China Unicom Unveils Massive Data Hub with Chinese AI Chips
China Unicom has confirmed the opening of a new data centre in Xining, Qinghai province.
The announcement, made via state broadcaster CCTV and reported by Reuters, describes the facility as “massive” and backed by around US$390m in investment.
Positioned as a flagship project, it highlights China’s drive to advance AI and high-performance computing through the use of domestically engineered technologies.
Built to operate on local processors rather than foreign-made chips, the facility reflects Beijing’s strategy to ease dependence on imported hardware.
The facility is currently equipped with 23,000 AI chips produced by Chinese manufacturers, delivering 3,579 petaflops of computing capacity.
Once fully operational, it is projected to reach 20,000 petaflops.
A petaflop is a measure of processing speed, equivalent to one quadrillion floating-point operations per second.
Domestic chips at the core
Most of the processors at the new centre are supplied by Alibaba’s semiconductor arm, T-Head, which accounts for roughly 72% of the total.
Additional suppliers include MetaX, Biren Tech and Zhonghao Xinying, with future upgrades set to incorporate chips from Tecorigin, Moore Threads and Enflame, broadening the ecosystem of domestic providers.
China Unicom, Alibaba, Biren, MetaX, Enflame, Tecorigin and Zhonghao Xinying all declined to comment to Reuters, while the official announcement omitted details about the facility’s power capacity or overall floor space.
It also remains unclear whether the reported compute output is measured in FP64 – the double-precision floating-point format standard in supercomputing benchmarks like the Top500 – or in FP8, which is widely adopted for AI training and inference workloads.
Local policy pushes domestic adoption
China is pushing operators of public data centres to prioritise locally developed hardware.
In August, Beijing introduced a mandate requiring publicly owned facilities to source more than half of their chips from domestic suppliers, a move designed to bolster the national semiconductor ecosystem and mitigate risks tied to export restrictions.
The policy shift comes as China’s internet regulator reportedly banned the country’s largest technology companies from purchasing Nvidia’s AI processors.
Widely used in AI systems worldwide, Nvidia chips have been at the centre of escalating trade tensions between Washington and Beijing.
Against this backdrop, the new data centre represents more than a facility launch: it signals a broader strategic direction, closely aligning infrastructure growth with the expansion of the domestic semiconductor industry.
Alibaba steps up chip development
Alongside its role in supplying the Xining facility, Alibaba is also pushing forward with its own semiconductor programme.
Earlier this month, the company disclosed that it is developing a new AI inferencing chip, already manufactured domestically and now undergoing testing.
The initiative aligns closely with government efforts to enhance China’s capabilities in AI-focused chip design and fabrication.
Although deployment timelines remain unspecified, the launch of new processors underscores Alibaba’s growing importance in delivering domestic compute capacity for data centres.
As the lead chip supplier for the China Unicom project, its T-Head unit highlights the strategic interplay between cloud providers, telecom operators and the evolving domestic semiconductor landscape.
Building for AI and HPC demand
China Unicom’s latest data centre underscores the scale of investment being channelled into AI and high-performance computing in China.
Designed to reach up to 20,000 petaflops of computing power, the facility supports national efforts to broaden infrastructure capacity while cutting dependence on imported chips.
By integrating processors from a range of domestic developers, the project not only advances policy objectives but also serves as a testing ground for China’s rapidly evolving semiconductor ecosystem.
