Z.AI, the Chinese artificial intelligence startup, has completed a 1 GW data centre that operates exclusively with Huawei's Ascend chips, completely avoiding Nvidia. The company expects to reach $1 billion in annual recurring revenue.
The Chinese artificial intelligence company Z.AI, formerly known as Zhipu AI, has finished constructing a data centre with a capacity of 1 gigawatt (GW) that operates entirely with Huawei's Ascend 910B accelerators. This milestone, reported by Bloomberg and covered by The Next Web, marks a significant step in the technological sovereignty strategy of the Asian country.
The data centre, whose power multiplies by ten or twenty that of typical hyperscaler facilities (between 50 and 100 MW), is designed for large-scale AI model training. Z.AI uses it for its family of GLM models, which directly compete with those of OpenAI and other Western firms.
The company is on a trajectory of rapid growth: according to data from HokAI and projections from JPMorgan, its revenues increased from $42 million in 2024 to an estimated $168 million for 2025. Forecasts point to 4.6 billion yuan (about $640 million) in 2026 and 30.9 billion yuan (approximately $4.3 billion) in 2028.
Z.AI is poised to become the first independent Chinese AI company to reach $1 billion in annual recurring revenue, a goal that analysts consider within reach.
Total sovereignty strategy
The decision to forgo Nvidia is a response to the trade restrictions imposed by the United States, which limit the export of advanced GPUs like the H100 and A100 to China. Faced with this scenario, Z.AI opted for a path of total sovereignty, exclusively employing Huawei's Ascend 910B accelerators.
This move is not only geopolitical but also pragmatic: Huawei's chips offer guaranteed availability within China, integration with local infrastructure, and government support. The trade-off is that the Ascend software ecosystem is less mature than CUDA, Nvidia's environment, which requires greater optimisation efforts from Z.AI.
The result, however, is a technology stack completely independent of foreign regulatory decisions, a value that the company exploits to access government contracts and align with the national agenda of technological sovereignty.
Business model and positioning
Founded in 2019 by researchers from Tsinghua University, Z.AI already has over 100,000 enterprise clients. Its business model combines the publication of open weights (licensed under MIT on Hugging Face) with an aggressive pricing strategy: it offers free access to its best models while monetising enterprise APIs and managed services.
Its most advanced model, the GLM-5.2, has 750 billion parameters and a context window of 1 million tokens, making it particularly competitive for coding tasks and processing long documents. According to Reuters, it positions itself as a viable alternative for companies looking to reduce their dependence on Western suppliers.
Z.AI faces competition in the Chinese market from tech giants like Alibaba (with its Qwen family), Tencent and Baidu, as well as other 'tiger' startups like DeepSeek, Moonshot AI, 01.AI and MiniMax. Its key differentiation lies in its commitment to open models, focus on coding and agents, and alignment with the Chinese government's industrial policy.
Lessons for the entrepreneurial ecosystem
The case of Z.AI offers lessons for AI startup founders worldwide. Diversifying infrastructure providers is no longer a strategic option but a resilience requirement, especially in a context of increasing geopolitical tensions.
Moreover, the downward pressure on model prices exerted by Z.AI and its Chinese competitors could benefit Spanish-speaking and LATAM startups, which could access high-level AI capabilities at reduced costs. Competition between technological ecosystems, far from being detrimental, expands the range of options for developers.
The company completed its IPO in Hong Kong in January 2026, with a valuation of $6.55 billion, after raising between $1.3 billion and $2 billion in previous rounds.
You may also like
