Generative AI in China: a revolution in the making?

Thanks to our group of Gen-AI ChangeMakers in Shanghai, I had the pleasure of organizing and hosting a panel discussion on the current state of Generative AI (GenAI) technology in China. We had an exceptional conversation with high-level representatives from SenseTime, Tencent, Canva, and LVMH.

My objective was to address the “elephant in the room”: Why do we not see more from the Chinese large players on Generative AI? What is the overall development of GenAI/AGI technology in China? What is the gap between the US and China?

Here are the key takeaways from our conversation with this exceptional panel:

1. China is not lagging behind in terms of knowledge and GPUs. Research is striving and of high level. Although there are some US restrictions on semiconductors, fast GPUs are available even if in limited quantity. Large Chinese internet companies have enough GPUs stocks for LLM in the current stage, however, many GPUs are “borrowed” from other business (e.g. Cloud, Ads, etc.) and the supply is very tight. On the” data to train front” iff few argue that high-quality Chinese texts is missing to train generative AI models, it is not really an issue as models can be trained in any languages, then fine tuned in Chinese. Maybe the real challenge is thatlarge Chinese players are listed companies that need to deliver results; therefore, their generative models tend to address verticals where ROI is faster to come by. The public may not see fully or even have access to what large players are using internally.
2. Large international players, like Canva with access to foreign GenAI models, are still evaluating how to implement similar access to these services in China. Chinese authorities are yet to clarify which foreign services will be available. In the meantime, other international players that have services depending on OpenAI are avoiding small companies proposing to bypass the Chinese firewall and are hoping that either Baidu, Tencent, ByteDance, Alibaba will release their own AGI services or that Microsoft Azure solutions powered by OpenAI will be soon available on the Chinese Market, though probably with limited functionalities.
3. LVMH clarified the view of established luxury brands on GenAI/AGI technology and its potential business use cases in China. Aside from OpenAI, other local solutions could help consumer goods players in two areas: FMCG could use GenAI on the product creation to accelerate the ideation process for new products. Jobs of Product designers will be to select from the ideas that can be produced by Gen-Ai. By contrast, luxury houses such as LVMH are probably going to be more focused on finding local Clienteling solutions powered by Gen-Ai that could help drive a more personalized dialogue with customers with tailored content.
4. The GenAI/AGI Chinese ecosystem and investment environment is aware that China is probably one or two years behind, but is very confident in the speed of Chinese startups and tech companies to catch up. There are numbers of open-source models that Chinese startups can leverage as foundations, as well as new large language models from Chinese very large players that have been launched in the last few days and weeks. For Chinese startups, it is challenging to compete on large models because those large models represent years of existing research, as well as a very large amount of capital and data. Also for start-ups and other tech companies, the GPUs will be a real blocking point because they cannot “borrow” from other divisions like the large players. Right now, the opportunity is rather for startups that have a very strong industry know-how, and with a strong understanding of building applications on the top of these existing models. Chinese companies are particularly innovative in sectors where they have access to the data they need for these models, so they tend to create GenAI solutions on industry verticals: healthcare, finance, retail to name a few. Ultimately, AGI models could become a threat to these segmented solutions, but if these verticals GenAI models can hold on their proprietary data and have strong customer relations then AGI pan solutions will face considerable challenges. Also, solutions that can integrate not only the formula creation but also the production environment may have a stronger hold in China, which is notoriously strong in the industrialization sector.

Speakers also expressed their personal views on AGI topics that are keeping them up at night. Right now, probably the Chinese authorities are in an aftershock mode, as the models are coming up too fast. Content moderation to keep up with Chinese regulations is, therefore, on everyone’s minds on an everyday basis. But on the longer view, some speakers were concerned that if only two AGIs models emerge, one coming from China and the other one from the US, it could reshape the landscape of the geopolitical situation and further divide our current environment.

Overall, it was a fascinating and thought-provoking conversation that shed light on the current state and future potential of GenAI and AGI in China. The Chinese GenAI/AGI ecosystem is aware of the gap with the US, but is confident in the speed of Chinese startups to catch up. The large players in China tend to focus on verticals where the ROI is faster to come by and may not make their generative models available to the public. For startups, the opportunity lies in building applications on top of existing models and having a strong understanding of the industry they operate in. Ultimately, the panelists believe that GenAI and AGI have immense potential for innovation and disruption in China, yet shared a worry of a future geopolitical bipolar AGI world environment.