China finds a way to bypass US sanctions in the AI race

China's Z.AI (formerly Zhipu) has launched a massive data center with a capacity of 1 GW, designed for training AI models. The most significant aspect of this project is that the complex was built entirely on Chinese-made technologies without a single NVIDIA chip. This event serves as clear proof that strict export restrictions imposed by Washington cannot stop China's technological development. This is reported by Ixbt.com reports .

According to Bloomberg, the new data center will serve to improve AI models in the GLM (General Language Model) series. The scale of the complex's energy consumption is staggering: 1 GW of power is enough to meet the simultaneous needs of approximately 750,000 households. This figure makes the facility one of the largest computing centers in China.

Victory for local chips

Currently, Z.AI is operating several computing clusters. Each of them contains over 10,000 chips, and sources confirm that these systems do not use NVIDIA products. Although it has not been disclosed which manufacturer's processors are installed, experts speculate that the main suppliers are Huawei, Cambricon, or Alibaba.

For a long time, the prevailing view in the global technology community was that Chinese chips were only capable of inference, but not powerful enough for training complex systems from scratch like NVIDIA. The Z.AI project shatters this stereotype, demonstrating that local hardware can handle even the most complex tasks.

Resource scarcity and future plans

This news comes after Beijing-based Moonshot introduced its Kimi K3 model. Notably, although Kimi K3's capabilities could compete with leading US analogs, it was forced to temporarily stop registering new users due to a lack of computing power. Z.AI's investment in its own infrastructure is aimed at preventing such problems.

The Chinese government views independence in the field of AI as a strategic goal. Therefore, it is planned to invest approximately 2 trillion yuan (around $295 billion) to build new data centers across the country over the next 5 years. These funds will be directed not only to strengthening the technical base but also to making software and the ecosystem independent of NVIDIA standards.

This trend is also important for rapidly digitizing countries like Uzbekistan. In the context of the global deficit and high prices for NVIDIA chips, China's alternative and competitive solutions may open new doors for regional IT projects.