China completes 1 GW AI data center using only domestic semiconductors, demonstrating successful deployment despite U.S. export controls.

Chinese artificial intelligence startup Zhipu AI (Z.AI) has completed and commenced full-scale operations of a massive 1-gigawatt (GW) data center built exclusively with domestically manufactured semiconductors. The project represents a direct challenge to US export controls on advanced chips and a symbolic milestone demonstrating the accelerating technological independence of China's AI ecosystem.

According to Bloomberg, Zhipu AI has finished construction and begun partial operations of the 1GW data center designed to support the development of its large language model (LLM) "GLM" series. One GW represents enough power to supply approximately 750,000 households simultaneously, making it the largest data center built by a Chinese AI company. The facility is reported to utilize over 10,000 Chinese-made AI semiconductors, though the specific location and total investment were not disclosed.

The launch of this data center represents the culmination of efforts by Chinese AI firms—long heavily reliant on Nvidia's high-performance graphics processing units (GPUs)—to secure domestic chips and independent infrastructure in response to tightening US export controls. Industry observers see the facility as a litmus test for whether China can build a self-sufficient AI computing ecosystem despite Washington's restrictions on advanced semiconductor exports.

Zhipu AI has also reportedly finalized the acquisition of "Zhongke Jiahe," a Chinese AI infrastructure software company founded by researchers from the Chinese Academy of Sciences. The firm possesses technology that enables the same AI model to run efficiently across different Chinese AI chips, such as those from Huawei and Cambricon. Industry analysts believe the acquisition was aimed at improving the operational efficiency of Zhipu AI's data center.

China's push for AI semiconductor independence extends beyond Zhipu AI. Huawei recently unveiled its next-generation AI supercomputing system, the "Atlas 950 SuperPod," which links thousands of its self-developed AI chips via an ultra-high-speed network to function as a single, massive AI entity. Proprietary packaging and interconnect technology significantly reduces reliance on Nvidia. Major cloud and telecommunications companies like Alibaba and China Telecom are also actively investing in data center construction, supporting the expansion of AI computing infrastructure.

The Chinese government is providing full-scale backing. It plans to invest approximately CNY 2 trillion (approximately $295.6 billion) over the next five years to promote nationwide data center construction. Around 20 Chinese AI semiconductor firms, including SenseTime, Huawei, Cambricon, and Moore Threads, recently launched the "Galaxy Initiative," which aims to jointly build a token operations center, establish five AI GPU clusters of 10,000 cards each, and support 200 AI startups.

China's AI surge is equally prominent on the software side. The open-source large language model "Kimi K3," released by Chinese startup Moonshot AI, has been recognized as competitive with top-tier US models in global AI performance evaluations. Kimi K3 is one of the world's largest open-source models, with 2.8 trillion parameters, and can process contexts of up to 1 million tokens at once. The launch triggered such a surge in users that new sign-ups had to be temporarily suspended. The model claimed the top spot on the global AI evaluation platform Chatbot Arena, surpassing Anthropic's "Claude Opus 5."

Alibaba also released a preview version of its AI model "Qwen 3.8 Max," claiming it is "one of the most powerful models currently available." Featuring 2.4 trillion parameters, the model has been transitioned to an open-weight format, allowing enterprises to run it on their own servers or modify it for specific purposes. With DeepSeek also expected to release the official version of its next-generation V4 model this month, competition among Chinese AI models is set to intensify further.

The successive adoption of open strategies by Chinese AI models is noteworthy. Following Kimi K3's open-source release, Qwen 3.8 Max has adopted an open-weight approach, contrasting with the US, where closed models like OpenAI's GPT, Anthropic's Claude, and Google's Gemini dominate. Analysts interpret this as a strategy by China to proliferate derivative models and services based on open models, aiming to preempt global technical standards and seize leadership in the AI ecosystem.

As China's high-tech offensive intensifies, the US government has begun considering a "ban on the use of Chinese AI." According to Axios, the Trump administration is currently reviewing options to legally restrict American companies from accessing or using Chinese AI models. Measures under discussion include applying government procurement restrictions or transaction limitation rules to US firms using Chinese AI models, or operating a de facto licensing system. This marks an expansion of the US-China AI competition from previously focusing on blocking semiconductor equipment exports to China to now blocking "service access" to prevent cost-competitive Chinese AI from penetrating the US market.

The Financial Times reported that the Chinese government is also considering tightening export controls on its domestic AI and semiconductor technologies, signaling the beginning of full-scale mutual containment between the US and China over AI technology.

The rapid progress of Chinese AI models has simultaneously exposed limitations in computing infrastructure. Moonshot AI's suspension of new sign-ups for Kimi K3 is analyzed as a result of its inability to secure enough computing chips to handle surging demand. Lian Jye Su, Chief Analyst at technology research and advisory group Omdia, noted that "Kimi K3 is a very computationally intensive model, so its operating costs are also substantial."

However, the market views this "computing bottleneck" as potentially beneficial for the AI semiconductor sector. Latest AI models require massive increases in high-bandwidth memory (HBM) and high-performance GPU usage not only for training but also for inference. Contrary to concerns raised during DeepSeek's emergence that "AI efficiency would reduce semiconductor demand," the intensifying super-scale AI race is bolstering expectations for expanded AI infrastructure investment.

China's AI technological advancement is yielding tangible results in the real economy. According to Capital Economics, China's electronics and information technology sector accounted for more than half of the country's quarter-on-quarter economic growth in the second quarter of this year. Estimates from China International Capital Corporation (CICC) suggest that AI-related exports alone contributed 1.1 percentage points to nominal GDP growth over a four-month period this year—a figure three times the total growth rate for all of last year.

Zhipu AI reportedly achieved its annual revenue target early this month, with its annualized recurring revenue (ARR) approaching $1 billion, underscoring AI's emergence as a new engine for China's economic growth.