China Touts AI Clout, But Its Openness Won’t Last
| AUTHOR | Pranay Kotasthane |
| DATE | July 23, 2026 |
| CATEGORIES | High Tech Geopolitics Siliconpolitik AI Geopolitics |
I spoke to Surya Gangadharan on The Gist at StratNews Global about what China’s moves at the 2026 World AI Conference in Shanghai mean and about India’s Semiconductor Mission 2.0.
The Shanghai summit has been running since 2018. At the first edition, Elon Musk was a speaker. By 2026, American presence had thinned out and Xi Jinping addressed the conference for the first time. More significant than Xi’s speech was the operationalisation of WAICO — the World AI Cooperation Organisation — agreed to by 29 countries. I think of it as the Belt and Road Initiative for AI, based out of China, created by China, and populated by countries already in China’s strategic ecosystem.
Also read Matt Sheehan’s excellent annotation of Xi’s WAIC 2026 speech.
What gives Xi’s statements weight is that Chinese models are doing quite well. They come at a fraction of the price of American proprietary models and are not far behind the frontier. So China is positioning itself as an open, reliable AI provider, directly attacking the US strategy of export controls, AI diffusion rules, and chip restrictions. Xi even used the word “open-source” (even the Indian PM said the same at the Paris AI summit).
But China is anything but open. In the last few month alone, Beijing brought out three decrees making it very difficult for investment to go out of China, with vague national security restrictions. If a Chinese company wants to establish a rare earth investment in India or anywhere else, they face layers of licences and checks, and most likely will not be granted approval.Right now, openness suits China’s catch-up strategy. But as AI becomes critical public infrastructure, I would not be surprised if China restricts frontier model access — partly because open-weight models are actually a narrative threat to the CPC. Once you download an open-weight Chinese model and run it outside Chinese servers, it will answer questions about Tiananmen Square that the domestically deployed version blocks.
My falsifiable prediction is CPC will impose export controls on its frontier open-weight models before January 1, 2028. This could take the form of licensing requirements, delayed releases, parameter-count thresholds, or end-use restrictions.
On India’s positioning: on semiconductors and critical minerals, India is clearly on the Western side — every top-25 fabless firm by revenue has a design centre in India, and India is part of both the Mineral Security Partnership and the Pax Silica framework. It is on the AI model layer that India takes a multialignment approach, and rightly so. Open-weight models can be dissociated from the risks of the China stack. You do not need to talk to a Chinese player — you download the weights and run inference locally if you have the compute. So the strategy is do not be dependent on China’s AI chips, but remain open to using any open-weight model.
Check out this post by my colleague Bharath answering the question: Where Censorship Lives On Chinese Open Weight AI Models?
On India Semiconductor Mission 2.0, the cabinet has finally put a number to the next five years. The pillars that stand out are design — our comparative advantage, where 20 per cent of the world’s chip designers already are in India — and talent development. Process knowledge ultimately lives in people; that is what makes TSMC irreplaceable. Indians being across the world in every segment of the semiconductor supply chain is what will eventually allow India to replicate that capability domestically. The Dholera fab will come online towards end-2028 and will need another year or two to improve yield. This is a decadal journey we have just begun.
Also see: Bharat Shakti’s article based on the same conversation.