Chinese AI offensive: Tech elites drive startup mania

 Chinese AI offensive: Tech elites drive startup mania

Traditionally, artificial intelligence (AI) ventures in China have been initiated by academics or professionals employed by foreign companies. However, the recent attention drawn to OpenAI's ChatGPT has inspired a new wave of AI entrepreneurs among China's tech giants, consisting of the elite members of the industry.

 

Individuals such as Jia Yangqing, who previously served as a vice-president at Alibaba and have spent years testing and refining AI technology at Chinese companies, are now venturing out to establish their own companies.

 

Prior to his role at Alibaba, Jia Yangqing had gained significant experience in the field of artificial intelligence. He had previously worked on computer vision and deep learning at Google Brain, the artificial intelligence and machine-learning research unit of the US-based tech giant. Jia also headed a research team at Facebook, now known as Meta Platforms, where he oversaw the development of a large-scale AI platform for the company's applications.

 

In March, Jia Yangqing departed from Alibaba to seek out the "next challenge" in his career, with reports indicating that he would join a new start-up that is focused on AI infrastructure.

 

Several other prominent figures have also left their positions in established Chinese tech companies to pursue their own AI ventures. Among them are Wang Changhu, who formerly served as the head of vision technology at ByteDance, and Li Yan, who led the multimedia understanding unit in the short video division of Kuaishou Technology.

 

A significant number of high-ranking officials in Chinese tech companies were originally recruited from universities or overseas corporations. Having gained extensive experience in researching and identifying practical applications for AI models, it is a logical decision for them to explore the possibility of establishing their own AI startups, according to a Chinese hardware investor.

 

The success of OpenAI has served as a significant source of inspiration for these individuals, and the influx of capital into the AI industry has certainly played a role in giving them the confidence to embark on their own entrepreneurial ventures, according to the investor.

 

The recent surge of interest in AI startups has also motivated tech specialists and investors to actively seek out promising ideas to support and invest in.

 

Wang Huiwen, retired co-founder of Meituan, recently announced a personal investment of $50 million to establish Beijing Lightyear Technology, with the intention of building China's equivalent of OpenAI. The company subsequently acquired OneFlow Technology, which specializes in developing open-source AI frameworks for machine learning, through a stock swap. The merger of the two companies has enabled them to secure ongoing financing at a valuation of $1 billion.

 

Wang Huiwen's foray into the AI industry with Beijing Lightyear Technology has served as a source of inspiration for others, including Lee Kai-Fu, the chairman and CEO of Sinovation Ventures. Lee has since launched his own AI startup, likely motivated by the success and potential of the AI industry.

 

The launch of Project Al 2.0, a business with the goal of creating platform-based AI and related applications, was announced by Lee Kai-Fu in March. Lee, a seasoned AI veteran and former head of Google China, is actively looking for investment possibilities in the sector, emphasizing the industry's enormous potential.

 

Due to the secretive nature of many new AI projects in China, it is challenging to track the exact number of such startups. However, industry insiders anticipate that more than 50 new ventures focused on building large-scale AI models could be launched within the next three months. This serves as further evidence of the rapid growth and immense potential of the AI industry in China.

 

Developing large-scale AI models is a costly endeavor, and it presents a significant challenge that only a select few are capable of tackling successfully.

 

According to Yin Qi, the co-founder and CEO of AI software developer Megvii Technology, building a GPT large language model would require a significant investment in hardware, including at least 10,000 Nvidia A100 graphics processing chips - which are widely regarded as one of the best chips for powering machine learning tasks. The cost of such a project would amount to around two billion yuan (S$383.8 million) in hardware investment alone. This highlights the immense challenge and significant financial commitment required to develop large-scale AI models.

 

The high cost of developing large-scale AI models often prices many AI companies out of the race, especially since most tech startups typically operate at a loss for several years. This further underscores the challenges and barriers that exist within the AI industry, even for those with considerable financial backing.

 

Many AI companies fear that if they don't invest in large-scale AI models like ChatGPT, they may become vulnerable to new entrants in the market, particularly if they have recently gone public and are under pressure to please their shareholders. As a result, there is a sense of urgency within the industry to keep up with the latest technological developments and invest in cutting-edge AI research, despite the significant financial costs involved.

 

SenseTime, a leading AI software developer, decided to take on the challenge of developing large-scale AI models like ChatGPT. In April, the company unveiled a suite of new services, including its large AI model SenseNova and a chatbot called SenseChat, which boasts around 180 billion parameters. By investing in cutting-edge AI research and technology, SenseTime is positioning itself to stay ahead of the competition and meet the evolving needs of its clients in industries such as healthcare, finance, and retail.

Machine learning algorithms rely heavily on parameters to evaluate the performance of an AI model. The company aims to establish SenseNova as a one-stop-shop for clients and partners seeking access to a diverse range of high-quality AI models. It envisions SenseNova as a comprehensive "supermarket of AI big models."

 

SenseTime is a prominent developer within China's AI sector, but it lags behind its major tech competitors in terms of financial performance. In 2022, the company generated 3.8 billion yuan in revenue but incurred a loss of over 6 billion yuan. By contrast, Baidu recorded 123.7 billion yuan in revenue and a net income of 20.7 billion yuan.

 

In 2018, Baidu announced the launch of its Kunlun AI chips, which it has since deployed extensively. In contrast, SenseTime only introduced its first hardware product, the AI-ISP chip, in the previous year.

 

In addition, SenseTime is currently facing US sanctions that restrict its access to capital and important American components. Despite this, the company remains undaunted, with co-founder Yang Fan stating that expertise in hardware engineering, algorithms, and big data are key factors in building large models.

 

According to an October 2022 report by IDC, China's AI investment is expected to reach $26.7 billion by 2026, accounting for roughly 8.9% of global investment. The report also projects that hardware will be the largest primary market in China's AI industry over the next five years, with investments surpassing $15 billion in 2026.

 

Zhou Ming, a former Microsoft Asia Research employee who resigned in 2020 to establish his own AI startup, anticipates significant opportunities for these models in the finance sector.

 

Considering the resource constraints typically faced by startups, Zhou recommends that they prioritize building models based on market demand. He has identified the finance industry as a particularly promising market due to its abundance of usable data, potential clients, and financial resources. Additionally, there are few companies that can provide ChatGPT-like, multi-round dialogue capabilities in this sector.

 

Zhou's own startup, Beijing Langboat Technology, offers various functions such as public sentiment analysis, research report analysis, and professional financial translations based on its self-developed Mengzi large language model.

 

Meanwhile, other firms seeking to overcome the barrier to entry are focusing on the consumer sector. Zhou Zhifeng, a partner at Chinese venture capital firm Qiming Venture Partners, notes that although the initial wave of AI hype lacked supporting infrastructure, there has been a proliferation of large model APIs.

 

API, which stands for application programming interface, is a software intermediary that enables communication between two applications.

 

Zhou highlights that startups do not need to hire a large team of software engineers to develop their own models. Instead, with modest funding, they can assemble a relatively small team to create applications based on existing APIs.

 

ZMO.AI, a Chinese startup specializing in AI image generation, has successfully followed this approach, with a team of approximately 20 employees. Founder and CEO Zhang Shiying reports that her company launched a marketing content generation product in September 2022 and attracted over one million monthly active users within just five months. The firm introduced a paid product in February, and it is already generating profits

 

According to Baidu co-founder and CEO Robin Li, startups should not concentrate on building models but rather focus on developing applications. He believes that this approach will create "brand new entrepreneurial opportunities that are 10 times greater than WeChat and Douyin".