The two financings exceeded the total amount of financing in China's large-scale model field last year.Text丨Wang and TongEditor丨Cheng Manqi
Just over one month into 2024, the capital competition in China's large-scale model field has entered a new stage.
LatePost learned that Moonshot AI, a large-scale model startup, recently completed a Series B financing of more than $1 billion, with investors including Alibaba, Meituan, Xiaohongshu and Sequoia China. At the end of this round, the Dark Side of the Moon is valued at over $2.3 billion.
This is currently the largest single round of funding in China's large model space, but this record will not necessarily last long.
It is understood that another large model head company is also carrying out new financing amounting to $1 billion, and the transaction has entered an advanced stage.
The two US$1 billion financings together have exceeded the total amount of public financing in China's large-scale model sector last year (about 10 billion yuan).
Large-scale model startups are competing for the little oxygen left: the overall downturn in US dollar investment, the RMB** is more focused on certainty, and the willingness and strength to invest heavily in large models are mainly large Internet or technology companies.
As a limited investor, China's major Internet companies are also engaged in fierce competition, and have differentiated into different actions and strategies: first, a large amount of foreign investment to serve strategic goals; The second is not to make a move for the time being, and concentrate on making your own large model.
The dark side of the moon, the new shareholders of this round, Ali and Meituan, both belong to the first faction.
Alibaba has previously invested in large-scale model startups Zhipu AI, Baichuan Intelligence, and Zero One Everything; Meituan, for its part, invested in Zhipu AI and invested in and acquired Lightyear Away.
Other companies that cast a wide net include Tencent and Xiaomi: Tencent invested in Baichuan Intelligence, Zhipu AI and MiniMax last year; Shunwei, founded by Lei Jun, has invested in Baichuan Intelligence, Zhipu AI and MiniMax, of which Baichuan and Zhipu AI have also been invested by Xiaomi.
And Byte has not yet invested in large model companies.
Alibaba is one of the most aggressive Chinese Internet companies investing in large models.
An investor said that in Baichuan and Zhipu AI, which Alibaba and Tencent invested in at the same time, Ali's stake is higher. This time, Alibaba participated in a new financing of the dark side of the moon with an amount of more than $1 billion.
The difference in action comes from different strategic intent and firmness.
In 2023, Alibaba has formed a relatively clear large-scale model strategy: the biggest goal is not to make money from self-developed models, but to provide cloud computing capabilities for many large models and seize the opportunity to become AI computing infrastructure.
A survey by Silicon Valley venture capital firm A16z found that 80%-90% of the early-stage financing of large models and generative AI startups is used to purchase computing power from cloud computing platforms; A16Z estimates that model and application companies will also send 10%-20% of their annual revenue to cloud computing companies when the industry is more mature. Equivalent to an "AI tax".
One of the ways to collect AI taxes is to cooperate deeply with potential large-scale model companies, that is, AI computing power buyers: for example, Microsoft pledged to invest $10 billion in OpenAI early last year, most of which is settled in cloud computing resources rather than cash; Google and Amazon also pledged a total of $6 billion in Anthropic's investment last year, and Anthropic's will spend more than $3 billion over four years on Google's cloud computing services, as well as Amazon's cloud computing services.
Investing heavily in large models will help Alibaba Cloud continue to strengthen its position as the No. 1 market in China and capture the incremental growth brought by AI computing.
Compared with Alibaba, Tencent's investment is weaker, which may be due to two reasons: first, the importance of cloud computing business to Tencent is not as important as Alibaba, and CSIG (Tencent Cloud and Smart Industry Business Group) has been streamlining personnel in the past two years; Second, Tencent hopes to focus more on its own hybrid model.
Those who have not yet invested in large-scale model startups have their own logic.
According to the Q3 2023 financial report, cash and cash equivalents at the end of the period were less than 35 billion yuan, while Byte's cash reserves exceeded 30 billion US dollars in the same period.
With limited resources, the priority is to do a good job of their own large models and corresponding products. A person close to him said that last year, it was clear not to invest in large model companies, focusing on application companies.
ByteDance, which has a large amount of investable funds, has wavered several times.
According to "LatePost", Byte once considered investing in large model companies Minimax and StepLeap Star in the first half of 2023, but ultimately did not make a move.
In mid-2023, Byte decided not to invest in large model companies, but to focus on making its own large models, which is the key for Byte to seize the opportunity of super apps in the AI era.
And just recently, Byte once again contacted the leading large model companies to reassess the necessity of investment.
The background is that the progress of byte's own development of large models has not met expectations; Alibaba's aggressive move and the recent undercurrent of large-scale investment and financing competition have also brought some pressure.
OpenAI's generative model, SORA, which was released just last week, is a new shock.
Previously, one view of the market was that Bytes, which owns Douyin and TikTok, and Google, which owns YouTube, has a lot of data and has the advantage to make multimodal models.
The strong performance of SORA shows that a strong enough large language basic model can cover other modal capabilities such as **, and the technical capabilities of the model may be more important than data resources.
Large Internet technology companies that originally had access to ready-made content or social products, and then massive amounts of data, have had to reassess whether they can stay ahead. "Bytes may be a little panicked right now, they don't have any investment cards in their hands before. A practitioner said.
The dark side of the moon, which was invested by Alibaba and Meituan, was established in March 2023 and was founded by Yang Zhilin, an assistant professor at the Institute for Interdisciplinary Information Sciences at Tsinghua University.
The 32-year-old graduated from Tsinghua University with a bachelor's degree in computer science and technology and pursued a Ph.D. from Carnegie Mellon University's Institute for Language Technology (LTI). During this time, he was the first author of Transother-XL and XLNet; The former is the first attention language model to surpass RNN across the board, and the latter surpasses the Google BERT model in 20 tasks. Yang Zhilin himself has more than 20,000 academic citations since 2019.
Many investors close to Yang Zhilin commented on him: he is rare and smart, and his growth rate is amazing. The Dark Side of the Moon features a focus on Chinese, long text, and to C product drives. In October 2023, the first model released by the dark side of the moon, Moonshot and the smart assistant Kimi, can support the input of 200,000 Chinese characters, which is 25x, OpenAI's GPT-4-32 K (about 2..)50,000 words).
An AI practitioner said that at this point in time in the second half of last year, the dark side of the moon is also one of China's leading large model companies with relatively high investment performance: after its 2 billion yuan Series A financing in October last year, it was valued at about $800 million, lower than Minimax and Zhipu AI, which were valued at nearly $2 billion at that time, and Baichuan Intelligence, which was already a unicorn.
A single round of $1 billion doesn't often happen early in an industry. Even in the last wave of AI boom of the financing boom, it took 4 years for SenseTime to raise a total of $1 billion, about 7 years for Megvii, and the dark side of the moon has only been established for 11 months.
Behind the abnormal investment is an extraordinary opportunity and high expectations of investors. Alibaba, Meituan, and others' Series B investments in the dark side of the moon may detonate the FOMO (Fear of Missing Out) dust that has long been permeated like a spark.
Title image**: Avengers 4: Endgame