Chinese GPU maker Moore Threads posts H1 revenue exceeding full-year 2025

The company attributed the growth to strong market demand for full-function GPUs, coupled with accelerated deployment and stable supply of its Kuae AI computing clusters. Moore Threads said it has been penetrating key customer segments such as internet and telecom operators, with richer application scenarios driving rapid revenue growth.

R&D spending in the first half totaled 769 million yuan, up 38.16% year-on-year. Since 2022, cumulative R&D investment has reached nearly 5.9 billion yuan. As of June 30, 2026, the company had filed 2,167 patent applications, including 2,049 invention patents, and obtained 788 granted patents, including 725 invention patents.

The company's flagship AI training-inference card, the MTT S5000, has entered volume production and passed the national "Safe and Reliable Evaluation." Its computing clusters have been deployed in Beijing, Wuxi, Hangzhou and other cities, maintaining up to 95% linear scaling efficiency in large-scale training and over 90% effective training time with checkpoint resume support.

Moore Threads said its clusters have trained a 236-billion-parameter MoE foundation model with over 25 trillion tokens, fully trained a 5D world model "Peking University EvoPhys-World," and open-sourced a code model MusaCoder with kernel generation performance comparable to international state-of-the-art.

Beyond cloud computing, the company's product line extends to edge and terminal AI through modules such as the MTT E300, the MTT AIBOOK and AICUBE. In embodied intelligence, it has established an innovation center in Wuxi and participates in a national AI application pilot base.

The company's MUSA ecosystem, compatible with CUDA, now supports 100% core math libraries, over 3,000 PyTorch operators and 55 core AI operators, with official support from vLLM and SGLang. Developer numbers have surpassed 800,000.