In a milestone for China's domestic semiconductor ambitions, Moore Threads has successfully integrated its flagship MTT S5000 GPU with China Mobile’s proprietary 'Jiutian' 35B large language model. This partnership marks the first time a domestically produced GPU has supported the large-scale deployment of an artificial intelligence framework developed by a central state-owned enterprise (SOE). The achievement signals a significant maturation of the Chinese AI ecosystem as it seeks to decouple from Western hardware dependencies.
The technical verification involved the full-process adaptation and inference validation of the 35-billion-parameter Jiutian model. By utilizing Moore Threads’ proprietary MUSA software stack, the MTT S5000 demonstrated its capacity to handle high-concurrency inference requests while maintaining low latency. This is particularly critical for China Mobile, which operates the world’s largest telecommunications network and requires high-reliability deployments across diverse industrial sectors.
This collaboration arrives at a pivotal moment as US export controls continue to restrict the flow of high-end NVIDIA and AMD chips into the Chinese market. Beijing has responded by mandating that its 'National Team' of state-owned firms prioritize domestic components. By successfully powering a major SOE’s AI initiative, Moore Threads is positioning itself not just as a hardware vendor, but as a viable architect for the nation’s sovereign computing infrastructure.
However, the ultimate success of this transition depends on the software ecosystem. The MUSA architecture aims to provide a seamless alternative to NVIDIA’s dominant CUDA platform, which has long been the industry standard. While compatibility with a 35B model is a promising benchmark, the true test will lie in whether Moore Threads can scale this performance to the trillion-parameter models that define the current global frontier of generative AI.