Chinese tech elites rush into their own AI ventures, but mania could cause the next tech bubble
China’s seasoned tech entrepreneurs are eagerly entering the domain of artificial intelligence (AI) startups, fueled by the ChatGPT frenzy, in order to establish themselves as influential players in China’s quest to challenge the United States’ dominance in generative AI.
Wang Huiwei, who formally departed Meituan in 2020, was one of the earliest Chinese tech veterans to enter the field with enthusiasm. In February, he made headlines by announcing the launch of Guangnian Zhiwai, an AI startup that aims to create China’s OpenAI. Notably, the Chinese startup has already been valued at $1 billion after acquiring OneFlow Technology Ltd., an open-source AI framework developer for machine learning, via a stock exchange.
Inspired by Wang’s entrepreneurial endeavor, Kai-Fu Lee, the chairman and CEO of Sinovation Ventures, embarked on his own startup venture.
Mid-March, 61-year-old Lee announced in a private WeChat post that he was actively preparing for the launch of Project AI 2.0. The primary focus of his startup will be the development of AI-based productivity tools with the objective of enhancing efficiency and effectiveness.
The former head of Google China and seasoned AI expert is looking for investment opportunities in the sector.
Vice President of Alibaba Group Holdings, Jia Yangqing, has departed the company to launch his own artificial intelligence business.
Jia, who oversaw Alibaba’s cloud computing services division’s computing platform division, reportedly to co-found a new startup with a focus on AI infrastructure.
Jia previously managed a research team at Facebook, which is now Meta Platforms Inc., and worked on computer vision and deep learning at Google Brain, the artificial intelligence and machine-learning research division of the American tech behemoth.
Several notable individuals have recently left their positions in prominent companies. Among them are Wang Changhu, previously in charge of vision technology at ByteDance Ltd., and Li Yan, who held a leadership role in the multimedia understanding unit of Kuaishou Technology’s short video division.
Since many of the key players at Chinese tech businesses were really hired underhandedly from universities or foreign organizations and had already spent a lot of time researching and identifying applications for AI models, it was not a surprise for them to try their hand at an AI startup.
The artificial intelligence startup gold rush has resulted in the emergence of a full-blown mania. Investors are fiercely competing to invest in these companies, granting some AI entrepreneurs valuations in the nine-figure range with little more than an idea and a resume.
“ChatGPT is generating a lot of excitement. ChatGPT has suddenly made it possible for us to communicate with computers, in contrast to the metaverse, which has a difficult time finding practical applications,” according to Ding Daoshi, an independent internet analyst and former director of Beijing-based consultancy Sootoo.
Most recently, Chinese startups MiniMax is almost done financing more than $250 million, which will value it at approximately $1.2 billion. MiniMax is working on AI solutions similar to Microsoft-backed OpenAI’s ChatGPT.
MiniMax’s latest fundraising attracted new investors such as an entity linked to Tencent.
Yan Junjie, a former vice president of the Chinese AI company, and other former SenseTime employees founded MiniMax in 2021.
When MiniMax first started off, miHoYo, the company behind the popular game “Genshin Impact,” provided funds.
Even though many AI startups have raised significant quantities of capital in the past, they cannot afford to ignore recent investor approaches. This is in part due to the high cost of computational capacity, which is required for artificial intelligence technologies such as ChatGPT.
Morgan Stanley analysts contrasted ChatGPT’s cost-per-query to Google’s cost-per-search in a research note. They calculated that OpenAI’s expenditures are around seven times greater than Google’s, at an estimated average of $0.02 per inquiry, compared to $0.003 for Google.
The analysts point out that the difference is due to “the compute intensity of ChatGPT’s natural language models.”
For a company to build a GPT large language model, according to Yin Qi, co-founder and CEO of AI software developer Megvii Technology Ltd., it would require at least 10,000 Nvidia A100 graphics processing chips, which are regarded as one of the best chips for powering machine learning tasks, and an investment of about 2 billion yuan ($289 million) in hardware.
Some investors questioned whether more seasoned competitors with bigger funds would crush AI start-ups. Some claimed that big players with deeper pockets had an unfair advantage given the steep computing cost
Last week, Baidu, Inc announced a 1-billion-yuan ($140 million) fund for financing Chinese startups that explore generative AI.
In deployments of up to 10 million yuan each, Baidu wants to use the pool to nurture companies based on its Ernie AI model.
Ernie will be used by potential startups to expand their services, and Baidu and its VC partners will review their business plans.
SenseTime Group Inc., a provider of AI software, introduced a number of new services in April, including their sizable AI model SenseNova and a chatbot named SenseChat that includes almost 180 billion parameters. Machine learning algorithms depend heavily on parameters, which are also used to gauge how well an AI model is doing. SenseTime hopes SenseNova could become a “supermarket of AI big models” for clients and partners.
Generative AI startups are already achieving extremely high valuation even though there is little evidence of comparable income, let alone profit. This is similar to the low-interest-rate boom times, when companies with unproven business ideas obtained valuations worth billions of dollars despite having little to show.
The technology behind generative AI is truly intriguing, and some of the hype surrounding it may be legitimate. But before assigning large valuations to generative AI startups, investors would be wise to investigate the facts in order to avoid another bubble.
Instead of creating models, startups should concentrate on creating applications, where “brand new entrepreneurial opportunities that are 10 times greater than WeChat and Douyin will emerge,” said Robin Li, co-founder and CEO of Baidu.
According to a report by state-run research institutes, Chinese organizations created 79 large-language models (LLMs) in the country over the previous three years as they stepped up their attempts to develop artificial intelligence (AI) algorithms.