On September 3, Moore Thread held a performance meeting for the first half of 2026 to announce that senior management exchanged views on topics of concern to investors, such as product launch, profit inflection points, and ecological construction. Referring to the progress of GPUs in large-scale model training and inference, Zhang Jianzhong, chairman and general manager of Moore Thread, explained that in the field of large-scale AI infrastructure, the company's MTT S series intelligent computing cards have been deployed in many intelligent computing centers and cloud service platforms, widely supporting the computing power requirements of large model training, inference, intelligent applications, and scientific computing. He said that the company is one of the few GPU vendors in the market that has actually implemented large-scale commercial applications of kilocalories and 10,000-card clusters under a single network for training. Moor Thread released the “Huagang” architecture in December 2025, which supports the expansion of intelligent computing clusters with a scale of 100,000 cards or more, and builds a “model factory” infrastructure capability for large-scale model training for post-training tasks such as large model pre-training and reinforcement learning, as well as cutting-edge directions such as MoE large models, VLA type physical intelligence models, and video generation models.
Zhitongcaijing · 3d ago
Share
Listen to the news
On September 3, Moore Thread held a performance meeting for the first half of 2026 to announce that senior management exchanged views on topics of concern to investors, such as product launch, profit inflection points, and ecological construction. Referring to the progress of GPUs in large-scale model training and inference, Zhang Jianzhong, chairman and general manager of Moore Thread, explained that in the field of large-scale AI infrastructure, the company's MTT S series intelligent computing cards have been deployed in many intelligent computing centers and cloud service platforms, widely supporting the computing power requirements of large model training, inference, intelligent applications, and scientific computing. He said that the company is one of the few GPU vendors in the market that has actually implemented large-scale commercial applications of kilocalories and 10,000-card clusters under a single network for training. Moor Thread released the “Huagang” architecture in December 2025, which supports the expansion of intelligent computing clusters with a scale of 100,000 cards or more, and builds a “model factory” infrastructure capability for large-scale model training for post-training tasks such as large model pre-training and reinforcement learning, as well as cutting-edge directions such as MoE large models, VLA type physical intelligence models, and video generation models.
Disclaimer:Webull uses external vendor Google Translation Service for news translations where we endeavour to ensure these are correct, however, we recommend that you please double-check this information accordingly. Webull is not responsible for translation errors or issues.
What's Trending