Science & Tech · AI · 8 hrs ago
China promotes open AI models as chip limits constrain their use
At a World AI Conference in Shanghai, Xi Jinping promoted Chinese AI as a global public good.
Chinese companies including DeepSeek, Qwen and Moonshot release model weights that users can download, customise and run on their own servers.
This differs from services by U.S. companies such as OpenAI, Anthropic and Google, which host their models and provide access through online services.
Downloadable models let users keep data on local servers and avoid relying on a provider to maintain access.
Beijing presents this openness as a way to share AI’s benefits and cast China as a cooperative alternative to the United States.
But U.S.-led chip export restrictions constrain China’s computing capacity, limiting access to advanced chips.
Training a model takes a large chip fleet for a limited period, while answering users’ queries requires continuous computing power that grows with demand.
The article says this gap allows China to build capable models but makes serving them at mass scale much harder; it gives no specific next step.
China is promoting open-weight AI models as an internationally accessible alternative while U.S. export controls restrict its access to advanced chips.
At the ninth World AI Conference in Shanghai, President Xi Jinping described Chinese AI models as a global public good.
Chinese models from labs including DeepSeek, Qwen and Moonshot release their weights for users to download, customise and run locally.
The article says chip limits are a constraint on China’s ability to train and serve AI models, even though they have not prevented it from reaching model-level parity.
Beijing’s promotion of open models also aims to present China as a cooperative AI power, in contrast with U.S. proprietary models.
- Who
- China, including President Xi Jinping and Chinese AI labs such as DeepSeek, Qwen and Moonshot.
- What
- China is promoting open-weight AI models as a global public good. The article says U.S. chip export restrictions constrain China’s ability to serve models at scale.
- When
- At the ninth World AI Conference, held in Shanghai in October 2026. The article says U.S. export restrictions have been imposed since 2022.
- Where
- Shanghai, China; the models can also be run locally by users in other countries.
- Why
- China aims to present itself as a responsible, cooperative AI power. Chip export restrictions also limit its access to advanced computing capacity.
China’s open-model approach
U.S. proprietary-model approach
Access
China’s open-model approach
Chinese labs release model weights for download, allowing users to customise and run models locally.
U.S. proprietary-model approach
U.S. companies including Anthropic, OpenAI and Google host their frontier models on their servers and offer access through APIs.
International role
China’s open-model approach
Xi presented AI as something that should be shared through international cooperation.
U.S. proprietary-model approach
Xi criticised what he described as the U.S. using national security concerns to take an exclusionary approach to AI.
The U.S. administration and its allies imposed export restrictions limiting China’s ability to import and manufacture high-end chips.
OpenAI crossed a million GPUs, while Meta targeted the equivalent of 1.3 million by the end of the year.
Xi addressed the World AI Conference in Shanghai for the first time in nine years and promoted Chinese AI models as a global public good.
- Conference
- Ninth World AI Conference, Shanghai
- Chinese AI labs
- DeepSeek, Qwen and Moonshot
- Export restrictions
- Imposed since 2022 by the U.S. administration and its allies
- DeepSeek V3 training chips
- About 2,000 export-compliant H800s
- Xi’s conference address
- First time in nine years
1 source
What people are saying
humanity’s collective wisdom
historic opportunity
bridge the AI and digital divides
should not be a solo performance by a single country but a symphony of international cooperation