The domestic large model is about to be decisive

The investment community

2024-07-02 16:28Official account of Beijing Qingke Entrepreneurship Information Consulting Co., Ltd

The full text is 3268 words, and it takes about 10 minutes to read

Highlight the point

01Alibaba invested about US$800 million in fiscal 2024 to acquire about 36% of the shares of the Dark Side of the Moon, and the two domestic large-scale model companies of the Dark Side of the Moon and MiniMax received more than 1 billion yuan in financing.

02 China's AI large model pattern is beginning to appear, and the valuation of five companies, Zhipu AI, Dark Side of the Moon, MiniMax, Baichuan Intelligence, and Zero One Everything, is more than 1 billion US dollars.

03 Large factories are advancing in research and investment, and Tencent, Alibaba, Meituan, etc. have bet on the large model track to compete for talent and technological advantages.

04At the same time, colleges and universities around the country have strengthened the cultivation of AI talents, and Tsinghua University and Harbin Institute of Technology have opened AI courses.

05The outbreak of AI large models has triggered the younger generation of entrepreneurial talents, and local governments have issued policies to seize the AI large model.

Technical support is provided by Tencent Hybrid Model

A detailed investment was exposed.

Recently, in its 2024 fiscal annual report, Alibaba disclosed for the first time its investment in the dark side of the large-scale model startup. According to the financial report, Alibaba invested about US$800 million (5.9 billion yuan) in fiscal 2024 to buy about 36% of the company's stake in the dark side of the month.

In the first quarter of this year alone, two domestic large-scale model companies, Dark Side of the Moon and MiniMax, received more than 1 billion yuan in financing. And this is not only a VC game, large manufacturers are advancing research and investment at the same time, and hot money in the market is entering the market in a big way.

At the beginning of April, the Cyberspace Administration of China (CAC) issued an announcement on the filing of generative AI services: as of March, a total of 117 large models of various types in mainland China have passed the filing. Among them, Beijing occupies the lead, with a total of 51 large models passing the record, accounting for 44% of the country's quota. Shanghai and Guangdong followed closely behind, filing 24 and 19 respectively.

On the back of the buzz, some changes are also happening. According to recent reports from foreign media, a number of American generative AI companies are in a dilemma of selling themselves due to the rupture of the capital chain. The integration of hot money, technology change, and clearing are carried out simultaneously, and the competition of large models is accelerating and differentiating. As Gan Jianping of INCE Capital said: Maybe mankind is getting closer and closer to this singularity.

Domestic large model five tigers

According to incomplete statistics from the investment community, so far in 2023, there have been 106 financing events in the field of large AI models in China. Quietly, the pattern of China's AI large model has begun to appear. The five companies of Zhipu AI, Dark Side of the Moon, MiniMax, Baichuan Intelligence, and Zero One Everything are all valued at more than $1 billion, and are known as the domestic "Big Model Five Tigers".

At the end of May, it was reported that Zhipu AI was in the latest round of financing, and Saudi Aramco's Prosperity7 Fund invested $400 million in Zhipu AI, which made Zhipu AI valued at $3 billion.

This is a large-scale model startup with Tsinghua pedigree. The company was incubated from Tsinghua KEG (Knowledge Engineering Laboratory), CEO Zhang Peng graduated from the Department of Computer Science of Tsinghua University with a bachelor's degree, and a Ph.D. in the knowledge graph research direction of the Department of Computer Science of Tsinghua University.

Since the start of the All in large model in 2020, the research of Zhipu AI technology has involved all aspects of large models, from the original GLM pre-training framework, domestic computing power adaptation, and general base large models, to semantic inference, multimodal models, to long contexts, visual understanding, agent capabilities, and other aspects. At the beginning of June, Zhipu AI launched Qingyan Flow, a multi-agent collaboration system composed of hundreds of thousands of AI bodies.

As the earliest company established in the "Five Tigers", the capital behind it is also extremely luxurious. At the beginning of this year, a round of financing was officially announced, with the participation of the Beijing Artificial Intelligence Industry Investment Fund, which is the first AI large-scale model company invested by the Beijing Artificial Intelligence Industry Investment Fund since its establishment. In 2023, Zhipu AI will receive a total of more than 2.5 billion yuan in financing, and its investors include the Social Security Fund, Zhongguancun Independent Innovation Fund, Meituan, Alibaba, Tencent, Xiaomi, Kingsoft, Shunwei, Sequoia, Hillhouse, etc.

Another VC project that is fiercely contested is the dark side of the moon. In February this year, the dark side of the moon completed more than 1 billion US dollars in financing, setting a new record for the largest single round of financing for a domestic AI large model company at that time. In May, the latest round of financing outflow from the dark side of the moon, with a valuation of $3 billion, and new entrants include Tencent and Gaorong Venture Capital.

The leader of the company is Yang Zhilin, a post-90s student at Tsinghua University. Growing up in Shantou, Guangdong, Yang graduated from high school and was sent to Tsinghua University, where he later pursued a Ph.D. at Carnegie Mellon University. In March 2023, Yang Zhilin officially founded the dark side of the moon, and the other two co-founders are his Tsinghua undergraduate classmate and friend Zhou Xinyu, and Wu Yuxin, who also graduated from Tsinghua University and Carnegie Mellon University. With such a team background, the dark side of the moon has been sought after by many investment institutions since its establishment, including Sequoia China, Zhen Fund, Alibaba, Xiaohongshu, Meituan Dragon Ball, Lisi Capital, etc.

In March, MiniMax officially announced a $600 million funding round with a post-investment valuation of $2.5 billion. Founded in 2021, Yan Junjie, who is at the helm of MiniMax, graduated from the Institute of Automation of the Chinese Academy of Sciences, and served as vice president of SenseTime, where he was responsible for building the tool chain and underlying algorithms of deep learning, as well as the technical development of general intelligence. The co-founder, Yang Bin, is an alumnus of Yan Junjie at the Chinese Academy of Sciences. Previously, MiniMax has completed multiple rounds of financing, with investors including miHoYo, IDG Capital, Hillhouse Ventures, Yunqi Capital, Mingshi Capital, Tencent, etc.

Baichuan Intelligence, founded by Wang Xiaochuan, is rumored to have carried out a new round of financing of hundreds of millions of dollars in April this year, and investors from the participating confirmed that the company's valuation has exceeded 1.8 billion US dollars (about 13 billion yuan) since the beginning of this year.

The venture capital circle is no stranger to Wang Xiaochuan. In 1996, Wang Xiaochuan was recruited to the Department of Computer Science of Tsinghua University, and joined Sohu after graduating from graduate school. From Sohu to Sogou, Wang Xiaochuan has left a lot of memories in the Internet era. In April 2023, he officially announced his large-scale model entrepreneurial journey, and co-founded Baichuan Intelligence with Ru Liyun, former COO of Sogou and 1998 alumnus of Tsinghua University.

On the other hand, 01000000 Things, founded by Kai-Fu Lee, released the AI large model series "Yi" in November last year, including two versions of Yi-6B and 34B with 6 billion and 34 billion parameters; In March of this year, Zero One Unveiled a new vector database based on a fully navigated graph, called Descartes. In May, it launched the first free one-stop AI work platform "Wanzhi".

Kai-Fu Lee publicly disclosed that the revenue of a single product of To C this year reached 100 million yuan. In November last year, it was officially announced that Alibaba Cloud would lead the investment and financing, and the valuation at that time exceeded $1 billion.

The financing of the large model track is hot, and the Chinese version of OpenAI makes people reverie.

The Dachang Legion is coming

According to the pedigree, the domestic large model can be divided into three categories: science and technology manufacturers, start-up companies, and university research institutions. With obvious advantages in computing power, platforms, models, applications, etc., large manufacturers have not reduced their voice in large models.

In April 2023, Zhang Yiming issued an open letter saying, "ByteDance can't miss AGI. Like any Chinese Internet company, AGI cannot be missed. AGI can solve an organization's second-curve growth dilemma. ”

In September of the same year, Wu Yongming, the new CEO of Alibaba Group, issued a letter to all employees, announcing the establishment of two strategic priorities: user-first and AI-driven.

In April last year, Alibaba Cloud's "Tongyi" large model series debuted, and since then, Tongyi Qianwen, Tongyi Tingwu and Tongyi Wanxiang have come out.

In August last year, Douyin launched the "Skylark" model, and then launched a number of AI products and services based on the Skylark model, including Feishu My AI, conversational large model Doubao, and various AI video tools for Jianying.

In September last year, Tencent released the "Hybrid Yuan" model. Combined with the Tencent Cloud MaaS solution of ToB released by Tencent two months ago, a complete set of large-scale product and service systems has been built.

In October last year, Huawei released the Pangu Model 3.0, a series of industry-oriented large models. Focusing on ToB, its basic model is combined with industry data for mixed training, and the industry model is used in mining, meteorology, medicine, molecular, electric power, finance and other industries.

In October last year, Baidu released the "Wenxin" large model 4.0, with parameters of more than one trillion yuan, which was one of the largest models in China at that time.

