profiles2 records
Muxi technology and API
TechnologyTechnology focussed Yes
API detail
- Has API
- No
- API docs
- API detail
Core technology
AI maturity
App detail
AI capability10 records
Feature10 records
Muxi partnerships and signals
Strategic signalPartnerships
Eleven partnerships are on record, tiered major and core.
- OpenAnolis (龙蜥社区)majorMuxi formally joined the OpenAnolis community, a leading domestic Linux open-source community co-founded by major OS, chip, and cloud companies. Collaboration aims to promote software/hardware ecosystem development and drive technological innovation in open-source.
- Shanghai Tsinghua International Innovation Center / Advanced Computing Architecture Joint LaboratorycoreMuxi co-founded the Advanced Computing Architecture Joint Laboratory with Shanghai Tsinghua International Innovation Center and Beijing Inspur. The lab focuses on GPU architecture design, software and system core technologies, AI algorithm applications, and advanced computing architecture innovation research.
- Tsinghua University School of Integrated CircuitscoreJoint research collaboration on reconfigurable GPU architecture design for homomorphic cryptography and genomic sequencing acceleration. Led by Founder Chen Weiliang, a doctoral candidate at Tsinghua's School of Integrated Circuits.
- Zhejiang University - GPU Chip Design and Application Joint Research CentermajorJoint research center with Zhejiang University focused on core GPU technology development and applications. Muxi CTO Yang Jian leads the collaboration from Muxi's side.
- Shanghai Jiao Tong University Alumni EcosystemcoreThree hundred-billion yuan listings (MiniMax, Biren Technology, Muxi) emerged from Shanghai Jiao Tong University alumni ecosystem within one month. Muxi is among the major tech unicorns originating from SJTU's entrepreneurial community.
- Shanghai Jiao Tong University (ICPA智算联盟)majorMuxi CTO Yang Jian serves as guest lecturer for Shanghai Jiao Tong University's 'Introduction to AI Chips' course through ICPA Intelligent Computing Alliance. Muxi participates in university classroom programs to cultivate future chip design talent.
- NineSource (九源联合体)majorNineSource consortium focuses on domestic intelligent computing system construction, unified software ecosystem, and technology transfer. Muxi is a partner at WAIC 2026 open-source zone showcasing NineSource's AI infrastructure capabilities.
- CNCF / HAMi (Kubernetes for Heterogeneous Computing)majorMuxi participates in CNCF ecosystem and collaborates with HAMi (heterogeneous AI infrastructure open-source project led by Dynamia), supporting AI infrastructure integration and heterogeneous computing management.
- PyTorch FoundationmajorMuxi is a contributor and ecosystem partner to PyTorch open-source ML framework. The MXMACA platform provides full PyTorch operator support and compatibility with mainstream PyTorch-based models (Llama, ChatGLM, DeepSeek).
- Meituan LongCat-2.0 EcosystemmajorMeituan open-sourced LongCat-2.0 trillion-parameter LLM with Muxi曦云C series GPU providing Day 0 inference adaptation. Industry analysis suggests this open-sourcing activates existing domestic chip ecosystems and advances 'national chips + national models' from breakthrough to large-scale deployment.
- Tencent Hunyuan EcosystemmajorMuxi曦云C series GPU achieved first domestic Day 0 adaptation for Tencent Hunyuan hybrid expert model Hy3, demonstrating seamless integration with Tencent's large model ecosystem.
Scale indicators9 records
Recent moves9 records
Expansion highlights6 records
Muxi competitors and assessment
Company assessmentBroad incumbents
- NVIDIA: Global GPU/AIChip incumbent dominating training workloads with H100/B200 and CUDA software stack; the structural competitor Muxi benchmarks against in terms of perf/$ and ecosystem compatibility.
- Cambricon Technologies: Listed Chinese AI chip designer (688256.SH) with established Siyuan/Cambricon-1A/1M accelerators serving government and data center customers; competes with Muxi in the same domestic AI accelerator procurement market.
