On
a 760-acre island on the Yangtze River in Wuhu, rice paddies are being
replaced by server farms. The Wuhu data island, operated by four of China’s
largest technology operators including Huawei, China Telecom, China Unicom,
and China Mobile, is the flagship project of what insiders are calling
“the Stargate of China”: a coordinated national effort to
consolidate and expand the country’s AI computing capacity in direct response
to the United States’ $500 billion Stargate infrastructure programme. With
total investment across Wuhu’s facilities reaching 270 billion renminbi,
equivalent to approximately 37 billion dollars, the scale signals that this
is not a regional project but a national strategic priority.

The context for this investment is a computing gap that is
significant and growing. According to research by Epoch AI, the United States
holds approximately 75 per cent of global AI computing capacity. China holds
approximately 15 per cent. That asymmetry has become a defining feature of
the AI competition between the two countries, and Beijing’s response is to
address it through infrastructure rather than through chip production alone,
given that US export restrictions continue to limit China’s access to the
most advanced semiconductors from Nvidia and other
manufacturers.

China Unicom’s Qinghai Data Centre and the Domestic Chip
Strategy

One of the clearest signals of China’s infrastructure ambition is
a data centre built by China Unicom in Xining, Qinghai province, which Reuters
reported in September 2025 as a deliberate demonstration of
domestic chip capability. Approximately 72 per cent of the facility’s
hardware uses chips from Alibaba’s semiconductor division T-Head, reducing
dependence on foreign suppliers. The centre is targeting a compute capacity
of 20,000 petaflops when fully operational, having reached 3,579 petaflops at
the point of the Reuters report. The project is explicitly framed by Beijing
as proof that China can build competitive AI infrastructure from domestic
components, a message designed both for domestic industry and for
international audiences watching whether US export controls are
working.

The infrastructure surge is documented at scale in Strider Technologies’
China AI Infrastructure Surge Report, which identified over 250 AI-capable
data centres either built or announced across China. These facilities span a
deliberate geographic strategy: energy-abundant remote western provinces for
large-model training, and densely populated eastern regions near commercial
hubs for inference workloads that serve end users with lower acceptable
latency.

Eastern Data, Western Computing: The Strategy Behind the
Grid

China’s “Eastern Data, Western Computing” initiative,
commonly abbreviated EDWC, is the organising framework for this geographic
distribution. Under EDWC, computation capacity is built in western regions
where land and electricity are cheap, then piped eastward to serve commercial
demand in Shanghai, Beijing, Guangzhou, and China’s other major tech hubs.
The Reuters
reporting on China’s surplus computing power strategy documented
the practical challenges this creates: western data centres frequently suffer
utilisation rates of between 20 and 30 per cent, because the demand they were
built to serve has not yet materialised or the latency to eastern markets is
too high for real-time applications.

To address this, Beijing is developing a national cloud platform,
coordinated through the Ministry of Industry and Information Technology
alongside China Mobile, China Telecom, and China Unicom. The goal is a
unified system that can schedule and redirect compute across facilities,
allowing surplus capacity in any region to be deployed where demand exists.
Full interconnection of public compute resources is targeted for 2028. The
ambition is analogous to a national electricity grid: distributed generation,
centralised coordination, delivered on demand.

What This Means for the Global AI Race

The implications of China’s data centre build-out extend beyond
the bilateral US-China competition. As we explored in our coverage of China’s
515 million gen-AI users, the scale of domestic AI adoption creates
a demand base that itself justifies infrastructure investment. Models need to
be served to users, and serving at scale requires compute that can handle
inference loads in real time. The infrastructure being built now is not just
for training frontier models but for deploying AI services to a population
whose uptake of generative tools is among the fastest in the world.

The energy implications are also significant, a dimension explored
in our analysis of the
AI emissions paradox and the carbon cost of computation. China’s
EDWC initiative routes much of its planned compute to western provinces where
renewable energy, particularly hydropower and wind, is abundant. The
environmental calculus of Chinese AI infrastructure is therefore different
from the US data centre model that depends heavily on fossil-fuel-intensive
grids in the American south and Midwest.

The Obstacles Are Structural

The Chinese plan faces genuine structural challenges. Integrating
facilities that use different chip architectures, some from Nvidia via grey
market channels, some from domestic suppliers like Huawei’s Ascend or
Alibaba’s T-Head, creates standardisation problems that are technically
difficult to resolve. The latency issue for western data centres is not
simply a network infrastructure problem but a physics constraint: the
distance from Qinghai to Shanghai creates irreducible delays for interactive
applications. And domestic chips, while improving, still trail Nvidia’s
latest hardware on performance per watt for most AI
workloads.

Local governments, having invested heavily in data centres that
are now underutilised, face financial stress that has prompted Beijing to
intervene with stricter licensing and the cancellation of smaller,
financially marginal projects. The national platform model requires a degree
of coordination between state telecoms and private operators that has
historically been difficult to sustain in Chinese infrastructure programmes.
The governance complexity is substantial even before the technical challenges
are considered.

The Bigger Picture

China’s data centre push is best understood not as a direct
replica of the US Stargate programme but as a different approach to the same
underlying problem: how to build the computational foundation for a national
AI strategy when access to the best hardware is constrained and domestic
alternatives are not yet fully mature. The approach is characteristically
pragmatic, consolidating what exists, subsidising domestic production, and
planning infrastructure that routes around the specific bottlenecks created
by export restrictions. Whether the result can close the three-quarter gap in
global computing capacity that Epoch AI’s research identifies is a question
that will shape the AI landscape for the next decade. What is already clear
is that the
regulatory and strategic frameworks governing AI now extend from
software and models all the way down to the physical infrastructure layer,
and that layer is where the most consequential competition is currently being
fought.

For Western technology companies and policymakers, China’s
infrastructure push raises questions that go beyond the bilateral computing
gap. If China successfully builds a national compute network that can
redirect capacity to wherever demand is highest, the result is an
infrastructure advantage that accumulates over time regardless of chip-level
comparisons. The ability to coordinate compute at national scale, directing
resources toward the most commercially valuable or strategically important AI
workloads, is itself a form of competitive advantage that semiconductor
export controls do not directly address. The geopolitical implications of compute
geography are only beginning to be understood, but the Wuhu data island and
the EDWC initiative together suggest that China is thinking about them
carefully.

About the Author

Stuart Kerr is the Technology Correspondent for LiveAIWire. He
writes about artificial intelligence, emerging technology, and the forces
reshaping work, business, and society.