China appears to be losing ground in the field of generative AI.
China has long been considered, along with the United States, a pioneer in the field of artificial intelligence (AI). But today, the country seems to be losing ground in the field of generative AI. Last summer, Chinese officials boasted that China had greater capabilities than the United States in generative AI, with around 80 different large language models (LLMs).
However, China has few, if any, of the four essential elements of LLM training. When it comes to generative AI, the Chinese ecosystem appears to be weak in all key areas.
Weaknesses in data diversity
Data sets are essential for generative AI. Having sufficient and quality data is irreplaceable so that the results provide added value. But as China has largely isolated itself from Western systems, the integrity of the data is questionable. The “hallucination” of an AI making up results is already a problem with Western systems, but it could be even more common with Chinese products.
Late software
The quality of artificial intelligence is closely linked to the software that accesses public clouds. It would allow “industrial” and optimized use of the long series of data necessary for LLM training. The providers of next-generation software solutions for public clouds are American companies and do not offer their services in China. China certainly has its own products such as Oushu* DB, which are comparable to American services, among others SnowFlake* and Databricks*. Technologically, these programs would be several decades behind schedule.
AI chips are firmly in American hands
Graphics cards, called GPUs, are extremely important for the computing power needed for generative AI. In fact, Nvidia and its GPUs, but also all the companies active in this field that are potential suppliers of AI chips, such as Broadcom*, Marvell*, AMD* or Intel*, are all American. As a result, China’s semiconductor industry cannot meet its own needs for AI chips. The US government has banned European and American equipment manufacturers from supplying key technologies to China, impeding the growth of the sector in the Middle Kingdom.
An obstacle: regulations
In the United States and China, public clouds are managed by the respective Internet giants. Chinese players have just gone through several years of sometimes very harsh regulations. In the case of Alibaba* and its cloud subsidiary Alicloud, this amounted to a takedown. Furthermore, the situation is aggravated by the fact that the cloud subsidiaries of Internet companies are not sufficiently optimized, since most client companies continue to operate with their own servers. In the future, regulation could also hinder China’s development in the field of generative AI. In fact, China, like the rest of the world, will sooner or later have to regulate its generative AI. Depending on the form this regulation takes in China, the Chinese development model could weaken again.
*The individual securities mentioned do not constitute an investment recommendation.
