Moore Threads builds general-purpose computing foundation with Universal GPUs

Innovative Breakthrough: Moore Threads

Moore Threads is one of China's few companies specializing in Universal GPUs. In six years, it has developed five generations of GPU architectures and brought five chips into mass production. As of June 2026, the company had invested nearly 5.9 billion yuan ($886 million) in R&D, with R&D personnel accounting for 74.2 percent of its workforce. It had filed 2,352 intellectual property applications and secured 951 grants, while building an active developer community of more than 800,000.

Guided by its vision to "Accelerate Computing for a Better World", Moore Threads provides advanced infrastructure and one-stop solutions for accelerated computing globally. It aims to build a general-purpose computing foundation for the AI era and enable the digital transformation across various industries.

MUSA architecture enables full functionality on a single chip

Built on its proprietary MUSA architecture, Moore Threads' Universal GPUs integrate AI acceleration, graphics rendering, physical simulation, scientific computing, and ultra-high-definition video encoding and decoding on a single chip. This creates an integrated platform for computing, rendering and simulation.

The company introduced China's first graphics acceleration engine supporting DirectX 12 and pioneered native support for AI generative rendering and hardware-accelerated ray tracing. Moore Threads is among the few domestic GPU makers whose products support the full range of computing precisions, from FP8 to FP64, and offer GPU capabilities for AI model training.

Full-stack products cover cloud, edge and device computing

MTT KUAE computing cluster

Moore Threads has built a full-stack solution covering cloud, edge and device computing, bringing general-purpose computing to a wide range of AI computing scenarios.

In cloud computing, its flagship MTT S5000 GPU for large model training, inference, and high-performance computing is now in mass production. It supports full-precision, general-purpose computing and was among the first products to pass China's security and reliability assessment, demonstrating strong technical maturity, stability and security.

The MTT KUAE 10,000-GPU computing cluster, built on the MTT S5000, has been deployed and is now operational, with several key performance indicators reaching international mainstream levels. Moore Threads is also working with leading internet companies to build 100,000-GPU computing clusters.

MTT AIBOOK (right) and MTT AICUBE

For edge and device computing, Moore Threads has developed its proprietary AI SoC, Changjiang SoC. Its products include MTT AIBOOK AI computing book, the MTT AICUBE home AI hub and the MTT E300 edge AI module, bringing AI into homes and industries.

The company has also launched the MTT S80 and S70 gaming GPUs, China's first graphics cards to support Windows and DirectX 11/12. They run leading AAA games such as Black Myth: Wukong and support major graphics and computing ecosystems, including OpenGL and Vulkan.

Three AI factories turn computing power into productivity

Built on the KUAE computing cluster, Moore Threads is developing three AI factories: a model training factory, a token production factory and an AI agent production factory. All three are moving toward large-scale application.

The model training factory has delivered results in training domestic models on domestic chips, covering foundation models, embodied AI models and world models.

The token production factory supports major open-source models, including DeepSeek, GLM, MiniMax, Kimi and Qwen, from launch. Moore Threads has also introduced a prefill-decode disaggregated inference solution based on the MTT S5000. Its overall cost-performance exceeds that of leading international computing solutions, and it has already handled live token traffic for leading model developers.

The AI agent production factory features Xiaomai, a proprietary general-purpose AI agent, and MT Lambda, a full-stack embodied AI simulation platform. Moore Threads is among the few domestic GPU companies capable of supporting the entire embodied AI training process.

As one of the few domestic GPU companies serving both enterprise and consumer markets, Moore Threads follows a dual ecosystem strategy: developing its proprietary MUSA platform while maintaining compatibility with mainstream international ecosystems, including CUDA. It is building an open, accessible and comprehensive domestic GPU ecosystem.

The company will continue to support internet and cloud service providers, token service providers, AI innovators, telecom operators, intelligent computing centers, research institutions, government and enterprise clients, and other industry customers. Together with its partners, it aims to strengthen China’s domestic computing ecosystem.