On March 5, according to sources, Alibaba led a new round of financing for Minimax, a large AI model company, with a post-investment valuation of more than $2.5 billion.
This is the second time this year that Ali has shot in the field of large models. In mid-February, according to a late report, the dark side of the moon completed a round of $800 million in Series B financing, this round of investors are Alibaba and Lisi Capital. At the end of this round, the Dark Side of the Moon is valued at over $2.3 billion.
Counting these two investments, Alibaba has invested in 5 AI large model companies, namely the Dark Side of the Moon, MiniMax, Zhipu AI, Baichuan Intelligence, and Zero One Everything. It is worth mentioning that these 5 companies are currently the only AI large-scale model unicorns in China.
In other words, Ali voted for the companies that are most likely to become China's OpenAI in the primary market with real money**, and also placed heavy bets on the dark side of the moon and minimax. It is no exaggeration to say that Ali is now the ETF of "China OpenAI".
01 Alibaba created the "China OpenAI" ETF
Before investing in the dark side of the moon and minimax, Alibaba invested in large-scale model startups Zhipu AI, Baichuan Intelligence and Zero One Everything, respectively, of which the investment in Zero One Everything was completed as Alibaba Cloud. Considering that Alibaba Cloud has not completed the split, it can be understood as an investment by Alibaba.
It is worth mentioning that compared with the past, Ali has made a more decisive and bold move in the field of large models this year. In the angel wheel of zero and one thousand things, there is only one investor in Alibaba Cloud. In the financing of Baichuan Intelligence and Zhipu AI, Ali participated with a number of institutions, and did not appear as the leader of the investment.
But in these two investments this year, Ali highlighted a "heavy start". According to the previous ** news, most of the funds in the ** financing of the dark side of the moon came from Ali, which also made Ali the largest institutional shareholder outside the dark side of the moon team after this round of financing. In the new round of financing of Minimax, Alibaba still plays the role of the lead investor.
Considering it in the context of the development of the large model industry, Ali has made continuous large-scale moves at the moment, not for no reason.
Because large models require large-scale computing power investment and continuous iteration, and leading models will form obvious advantages in terms of capabilities and costs, it will be difficult for players behind them to survive, resulting in a high threshold for entrepreneurship and a very short investment window for large models. In the words of Chen Yu, a partner of Yunqi Capital, the investment in the large model actually ended before the release of ChatGPT.
Today, the play-in competition in the field of large models has ended and is entering the knockout stage. The five companies invested by Ali are the companies most likely to become openai selected by the primary market with real money**, although three of the five were only established in 23 years, and the other two only started to make a lot of money last year.
02 Casting a net to cast AI, what is the picture?
In the Internet industry, such a net-wide investment like Alibaba is still Tencent's investment in the game field around 2020.
The shift in Tencent's investment strategy stems from the reflection of James Michel, the company's chief strategy officer: the gaming industry should not have such a high ROI, which only shows that Tencent has not failed enough. Why aren't there so many cases of failure? That's because we don't look enough and invest too conservatively. 」
The speech took place in early 2020. Since the beginning of this year, Tencent's investment pace has significantly accelerated, shifting from conservative to casting a wide net, including domestic and foreign markets. Daniel Ahmad, an analyst at Niko Partners, a data agency, pointed out that in 2021 alone, Tencent acquired and invested in more than 100 game companies, equivalent to one every three days, of which 30% were game companies outside of China, setting a record for the fastest investment in the game field in the world.
At that time, Tencent's logic was that investment was the most efficient means to promote the globalization of the game business. So, why does Alibaba invest in the field of AI large models?
A very important reason is that for Alibaba, which has the strongest cloud computing business in China, investing in a large model enterprise is an extremely cost-effective business.
On the surface, Ali has invested a lot of money in the large model, but because the large model company needs a lot of cloud computing resources in the training and inference process, it means that the money invested by Ali can finally be returned to Ali in the form of cloud computing income, which does not include part of the investment amount itself is paid with computing power. Back and forth, there is no shortage of money, and it can also drive the growth of Alibaba Cloud's revenue, no matter how you count it.
