China’s widespread release of open-weight artificial intelligence models is strengthening its global soft power, according to Andrew Ng, an influential AI technologist and former co-founder of Google Brain. While American companies largely prioritize closed-source, proprietary models that are expensive and less adaptable for international users, Chinese firms like DeepSeek and startup Z.ai have gained significant influence by providing free access to high-performing open-weight models. Ng noted that many nations in Africa are now adopting these Chinese alternatives, which allows developers to run sophisticated AI locally on their own hardware rather than relying on US-based API services that can be restricted by export controls.

This shift in AI infrastructure is viewed as a strategic challenge for the United States. Ng emphasized that when nations rely on Chinese-developed models, the AI responses often reflect China's cultural and political values, such as regarding the 1989 Tiananmen Square events, rather than Western perspectives on liberty. Furthermore, Ng argued that the US government's occasional revocation of foreign access to specific models—such as the restriction on the Anthropic Fable 5 model—has unintentionally accelerated the urgency for global nations to secure their own AI supply chains through open-weight technologies that cannot be remotely withdrawn.

The competitive landscape is becoming increasingly complex as different cultural zones embed their own values into LLM (large language model) algorithms. Kai-Fu Lee, a noted computer scientist, observed that just as China censors content critical of the Communist Party, Western models often reflect cultural sensitivities regarding race and gender, while models in the Islamic world may restrict content perceived as blasphemy. Consequently, the adoption of open-weight models is driven by this inherent diversity and the ability for nations to align AI tools with local circumstances.

Despite the strategic advantages of open models, experts warn of significant dual-use risks including misuse by terrorists or fraudsters. Former Google CEO Eric Schmidt expressed concern over the proliferation of open-source technology, advocating for the mandatory implementation of robust safety guardrails through RLHF (training technique where human feedback shapes the model's behavior). Schmidt suggested that the US and China should establish a 'no-surprise rule,' requiring both superpowers to provide transparency regarding the development of new frontier AI systems, mirroring Cold War-era efforts to manage the proliferation of nuclear capabilities through mutual understanding and communication.

China’s widespread release of open-weight artificial intelligence models is strengthening its global soft power, according to Andrew Ng, an influential AI technologist and former co-founder of Google Brain. While American companies largely prioritize closed-source, proprietary models that are expensive and less adaptable for international users, Chinese firms like DeepSeek and startup Z.ai have gained significant influence by providing free access to high-performing open-weight models. Ng noted that many nations in Africa are now adopting these Chinese alternatives, which allows developers to run sophisticated AI locally on their own hardware rather than relying on US-based API services that can be restricted by export controls.

This shift in AI infrastructure is viewed as a strategic challenge for the United States. Ng emphasized that when nations rely on Chinese-developed models, the AI responses often reflect China's cultural and political values, such as regarding the 1989 Tiananmen Square events, rather than Western perspectives on liberty. Furthermore, Ng argued that the US government's occasional revocation of foreign access to specific models—such as the restriction on the Anthropic Fable 5 model—has unintentionally accelerated the urgency for global nations to secure their own AI supply chains through open-weight technologies that cannot be remotely withdrawn.

The competitive landscape is becoming increasingly complex as different cultural zones embed their own values into LLM (large language model) algorithms. Kai-Fu Lee, a noted computer scientist, observed that just as China censors content critical of the Communist Party, Western models often reflect cultural sensitivities regarding race and gender, while models in the Islamic world may restrict content perceived as blasphemy. Consequently, the adoption of open-weight models is driven by this inherent diversity and the ability for nations to align AI tools with local circumstances.

Despite the strategic advantages of open models, experts warn of significant dual-use risks including misuse by terrorists or fraudsters. Former Google CEO Eric Schmidt expressed concern over the proliferation of open-source technology, advocating for the mandatory implementation of robust safety guardrails through RLHF (training technique where human feedback shapes the model's behavior). Schmidt suggested that the US and China should establish a 'no-surprise rule,' requiring both superpowers to provide transparency regarding the development of new frontier AI systems, mirroring Cold War-era efforts to manage the proliferation of nuclear capabilities through mutual understanding and communication.