Z.AI constructed a large AI data center utilizing chips manufactured in China.

      For a year now, one pressing question has persisted regarding China's rapidly advancing AI models: what are they actually powered by? Z.AI has provided part of the answer by constructing a large data center that exclusively utilizes Chinese-made chips.

      The company, previously known as Zhipu, has begun partial operations at this facility, according to Bloomberg, which cited an insider. This 1-gigawatt center is designed to train Z.AI’s GLM models. Z.AI did not respond to inquiries for comments.

      What has been constructed

      The scale is noteworthy. A gigawatt represents approximately the power consumption of 750,000 homes at any given time, positioning this site among the largest server hubs established by a Chinese AI lab.

      Equally impressive is the number of chips involved. Z.AI reportedly operates several computing clusters, each containing over 10,000 chips, none of which are from Nvidia, as noted by the source. Bloomberg did not specify the exact type of chips. Huawei, China's leading AI accelerator designer, competes with local companies like Cambricon and Alibaba in the race to close the gap with Nvidia.

      Why this is significant

      US export controls have prevented Chinese labs from acquiring Nvidia's most potent chips, raising the question of whether domestically produced components could handle the demands of training advanced models as opposed to merely operating them.

      A 1GW cluster entirely composed of domestic silicon offers a definitive answer to that question, indicating that Chinese labs can continue to expand even when cut off from the industry-standard chips. The timing of this development is particularly noteworthy, coming just days after Moonshot's Kimi K3, based in Beijing, matched leading US models but then faced computing limitations and paused new registrations. Z.AI is competing against the same rivals and is banking on owning its hardware.

      The larger development

      Z.AI is not undertaking this initiative alone. China plans to invest around 2 trillion yuan, approximately $295 billion, over five years on data centers nationwide. So far, cloud giants Alibaba and China Telecom have been the primary builders.

      Additionally, Z.AI has the financial resources to compete effectively. Following a listing in Hong Kong and a subsequent share sale, Z.AI is on track to achieve $1 billion in annual recurring revenue, potentially becoming the first Chinese AI company to reach that milestone. Its stock surged nearly 20% on that day, as it positions itself as an enterprise AI provider, drawing comparisons to Anthropic.

      The caveats

      However, a few reservations should be noted. The information regarding the chips comes from an unnamed source, and Z.AI has not verified these details. Constructing the cluster does not equate to demonstrating that it can train a top-tier model as efficiently as one powered by Nvidia.

      Domestic chips may still fall short in terms of raw performance, and successfully integrating thousands of them into a cohesive system poses challenges. The true test will be the next GLM model and how it stands up against those produced by US labs.

      Nonetheless, the trajectory is clear. The initiative to create a Chinese alternative to Nvidia has progressed from mere concepts to a functioning, gigawatt-scale data center. This shift alone alters the conversation about the effectiveness of export controls.