As the primary investors in domestic large model startups, Tencent and Alibaba once again stand in the position of 'sugar daddies.'
From 2023 to the present, these two former investment giants have coincidentally and significantly reduced their investment frequency. According to public data from Tianyancha, Tencent Investments made only 33 moves in the entire year of 2023, averaging fewer than three per month. In contrast, Tencent made 302 and 91 investments in 2021 and 2022, respectively.
However, when focusing on the large model-related sector, Tencent and Alibaba remain the most 'generous' investors in China. Earlier, renowned investor Zhu Xiaohu stated in an interview with Tencent Technology: 'These (Chinese large model) companies have no scenarios, no data—what value do they have? And their valuations are so high right from the start.' But obviously, Tencent and Alibaba don't think so.
Among the currently known five large model unicorn companies, Alibaba's investment participation rate is as high as 100%, while Tencent has participated in the investment of three, with a participation rate of 60%. In contrast, Baidu and ByteDance have not invested in any of them so far.
Choice is the most common keyword that internet giants face when the wave of large models comes. For corporate leaders, the choice lies between developing related businesses in-house or investing heavily in startups. For some executives, it's about deciding whether to leave the security of big tech firms to venture into what's considered the most promising entrepreneurial field in recent years. For ordinary employees, the dilemma is whether to stake their limited career prospects on the uncertain outcomes of new industry trends.
In recent years, a popular saying suggested that while 'AT' in 'BAT' still referred to Alibaba and Tencent, the 'B' had shifted from Baidu to ByteDance. Some Baidu employees even openly mocked their own company, claiming it no longer qualified as a major tech firm, and used '1 degree' (a pun on Baidu's name and stock value) as a unit to measure the market capitalization of internet companies.
In this race to invest in large AI models, AT continues to charge ahead, while both Baidu and ByteDance have chosen to take a wait-and-see approach. Currently, there are five large model companies in China with valuations reaching $1 billion: Moonshot AI, Zhipu AI, Minmax, 01.AI, and Baichuan Intelligence. As mentioned earlier, Alibaba has invested in all five of these companies, while Tencent has invested in three, excluding Moonshot AI and 01.AI.
It's noteworthy that neither Alibaba nor Tencent were early investors in these five companies. Taking Zhipu AI, the longest-established among them, as an example, Alibaba-affiliated entities joined in August 2023, with Ant Group becoming an investor in its B+++ round of financing. Prior to this, Zhipu AI had already undergone five rounds of financing over four years.
The investment in Zhipu AI in August 2023 also marked the first appearance of Alibaba-affiliated investment institutions among the investors of these five companies. Expanding the scope from the current five unicorns to all large model startups, Alibaba's first investment occurred in March 2023, when Ant Group participated in incubating Shengshu Technology, which focuses on multimodal generative large models and application product development. When Alibaba's first major investment in large models landed, its self-developed large model "Tongyi Qianwen" was soon to be officially integrated into DingTalk, and the standalone app was about to be opened to the public.
Tencent's investment actions in large models occurred almost simultaneously. In June 2023, