The Mysterious Rise of China’s Desert AI Hubs

Bloomberg OriginalsAbout 3 min readAug 1, 2025Watch original
THE SUMMARYAI-generated

Key Concepts:

  • AI Supremacy
  • Data Centers
  • US Chip Ban
  • Nvidia H100/H200 Chips
  • Semiconductor Manufacturing
  • DeepSeek
  • US-China Tech Race
  • Smuggling Networks

China's AI Ambitions and Data Center Construction in Xinjiang

China aims to become a global leader in AI by 2030, heavily investing in high-tech industries. A Bloomberg News investigation reveals the construction of massive data centers in Xinjiang, a sensitive region where China has been accused of human rights abuses. These data centers aim to install over 115,000 high-end, US-designed Nvidia chips, specifically H100 and H200, which are subject to a US export ban.

The Role of Banned US Chips

The Nvidia H100 and H200 chips are considered the industry standard for training large language models. The US ban aims to prevent China from accessing these advanced semiconductors, restricting their potential military applications and hindering their AI sector. Project approval documents indicate that local governments in Xinjiang and Qinghai have approved 39 data centers intending to use these banned Nvidia processors.

Challenges in Verifying Chip Access

Bloomberg News has been unable to verify whether the data centers actually possess the claimed Nvidia chips. Initial invitations to inspect server rooms were revoked, making it difficult to confirm the presence of the banned hardware. Nvidia claims there is no evidence of smuggling and argues that their technical support is necessary to keep the chips running, which they are not providing to China.

US Efforts to Restrict Chip Exports

The US government has implemented export restrictions on advanced semiconductors to China since 2022, leveraging the fact that most advanced chips are designed in the US and controlled by US law. While the White House has loosened some restrictions, allowing the export of the inferior H20 chip, the goal is to keep China a generation behind in chip design. The US aims to maintain its technological edge and prevent China from acquiring advanced chips with potential military applications.

China's Semiconductor Manufacturing Capabilities

China has created a $48 billion semiconductor investment fund to boost its chip manufacturing capabilities. However, despite the significant investment, China still lags behind other manufacturers in terms of precision and technology. The process of manufacturing advanced chips is described as incredibly complex, requiring specialized lasers and intricate mirror systems. Huawei's Ascend series, considered China's best chip, is still at least a generation behind.

DeepSeek and China's AI Advancements

Chinese AI startup DeepSeek has challenged perceptions of US tech dominance by developing a competitive large language model. DeepSeek claimed to have used lower-efficiency chips that were legally available to China. However, Washington suspects that DeepSeek may have had access to H100 chips. DeepSeek has also expressed interest in the data center facilities in Yiwu.

Global Implications and the US-China Tech Race

China aims to lead the world in AI, targeting the global south and emerging markets to rely on Chinese AI technology and infrastructure. However, the US is also investing heavily in AI, with Nvidia alone promising half a trillion dollars in US chip manufacturing. The Stargate project, backed by Trump and involving OpenAI, plans to build data centers in the US with 400,000 of Nvidia's most advanced chips, surpassing the scale of the Xinjiang data centers.

Conclusion

The construction of data centers in Xinjiang highlights China's determination to advance its AI capabilities despite US export restrictions. While the presence of banned Nvidia chips remains unverified, the scale of the construction and the potential for China to overcome technological hurdles pose a significant challenge to US efforts to maintain its technological edge. The US-China tech race continues, with both countries investing heavily in AI and semiconductor manufacturing.

AI summaries can miss context or contain errors. Check important details against the original video.

MAKE IT YOURS

Read. Remember. Reuse.

Free tools

Go a little deeper.

Have a question about this video? Load its transcript to open the video chat.