Nvidia’s powerful AI chips are at the center of a quiet construction boom unfolding on the edge of the Gobi Desert in Xinjiang.
In the county of Yiwu, China is racing to build massive data centers: high-tech hubs that reflect its growing ambition to become a global leader in artificial intelligence, even as US export bans on those chips, imposed in 2022, remain firmly in place over national security concerns.
A Bloomberg analysis shows that Chinese tech firms are planning to roll out more than 115,000 Nvidia AI chips.
At the heart of this hardware push are Nvidia’s H100 and H200 models, the same chips that drive cutting-edge AI systems like OpenAI’s GPT-4o and Google’s Gemini.
Eager to close the gap with Western rivals, Chinese companies such as DeepSeek are racing to tap into this computing power.
Still, getting hold of these chips is anything but straightforward.
The US first imposed export restrictions in 2022 and has since tightened the rules, barring Nvidia and its partners from selling their most advanced AI hardware to Chinese buyers.
To close potential loopholes, the US Commerce Department has cracked down on indirect routes such as transshipments through Malaysia and Thailand that Chinese firms have reportedly used, often via shell companies, to sidestep the bans.
One high-profile case in 2024 saw Singaporean authorities charge three people with helping funnel $390 million worth of Nvidia chips to China through Malaysia.
Nvidia Chips in short supply as China’s AI data center boom
Copy link to section
Yet despite the hurdles, data center construction is charging ahead in Xinjiang and neighbouring Qinghai.
In late 2024, local governments greenlit 39 new projects many of which openly referenced plans to deploy Nvidia chips, even though those remain subject to US export controls.
By mid-2025, at least seven sites in Xinjiang had either broken ground or signed contracts to provide AI computing services.
One operator even claims it’s running DeepSeek’s latest model on state-of-the-art hardware, though the specifics remain murky.
Still, serious doubts remain about whether China can secure these chips at the scale it needs.
US officials estimate that just 25,000 of the restricted Nvidia processors are currently inside the country, a fraction of what the Xinjiang data center projects alone would require.
There’s also scant evidence of a black market capable of delivering over 100,000 top-tier chips to a single region.
Some analysts suggest that many of these initiatives may be more about signalling alignment with Beijing’s AI goals than about actual computing power on the ground.
‘Made in China 2025’
Copy link to section
China’s swift buildout of data centers in the remote deserts of Xinjiang is part of a much larger ambition: turning the country into a global AI superpower.
This push is rooted in strategic plans like “Made in China 2025” and the 2017 New Generation AI Development Plan, both of which emphasize self-reliance, cutting-edge innovation, and a reduced dependence on foreign technology.
Backed by heavy state investment, sprawling infrastructure projects, and efforts to weave AI into everything from industry to national defense, the campaign is moving fast.
But US export curbs on advanced semiconductors have forced Beijing to rethink its sourcing and double down on domestic chip production deepening the fault lines in the global tech rivalry.