Trump wants to impose a binary choice on AI. It is not certain that this will work

It is not possible to be part of a US-led coalition on artificial intelligence and, at the same time, part of a rival Chinese coalition, the US State Department wrote in a draft letter seen by Reuters in recent days. The coalition referred to in the letter is Pax Silica, the initiative launched last year by the United States aimed at safeguarding supply chains for artificial intelligence models and ensuring supply chain security, which has been joined by dozens of countries, including Japan, Australia, South Korea and Italy. Joining Pax Silica is not merely a matter of signing up to a pact, “but a commitment”, the draft states: this is how US President Donald Trump is reportedly preparing to force his allies to take sides. Washington is carefully avoiding any mention of competition with Beijing, just as Chinese leader Xi Jinping did during his first speech at the World Conference on Artificial Intelligence in Shanghai last July, but the reference to the technological challenge posed by China is clear: “It is not possible to combine such membership with duplicate initiatives whose expectations conflict with our own,” the letter states. For this reason, the 35 signatories of the US “Declaration on AI Opportunities” are being called upon to choose sides: either America or China.  

In recent months, the gap between American and Chinese companies in the field of artificial intelligence has narrowed, and the latest models unveiled by Beijing – such as Moonshot AI’s Kimi K3 and Alibaba’s Qwen 3.8 – have come ever closer to the cutting-edge American models from OpenAI and Anthropic, despite the restrictions imposed by the United States on China regarding advanced chips and the threat of further sanctions against Chinese start-ups by the US Treasury. In July, Xi Jinping responded to the US restrictions by launching a coalition of 29 countries (including Russia, Belarus, Serbia, Cuba, Pakistan, Brazil and Venezuela) for cooperation on AI, offering ‘open-source’ technology and training to countries in the Global South – those with which cooperation is ‘mutually beneficial’.

The risk is not only that more and more countries are adopting Chinese artificial intelligence systems, because they are cheaper and more accessible, but also that Chinese data is being used to train chatbots worldwide. According to the New York Times, the fact that the main chatbots are trained on English-language datasets poses a threat to China, which “to bridge this gap and develop more powerful AI tools, aspires to become a leading data provider”. Earlier this year, the country’s National Data Administration unveiled a plan to transform China into a powerhouse in the sector by the end of 2028, “to draw more users into the orbit of Chinese AI and narrow the gap with the United States in access to high-quality training data”. The largest of these datasets is called WanJuan, which in Chinese means “ten thousand scrolls”; as well as Chinese, it is available in Arabic, Korean, Russian, Thai and Vietnamese and is designed to be in line with “dominant Chinese values”. Yesterday, Lizzi Lee, a researcher at the Centre for China Analysis at the Asia Society Policy Institute, wrote in the Financial Times that “China is beginning to build its own self-sufficient artificial intelligence ecosystem capable of strengthening itself internally, and is increasingly offering it abroad”; for this reason, the United States, with its ultimatum regarding Pax Silica, hopes to curb its advance.

For the time being, the only country to have joined both the Chinese and American initiatives is Kazakhstan, but the Director-General of the Agency for Digital Italy, Mario Nobile, was also present at the Shanghai Conference. Following the summit, he stated: “I fully agree with the position expressed by President Xi Jinping: the development of AI must serve the common good.”