The craze for generative artificial intelligence in China has led to frequent product announcements from both startups and tech giants. However, investors are cautioning that a shakeout is likely as costs and the need for profits increase.
The initial excitement in China, sparked by the triumph of OpenAI’s ChatGPT nearly a year ago, has led to what a high-ranking Tencent executive recently referred to as a “battle of a hundred models.” As a result, companies like Baidu, Alibaba, and Huawei are all promoting their respective products.
According to brokerage firm CLSA, China currently possesses a minimum of 130 significant language models, making up 40% of the world’s total and closely trailing the United States’ share of 50%. Furthermore, numerous companies have introduced numerous “industry-specific LLMs” that are connected to their main model.
Investors and analysts believe that many have not yet found successful business models, are too alike, and are currently dealing with rising expenses.
The relationship between Beijing and Washington has impacted the industry, causing U.S. dollar investments in early-stage projects to decrease and creating challenges in acquiring AI chips from companies like Nvidia.
Esme Pau, the leader of China’s internet and digital asset research at Macquarie Group, predicts that only the most capable individuals will thrive. She anticipates a competition among players for users, resulting in consolidation and a battle over prices.
She mentioned that many top companies have indicated their intention to engage in price competition in order to increase their market share, similar to the strategies employed by cloud services like Alibaba and Tencent.
“According to Pau, within the next six to 12 months, LLMs of smaller capacities will be phased out due to limitations in chips, increased expenses, and growing competition.”
FOUNDERS AND INCUMBENTS
There are conflicting views on the longevity of different companies.
According to Yuan Hongwei, who leads the venture capital firm Z&Y Capital in Shenzhen, she predicts that only a small number of versatile LLMs will become the dominant players in the market.
This is the reason why her company seeks out seasoned entrepreneurs when selecting which startups to fund.
Z&Y, whose past investments include drone maker DJI and autonomous driving startup Pony.ai, eventually decided to back Baichuan Intelligence, a five-month-old firm looking to build an open-source AI model to rival Meta Platform’s Llama 2.
Baichuan was started by Wang Xiaochuan, founder of China’s No.2 internet search engine Sogou Inc, and became one of the first five companies to receive Beijing’s approval to release a public chatbot in late August. The company is on track to close a second round that will value it at $1 billion, Wang said.
“We have identified an opportunity,” stated Yuan. “Wang is spearheading this project and with his expertise in the digital industry, proven track record with Sogou, and influential presence, we believe this is our most promising option.”
Prominent entrepreneurs and technology executives, including Kai-Fu Lee, former head of Google China, and Yan Juejie, former vice-president of SenseTime, are backing a number of emerging Chinese AI startups.
Some people claimed that the top tech companies in China, including Alibaba, Tencent, and Baidu, had the greatest advantage and financial resources to thrive due to their extensive user base and diverse range of services. For example, they could seamlessly provide generative AI services as an extra feature for their cloud customers.
According to Tony Tung, managing director at Gobi Partners GBA, the current tech giants have an unfair advantage due to their established ecosystems and the abundance of easily attainable business opportunities.
According to Tung, certain investors have expressed remorse for investing too early in LLM companies during the height of the excitement earlier this year. Many of these startups are currently facing challenges in creating solid business plans and are seeking partnerships with larger technology companies to find ways to apply their services or potentially be acquired by them.
He stated that numerous individuals often create comparable LLMs in search of similar issues to address, utilizing minor innovations in data processing techniques or model architecture.
“At present, investors have significantly become more cautious compared to earlier this year.”