> ## Content Index
> Fetch the complete content index at: https://nordisk-ai.ghost.io/llms.txt
> Use this file to discover other available public pages before exploring further.

# China Trains AI Across Multiple Data Centers and GPU Architectures
- URL: https://nordisk-ai.ghost.io/china-trains-ai-across-multiple-data-centers-and-gpu-architectures/
- Published: 2024-09-30T20:04:12.000Z
- Updated: 2026-08-27T17:21:22.000Z
- Description: China has achieved an AI breakthrough by training a generative AI model across multiple data centers and GPU architectures, despite US sanctions limiting access to powerful chips.
- Author: Anders Hagen
- Tags: Society and Politics, AI, AI Development, Data Centers, Generative AI, Huawei, China, Nvidia, sanctions, Sweden, Sweden, USA, #Migrated-1787841634512, #wp, #wp-post, #Import 2026-08-27 14:41, #Import 2026-09-07 13:41

China has taken a major step forward in artificial intelligence by developing a generative AI (GAI) model trained across multiple data centers and GPU architectures, according to industry expert Patrick Moorhead. This is a significant breakthrough, especially given the technical complexity of using different GPU systems within a single data center, let alone across geographically dispersed data centers. Moorhead, chief analyst at Moor Insights & Strategy, shared the news on X (formerly Twitter) after learning of the development during a meeting.

### The Technology Behind Training Across Multiple GPU Architectures

The ability to train GAI models across multiple locations and architectures is crucial for China's continued AI development, especially since US sanctions have restricted its access to the most powerful chips. Nvidia has created a scaled-down version, the H20, that complies with the restrictions set by Washington, but even these chips risk being banned, creating further uncertainty for Chinese tech companies.

### US Sanctions Are Driving Innovation

To counter the chip shortage, Chinese researchers have developed methods for combining GPUs from different manufacturers within a single training cluster. This includes mixing high-performance chips like the Nvidia A100 with less powerful but more accessible alternatives like Huawei's Ascend 910B or Nvidia's H20\. Although this technique has historically caused efficiency losses, China appears to have solved the problem, as shown by its success in training a GAI model across multiple data centers.

### Combining High-Performance and Simpler GPUs

The combination of high-performance and simpler GPUs, provided by both Huawei and Nvidia, lets Chinese researchers maximize their resources despite the sanctions. Similar attempts in the past have led to major efficiency losses, but researchers now appear to have overcome this obstacle. It's an important milestone that allows China to keep developing advanced AI technology without access to the latest hardware from the West.

### Solving the Efficiency Loss in Mixed GPU Use

Efficiently training AI across different GPUs and locations has been a challenge, since each type of chip has different capabilities and features. Traditionally, this has led to significant inefficiency, but by solving this problem, China has shown it's possible to continue its AI development despite harsh external restrictions. This success signals a new phase in how AI technology can be trained and developed, even under difficult conditions.

### China's Determination to Continue Its AI Development

This breakthrough demonstrates China's determination to continue its AI development despite political and technical obstacles. As Huawei has put it: China will find ways to push its AI agenda forward, despite the restrictions imposed on it. Necessity drives innovation, and this latest advance is a clear example of that.