- AMD: Global GPU incumbent with Instinct MI300/MI325 accelerators and ROCm software; relevant benchmark because Muxi's CTO Peng Li was AMD's first Chinese female Fellow, and AMD's GPGPU history shapes Muxi's architectural approach.
- Huawei (HiSilicon / Ascend): Massive incumbent offering Ascend 910/310 AI accelerators integrated with Huawei Cloud and CANN software stack; dominates Chinese AI compute procurement through scale and group-level demand.
Direct peers
- Moore Threads: Chinese fabless GPU designer founded 2020; offers MTT S-series and KUAE inference platforms targeting the same domestic AI/data center customers as Muxi with comparable Day-0 adaptation claims and recent STAR Market listing.
- Biren Technology: Chinese fabless GPU startup developing training/inference accelerators (BR100/BR104 series) with similar funding milestones and Shanghai-based operations; positioned as a near-direct competitor in domestic sovereign AI compute.
Regional players
- Hygon Information Technology: Chinese HPC/CPU/GPU vendor combining x86-licensed CPU cores with domestic DCU accelerators; competes in Chinese data center procurement as a state-aligned alternative with broader product portfolio.
Emerging players
- Iluvatar CoreX: Chinese AI chip startup offering cloud training (BI) and edge inference (CI) accelerators; targets the same domestic AI compute customer base with similar Day-0 ecosystem support claims.
Market position
Strengths5 records
Weaknesses5 records
Competitive moat6 records
Key risks6 records
Key highlights7 records
Customer concentration
Muxi social profiles
Digital presenceMuxi compliance and trust
Trust signalCompliance2 records
Muxi financial estimates
Financial estimateRevenue estimate
Valuation estimate
Muxi leadership team
Management profileNumber of profiles
Profiles7 records
Muxi subsidiaries and ownership
Company hierarchySubsidiaries8 records
Muxi funding detail
Funding detailFunding overview
Funding rounds5 records
Investors20 records
Funding detail is available on the Subscription and Enterprise plan.Contact sales →
Muxi M&A and investment
M&A and investmentM&A
Investments4 records
M&A and investment is available on the Subscription and Enterprise plan.Contact sales →
Frequently asked questions about Muxi
What does Muxi do?
Muxi designs, manufactures, and sells full-stack GPU chips and integrated compute systems for AI training and inference, graphics rendering, and scientific computing. Its portfolio includes four GPU accelerator card lines (曦思 N-series inference, 曦云 C-series training-inference, 曦彩 G-series graphics, 曦索 X-series scientific AI), packaged into GPU servers, workstations, and supercomputing cabinets, and supported by the proprietary MXMACA heterogeneous computing software platform. Customers are primarily Chinese data centers, cloud providers, AI research institutes, and enterprise verticals.
Is Muxi a public or private company?
Muxi is a public company. It is classified as public and is currently operating.
When was Muxi founded?
Muxi was founded in 2020. It employs 501 to 1,000 people.
Where is Muxi based?
Muxi is headquartered in Shanghai, China, in the Asia region.
How does Muxi make money?
Two revenue lines are on record. GPU Hardware Sales (Accelerator Cards, Servers, Workstations, Supernodes) is the primary driver. The others are software and Ecosystem Licensing.
Who are Muxi's main competitors?
Broad incumbents on record are NVIDIA, Cambricon Technologies, AMD and Huawei (HiSilicon / Ascend). Direct peers are Moore Threads and Biren Technology. Hygon Information Technology is listed as a regional player. Iluvatar CoreX is listed as an emerging player.
Does Muxi have an API?
No public API is recorded for Muxi.
What industry is Muxi in?
Muxi's product category is GPU Chips and AI Compute Hardware. Its primary akta.pro industry code is HDAHABAC, Compute & Acceleration ICs (CPU/GPU/AI/FPGA/ASIC), with a secondary code of HDACAGAA, Graphics & Accelerator Cards (GPUs, compute accelerators). Its NAICS code is 33411 and its SIC code is 3571.