China is now home to 1,509 AI models, which accounts for more than 40% of the world’s total, according to the World Artificial Intelligence Conference (WAIC) in Shanghai, showing the country’s great leap forward in the fast-growing technology. There are 3,755 AI models worldwide, according to the WAIC. Tencent Holdings and SenseTime launched new AI models at the conference. Tencent’s Hunyuan 3D World Model 1.0 showed its prowess in generating detailed three-dimensional environments, while SenseTime touted improvements in reinforced learning efficiency delivered by SenseNova V6.5. “We see AI enabling scalable and cost-efficient production of multimodal content across text, images, audio and video,” said UBS Securities Analyst Wei Xiong, who pointed out that Chinese models were “showing early success in AI video generation”. The number of AI models reflects the country’s big strides in transforming into an AI powerhouse through various private and publicly backed open-source development initiatives, narrowing the gap with the U.S.
The open-source approach gives the public access to a program’s source code, allowing third-party software developers to modify or share its design, fix broken links or scale up its capabilities. According to the American benchmarking platform LMArena, created by researchers from the University of California, Berkeley, China has become home to the world’s top open-source AI models.
While American AI giants such as OpenAI and Meta Platforms did not take part in this year’s WAIC, which concluded on July 28, Chinese Big Tech firms from Alibaba Group Holding to Huawei Technologies showcased their latest AI-related developments. Hangzhou-based Alibaba unveiled its latest large multimodal model, powered by its Qwen AI model family, which was specifically designed for intelligent automotive cockpits. This was jointly developed with U.S. mobile chip firm Qualcomm and Banma, Alibaba’s intelligent cockpit solution provider.
On the AI hardware side, domestic semiconductor enterprises – including Huawei and start-ups such as Moore Threads and Tencent-backed Enflame – displayed their latest AI processor technologies. Huawei, which occupied a central space at the Shanghai World Expo Exhibition and Convention Center, showed to the public for the first time its Supernode 384 system – which is a cluster of 384 Ascend AI processors – spread across 12 computing cabinets and four bus cabinets. That system delivers 300 petaflops of computing power and 48 terabytes of high-bandwidth memory, gaining attention in China as an alternative to Nvidia’s NVL72 system.
WAIC also showed a rising trend among domestic AI infrastructure suppliers to join forces in building commercial computing resources amid the strong demand for advanced computing power in China’s AI sector. SenseTime unveiled a “compute mall” initiative with more than 10 domestic partners – including chip design firms Huawei, Hygon Information Technology, Cambricon Technologies, Biren Technology and Moore Threads – that would allow AI developers to freely combine and allocate a variety of computing resources, platform tools and AI model services just like “purchasing goods” at a supermarket. “The overall share of domestically produced computing power remains very low, likely even below 10%,” said Chen Daliang, CEO of Suanova – a Shanghai-based AI computing resources provider and a subsidiary of Hong Kong-listed Yeebo Technology.
