Zhipu, a Chinese artificial intelligence (AI) model developer, has reportedly completed the construction of a massive data center that exclusively uses domestically produced semiconductors and has begun partial operations.

Bloomberg reported on July 20 (local time), citing sources, that Zhipu has finished building and started partial operations of a 1-gigawatt (GW) data center designed to support the development of its advanced large language model (LLM), the GLM series.

Bloomberg explained that 1 GW is enough power to supply approximately 750,000 households simultaneously, making it one of the largest server facilities ever built by a Chinese AI research firm.

While the specific location and investment amount have not been disclosed, it is reported that more than 10,000 Chinese-made semiconductors were used in the completed facility.

This move is interpreted as an effort by Chinese AI companies to reduce their reliance on Nvidia chips and build their own AI ecosystems and computing infrastructure, amid ongoing U.S. restrictions on the export of advanced semiconductors to China.

In particular, as the Chinese AI startup Moonshot recently unveiled its new AI model, Kimi K3, which is evaluated as being close to the world's top-tier level, interest is growing regarding what computing infrastructure these companies are utilizing and whether Chinese-made AI semiconductors can support the development of advanced AI models in the long term.

Within China, Huawei is considered a leading company in the AI accelerator sector, and Cambricon is also engaged in a race to develop AI semiconductors to narrow the performance gap with Nvidia.

Large cloud and telecommunications companies with significant capital, such as Alibaba and China Telecom, are also leading the construction of computing infrastructure.

This is because building large-scale data centers requires not only the cost of purchasing massive amounts of high-performance servers but also enormous investments for stable power and water supply.

Zhipu raised billions of dollars earlier this year through a Hong Kong initial public offering (IPO) and subsequent stock sales.

Bloomberg previously reported, citing sources, that Zhipu had already achieved its annual revenue target early this month, and its annual recurring revenue (ARR) is expected to reach 1 billion dollars (approximately 1.4 trillion won) this year.

The Chinese government has also announced that it will invest approximately 2 trillion yuan (approximately 436.6 trillion won) over the next five years in nationwide data center construction to resolve computing bottlenecks in the AI industry.

Meanwhile, prior to this, some 20 Chinese companies related to AI chips, including SenseTime, Cambricon, Huawei, and Moore Threads, launched the Galaxy Project on July 18, a corporate cooperation project to build an independent Chinese AI ecosystem.

The core contents of this plan include the joint construction of one token operation center, the establishment of five intelligent computing clusters with a capacity of 10,000 Chinese-made AI graphics cards each, and support for 200 AI startups.

(Photo: Yonhap News)