On September 3, Moore Threads convened its performance review meeting for the first half of 2026. During the event, company executives delved into topics of keen interest to investors, including product deployment strategies, anticipated profitability milestones, and the progress of ecosystem development. Zhang Jianzhong, the Chairman and General Manager of Moore Threads, shed light on the latest breakthroughs in GPU technology tailored for large-scale model training and inference.

He highlighted that, within the realm of expansive AI infrastructure, Moore Threads' MTT S Series AI computing cards have been successfully integrated into numerous AI computing centers and cloud service platforms. These deployments effectively address the computational demands for extensive model training, inference processes, intelligent agent applications, and scientific computing tasks.

Zhang underscored Moore Threads' unique position in the market as one of the select few GPU providers capable of commercializing large-scale clusters boasting thousands, or even tens of thousands, of cards. Furthermore, Moore Threads unveiled its 'Huagang' architecture in December 2025. This innovative architecture facilitates the expansion of AI computing clusters to accommodate over 100,000 cards. Its primary objective is to establish the foundational infrastructure for a 'model factory' dedicated to ultra-large-scale model training in cutting-edge domains. These domains encompass post-training tasks for large model pre-training and reinforcement learning, as well as MoE large models, VLA-type embodied intelligent models, and video generation models.