Not only that, Ali can also become a distributor of large-scale enterprise products, and keep up with the latest technology of large-scale model companies in a timely manner. This can be referred to Microsoft's investment in OpenAI.
After a large investment, Microsoft not only enjoys the priority right to use OpenAI's advanced technology, but also obtains the exclusive commercialization authorization of OpenAI's technology, and its cloud computing platform Azure has become the exclusive cloud service provider of OpenAI.
From the current point of view, due to customer resources and security considerations, the Azure OpenAI API sells better than the OpenAI API. According to Microsoft, the company is on track to earn $10 billion or more in annual revenue from developers using Azure cloud or OpenAI models.
Aside from business-level considerations, Alibaba's largesse on the big model also has a lot to do with the management background. Now, Tsai Chongxin, chairman of Alibaba Group, is from a VC background and has served as Ali's war capital, while Wu Yongming is the founder of Yuanjing Capital. In other words, the two have a relatively deep investment background and rich investment experience, which can be regarded as a way for Alibaba to enhance its influence in the field of AI and cloud computing.
And back to the perspective of large-scale model unicorns, Ali, who is willing to enter the game, is also one of the few ideal choices for them.
On the one hand, the amount and quantity of investment in the country of US dollar VC are shrinking; On the other hand, most of the LPs of RMB are **guidance** or industrial capital. According to Qingke data, in 2023, in the distribution of new RMB **LP raised by Chinese equity institutions, the proportion of ** institutions and industrial capital will be as high as 406% and 267%。The typical large model of heavy investment, inability to see the business model in the early stage, and great risk is not in line with the investment preference of RMB**.
At least for now, this is a win-win situation for Ali and the big model unicorns.
03 Giants in the narrative of large models
Throughout the development of large-scale model companies, giants are an unavoidable topic. In the past wave of technology, large companies tended to try their own teams first, but in the LLM era, everyone regarded time and capital as the most valuable resource, and preferred to invest in and acquire companies that can run faster than themselves.
This is the case for two reasons:
In this technology-driven generative AI competition, large-scale model startups play a more leading and important role than traditional technology giants. Needless to say, Wensheng's ** large model sora put Google's gemini 15 stole all the limelight. And just the night before, Anthropic released the Claude 3 series, which even showed capabilities beyond GPT-4 in terms of multimodality and language capabilities.
Due to the solidification of the organizational structure, it is difficult for giants to cope with the changing competition in the field of large models. For large companies with a highly mature organizational structure, it is difficult to support them from the organizational structure to the incentive mechanism.
But what is interesting is that in terms of giant investment in large models, domestic and overseas are also showing a completely different situation.
Overseas, large-scale model companies are almost the "spokespersons" of giants competing in the AI field, and they are clearly distinguishing from each other. For example, OpenAI has Microsoft behind it, Google and Amazon behind Anthropic, and XAI has Truss's support behind it. Not long ago, Mistral AI, known as the "French version of OpenAI", announced that it had accepted Microsoft's investment.
In China, Ali has voted for all the domestic AI large model unicorns, among which Tencent has also appeared in Zhipu AI, Baichuan Intelligence, and Minimax. Of course, this has a lot to do with changes in the external environment, but at the same time, it also shows one thing: the competition in the field of domestic large models is far less fierce than that of foreign countries.
Big model companies are getting closer and closer to the giants, even subverting the conventions of the past history of technology competition - the rise of new giants is often achieved by subverting the old giants. In this case, how should we view the role of giants in the era of large models? Consensus Shredder makes an interesting point about this issue: big companies in the LLM era are like snakes.
On the one hand, they are trying to occupy the entrance to the scene, trying to see where LLM can be applied. On the other hand, they continue to absorb various industrial resources to build their own ecology: it may be the software ecology represented by the public cloud, the creative ecology represented by the content, or the first-chain ecology represented by the end side.
To a certain extent, Alibaba's comprehensive investment in large-scale model companies is a kind of scene occupancy and ecological construction. Compared with Tencent, which is completely self-developed and more low-key, Ali seems to be more willing to bet heavily on the large model table. Whether the abnormal investment can bring more returns than expected can only be answered by time.