Daily Management Review

The 'Battle Of A Hundred Models' Among China's AI Models Is About To Come To An End


The 'Battle Of A Hundred Models' Among China's AI Models Is About To Come To An End
A flurry of product announcements from startups and IT giants have been made nearly daily as a result of China's obsession with generative artificial intelligence, but investors are warning that a shakeout is about to occur as cost and profit demands increase.
A senior Tencent executive called this month's competition as a "war of a hundred models" as it involves rivals such as Baidu, Alibaba, and Huawei promoting their respective products. The excitement in China was first sparked by the success of OpenAI's ChatGPT almost a year ago.
According to brokerage CLSA, China already possesses at least 130 large language models (LLMs), accounting for 40% of the global total and trailing only the United States' 50% share.
Companies have also revealed a large number of "industry-specific LLMs" that connect to their main models.
However, according to investors and analysts, the majority had yet to develop workable business models, were overly similar to one another, and were currently coping with rising costs.
The industry has also been affected by the tensions between Beijing and Washington, as American dollar funds invest less in early-stage ventures and access issues to AI processors created by companies like Nvidia start to materialise.
"Only those with the strongest capabilities will survive," said Esme Pau, head of China internet and digital asset research at Macquarie Group, who expects consolidation and a price war as players compete for users.
She continued by saying that some major corporations have made it clear they intend to compete on price to acquire market share, much like cloud services like those run by Alibaba and Tencent have.
"In the next six-12 months, LLMs with lower capacities will gradually be eliminated due to chip restrictions, high costs and intensifying competition," Pau said.
Views on which companies will endure widely differ.
Only two to three general-purpose LLMs will ultimately dominate the market, according to Yuan Hongwei, chair of Shenzhen-based startup fund Z&Y fund.
That is why when choosing which firms to invest in, her company searches for experienced entrepreneurs.
Z&Y, which has previously backed companies like DJI, a maker of drones, and Pony.ai, a startup focused on autonomous driving, ultimately chose to support Baichuan Intelligence, a five-month-old business trying to develop an open-source AI model to compete with Meta Platform's Llama 2.
Wang Xiaochuan, the creator of China's No. 2 internet search engine Sogou Inc., founded Baichuan. In late August, Baichuan was one of the first five businesses to get Beijing's licence to launch a public chatbot.
According to Wang, the business is on track to close a second round that will value it at $1 billion.
"We see an opportunity here," Yuan said. "Wang himself is leading this project. Given his understanding of the digital business, his success with Sogou and how he commands attention industry-wide, we think it is our best bet."
New Chinese AI businesses are led by a number of other well-known businesspeople and IT leaders, like Yan Juejie, a former vice president of SenseTime, and Kai-Fu Lee, the former president of Google China.
Others argued that with their sizable user bases and broad range of services, China's leading digital firms Alibaba, Tencent, and Baidu ultimately had the biggest head start and substantial funds to succeed. For instance, they might easily provide generative AI capabilities to their cloud users as an additional plug-in.
"The incumbent tech giants have inherited an unfair advantage of a majority of low-hanging fruit business scenarios from their established ecosystems," said Tony Tung, managing director at Gobi Partners GBA.
Tung continued by saying that some investors now look back on their hasty investments in LLM firms made at the height of the frenzy earlier this year and that many of these startups are now trying to work with big giants to develop use cases or be potentially acquired by them.
"Many people end up developing similar LLMs, that are looking for similar problems to solve, with micro innovation in the data processing technique or model architecture," he said.
"At this particular moment, investors have certainly sobered up quite a bit compared to early this year."