By
Stuart Kerr, Technology Correspondent,
LiveAIWire
Building the Super-Compute Backbone
China’s ambition in compute infrastructure is being supercharged. The latest example is China Unicom’s new data centre in Xining, Qinghai: it’s built using almost entirely domestically manufactured AI chips, including units from Alibaba’s chip division T-Head, and is on track for a compute capacity of 20,000 petaflops when fully operational. So far, at 3,579 petaflops built using nearly 23,000 domestically made AI chips, it already signals a push toward technological self-sufficiency that is central to Beijing’s AI strategy, according to Reuters.
This isn’t happening in isolation. As documented in the China’s AI Infrastructure Surge Report from Strider Technologies, over 250 AI-capable data centres have either been built or announced across China, spanning remote western provinces and densely populated eastern regions alike. These facilities are part of the national plan to balance energy, geography, and China’s AI datacentre demand.
Eastern Data, Western Computing: Strategy and Strain
China’s “Eastern Data, Western Computing” initiative is foundational to its compute strategy. Under this approach, data centres are built in western areas where energy is cheaper and land is more available, intended to serve demand from the east, where many of China’s tech hubs are located.
But the plan has exposed tension points. Many of these new western data centres suffer from high latency to coastal demand centres, mismatched hardware, and under-utilisation hovering between 20 and 30 percent, according to Reuters reporting on the surplus capacity crackdown.
To address this, Beijing is working with its major state telecoms, China Mobile, China Telecom, China Unicom, and the Ministry of Industry and Information Technology to build a national cloud platform. This network would unify, orchestrate, and schedule compute power across facilities, enabling surplus capacity to be sold or redirected. The goal is full interconnection of public compute resources by 2028.
Chips, Scale, and Technological Sovereignty
Central to China’s AI datacentre push is reducing dependence on foreign semiconductor supply chains. The Qinghai data centre is powered mostly by home-grown chips, with Alibaba’s T-Head alone supplying roughly 72 percent of the hardware. This move is part of a larger effort to respond to US export restrictions and enhance domestic chip capabilities.
Simultaneously, the country is accelerating regulation: more stringent licensing, cancelling or freezing smaller, financially risky projects, and forbidding local governments from involvement in certain compute infrastructure to avoid overbuilding.
Implications: Beyond Borders and Downloads
China’s grand plan has ripple effects. It’s not only about powering domestic AI models; it’s reshaping geopolitical assumptions about compute capacity. A recent World Economic Forum report on China’s path toward AI-powered industrial transformation highlights how infrastructure upgrades are central to economic, security, and diplomatic leverage.
From a global perspective, this may force Western cloud providers and AI labs to rethink supply chains, latency costs, and the ethics of compute geography. If compute becomes a national infrastructure asset, questions of access, standardisation, and governance come into sharper focus.
Challenges on the Road to ‘Stargate’
Yet the path is not without obstacles. Integrating thousands of data centres with different chip architectures and performance profiles into one unified cloud is technically complex. Latency, hardware standardisation, grid stability, energy supply, and environmental impacts loom large. The western sites, while cheap in energy, still struggle to achieve latency expectations for eastern consumers.
Moreover, economic viability remains uncertain for many under-utilised centres. Low utilisation rates strain budgets, especially when maintenance, cooling, and staffing costs are high. Local governments are now being more cautious, cancelling projects or imposing stricter criteria.
Moving Forward: What to Watch
What matters now is whether China can successfully operationalise its compute ambition for its AI datacentre network. Key indicators will be: how well the national cloud platform performs, whether usage climbs, whether hardware becomes more standardised, and if exports or foreign usage agreements emerge.
From LiveAIWire’s earlier reporting on efficient, smaller AI models, efficiency and proximity of compute can be as influential as raw scale. Internal demand from industries like biotech, climate modelling, and defence will test whether these super-datacentres serve real needs or are symbols of prestige.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity, and the social impact of emerging technology. He publishes daily at LiveAIWire.com.