In addition, large manufacturers have also bet on the large model track.

The "King of Sweepers" is Ali. From September last year to May this year, according to incomplete statistics from the investment community, Ali has successively launched 6 large-scale model companies: Zhipu AI, Baichuan Intelligence, Zero One Everything, Dark Side of the Moon, Minimax, and Precision Chemistry. According to Alibaba's financial report, the company invested a total of about US$800 million (about 5.9 billion yuan) in the dark side of the moon in fiscal year 2024 and purchased about 36% of the shares. In addition, Ant Group has also invested in two large-scale model companies, Shengshu Technology and Zhipu AI.

Tencent is not to be outdone. In 2023, it will take the lead in Shenyan Technology, Lightyear Beyond (which has been acquired), Minimax, Zhipu AI, Suiyuan Technology, and Wuwen Core Dome, and the latest round of financing on the dark side of the moon this year.

Meituan Dragon Ball was deeply involved in the Series A financing of the large-scale model star company "Dark Side of the Moon" and was the lead investor in its Series A1 financing. In addition, light years away, Meituan also appeared behind Zhipu AI. It is worth mentioning that as an idealized carrier of large models, humanoid robots ushered in new opportunities, and Meituan took the lead in launching two humanoid robot companies, Galaxy Robotics and Unitree Technology.

Unconsciously, China's head AI unicorns are behind the figures of large manufacturers.

Scramble for the big model

AI is a highly talent-intensive discipline, and the siphon effect of the stronger is obvious, and universities around the world are strengthening the cultivation of AI talents.

Just this month, Harbin Institute of Technology announced that it would officially establish the School of Artificial Intelligence, and at the same time set up the "AI+ Advanced Technology Leading Class", which will recruit outstanding undergraduates from all over the country from this year, and is committed to cultivating outstanding talents in the field of AI with a broad international vision, a deep sense of social responsibility, and leading the future development.

Tsinghua University also established the School of Artificial Intelligence in April this year, with Yao Qizhi as the first dean. Focusing on the two key directions of "Artificial Intelligence Core Basic Theory and Architecture" and "Artificial Intelligence + X", the School will build China's independent "AI Top Talent and Original Innovation Base" with high positioning and new mechanism, so as to provide strong support for the realization of high-level scientific and technological self-reliance and self-reliance.

Not only that, Fudan University announced that Fudan will launch at least 100 "AI courses" in the new academic year (20242025 academic year) from September this year.

and the School of Artificial Intelligence of Shanghai Jiao Tong University was also unveiled not long ago; Beijing University of Aeronautics and Astronautics also announced that this year's high recruitment will add 4 pilot classes for the cultivation of outstanding talents, including artificial intelligence majors, and implement the eight-year integrated training of this doctorate.

The explosion of AI models has made entrepreneurial talents younger and younger. High IQ and high education make it easier for this group of young people to embrace the rapid wave of change.

In addition, various localities have also issued policies to seize the AI model.

In May last year, Shanghai issued the "Several Measures for Promoting the Innovation and Development of Artificial Intelligence Large Models in Shanghai (2023-2025)", creating the country's first large model innovation ecological community "model speed space", providing all-round service guarantee in computing power scheduling, open data, evaluation services, financial services, comprehensive services, etc.

At the same time, the Beijing Municipal People's Government issued the "Implementation Plan for Beijing to Accelerate the Construction of an Artificial Intelligence Innovation Source with Global Influence (2023-2025)". It points out a number of cutting-edge directions, and achieves new guidance in technological innovation.

And Guangdong. According to the "Implementation Rules for the Construction of Artificial Intelligence Large Model Application Demonstration Zone in Haizhu District, Guangzhou", the cumulative annual reward amount for a single large model enterprise is up to 10 million yuan. Shenzhen issued the "Shenzhen Action Plan for Accelerating the High-quality Development and High-level Application of Artificial Intelligence (20232024)". Focus on the fields of general large models, intelligent computing power chips, intelligent sensors, intelligent robots, intelligent networked vehicles, etc., and focus on supporting the creation of open-source general large models based on domestic and foreign chips and algorithms.

In June this year, Shandong issued the "Guiding Opinions on Accelerating the High-quality Development of the Large Model Industry" to accelerate the breakthrough of a number of forward-looking, original and leading key technologies of large models, create a basic large model with a certain international influence, and cultivate a number of industry-level large models and scenario-level large models with wide coverage and high product energy efficiency in key areas and key links.

This is a "battle for the future".

View original image 98K

•