Wire flash
TechZ.ai Activates 1-Gigawatt AI Data Center Using Only Chinese Chips
Editorial responsibility
- No named human review is recorded for this page.
- Source reporting is collected, normalized, translated or condensed automatically when needed.
- Automatically published source-backed update
Chinese AI developer Z.ai (formerly Zhipu) has completed construction of a 1-gigawatt data center equipped exclusively with domestically produced chips and partially activated it, according to a Bloomberg report. The facility will train the company's GLM model family. Z.ai now operates several computing clusters with over 10,000 chips each. While the chip supplier was not named, evidence points to Huawei's Ascend accelerators, as Z.ai's GLM-5.2 model was trained entirely on Huawei hardware. The company has been on the U.S. entity list since January 2025, cutting off access to Nvidia silicon. However, Chinese chips trail Nvidia on performance per watt, meaning a gigawatt of domestic silicon delivers less usable compute. The development comes as Beijing plans to spend roughly $295 billion over five years on a nationwide AI data center grid, with at least 80% domestic technology. Supply constraints persist due to SMIC's high utilization and scarce domestic HBM memory.
Source report
Beijing, China — Chinese AI developer Z.ai (formerly Zhipu) has completed construction of a 1-gigawatt (GW) data center stocked exclusively with domestically manufactured chips and has partially activated the facility, Bloomberg reported Monday, citing a person familiar with the matter.
The facility will be used to train the company's GLM model family. According to the source, Z.ai has now built or operates several computing clusters, each containing more than 10,000 chips. For context, one gigawatt of electricity is sufficient to power approximately 750,000 homes, placing this site among the largest ever established by a Chinese AI lab.
Chip Supplier and U.S. Restrictions
The source did not name the chip supplier, but Z.ai's recent training history points to Huawei. In June, the company released GLM-5.2, an open-weight model reportedly trained entirely on Huawei Ascend accelerators with no Nvidia hardware involved. The model topped open-weight leaderboards within a week of release.
Z.ai has been on the U.S. Commerce Department's entity list since January 2025, which legally prohibits access to Nvidia silicon, leaving domestic components as its only viable supply chain.
Performance Considerations
Despite the scale, raw power draw alone does not equate to U.S. facilities of similar size. Chinese accelerators, such as Huawei's Ascend line, trail Nvidia's current Blackwell parts in performance per watt. As a result, a gigawatt of domestic silicon delivers less usable training compute than a gigawatt consumed by Nvidia-based systems.
Broader Industry Context
- National AI Infrastructure Plan: Beijing is drafting a plan to spend approximately 2 trillion yuan ($295 billion) over five years on a nationwide grid of AI data centers, with at least 80% of underlying technology sourced from Chinese suppliers.
- Manufacturing Constraints: SMIC's most advanced stable node — the roughly 7nm-class N+2 process — is running above 93% utilization, making it difficult to fill new facilities.
- HBM and Chip Supply: Scarce domestic high-bandwidth memory (HBM) limits how many Ascend-class accelerators Huawei can assemble. Huawei shipped approximately 812,000 AI chips last year.
- Capacity vs. Production: China can build a 1GW shell much faster than it can produce the chips needed to draw 1GW of power.
Competitor and Financial Updates
- Moonshot, a rival AI firm, suspended new subscriptions on Sunday, stating in a social media post that it wanted to prioritize compute for existing members following the launch of its Kimi K3 model.
- Z.ai is on track to achieve $1 billion in annual recurring revenue, having hit its 2026 sales target in July. The company recently raised billions of dollars through a Hong Kong IPO and a follow-on share sale.
Unconfirmed Details
The source did not disclose the data center's location, cost, or construction timeline.
Stay on the cutting edge: Get the Tom's Hardware Newsletter — best news and in-depth reviews, straight to your inbox. Follow Tom's Hardware on Google News or add us as a preferred source for the latest news, analysis, and reviews.
Source
Latest from Tom's HardwareNeutral / independent
Part of this Story
Z.ai Activates 1-Gigawatt AI Data Center Using Only Chinese Chips