TLDRs
- Chinese tech giants formed AI alliances to bolster domestic chip and LLM integration.
- Huawei’s new CloudMatrix system challenges Nvidia’s performance benchmarks.
- US sanctions drive Chinese AI firms to prioritize national tech self-sufficiency.
- Huawei eyes global exports with older AI chips despite limited interest abroad.
Chinese technology leaders have unveiled two major AI alliances aimed at fortifying the country’s domestic tech ecosystem and minimizing reliance on U.S. components.
These moves, made at the World Artificial Intelligence Conference (WAIC) held in Shanghai on Monday, July 28, come amid increasing pressure from U.S. sanctions that have disrupted Chinese access to advanced chips and model training infrastructure.
The first alliance, named the “Model-Chip Ecosystem Innovation Alliance,” seeks to bridge the gap between large language model (LLM) developers and AI chipmakers. The goal is to create a tightly integrated ecosystem that boosts efficiency and innovation in both hardware and software development.
Founding participants include major chipmakers such as Huawei, Biren, Moore Threads, and Enflame, all of which have been impacted by U.S. export restrictions. The alliance was initiated by StepFun, an emerging force in LLM development.
Chamber Initiative Targets Industrial AI
Contents
The second initiative, the Shanghai General Chamber of Commerce AI Committee, focuses on integrating AI into China’s broader industrial sectors. This group includes prominent AI companies such as SenseTime, MiniMax, and hardware players Metax and Iluvatar CoreX.
SenseTime, once known primarily for facial recognition tech, has shifted its core strategy toward LLMs in response to export restrictions. These firms now aim to align AI capabilities with industrial transformation goals under Beijing’s broader “intelligent manufacturing” vision.
This alliance represents not just a strategic pivot but a long-term national objective: building a fully self-sufficient AI supply chain from silicon to software.
Huawei Showcases Cutting-Edge CloudMatrix
The WAIC also featured a wave of hardware innovation. Notably, Huawei unveiled CloudMatrix 384, a supercomputing system powered by 384 of its Ascend 910C chips, a homegrown alternative to U.S.-made GPUs. Research firm SemiAnalysis noted that the system outperformed Nvidia’s GB200 NVL72 in select benchmarks, signaling China’s maturing chip design prowess.
Metax also made headlines by debuting its AI supernode, equipped with 128 C550 chips, which is tailored for large-scale data center workloads.
These announcements underscore China’s determination to meet or exceed Western tech standards through domestic innovation.
Huawei Eyes Overseas Markets with Legacy Chips
Beyond its domestic strategy, Huawei is also exploring limited exports of its older Ascend 910B AI chips to countries like the UAE, Saudi Arabia, Thailand, and Malaysia.
According to reports, Huawei is offering a few thousand units to universities and AI research labs, including the Mohamed bin Zayed University of AI in Abu Dhabi.
Although the interest so far has been tepid and no deals have been finalized, these exports are part of a broader strategy to find markets for earlier-generation technology while reserving the advanced 910C chips for Chinese firms that cannot access cutting-edge U.S. chips due to sanctions.
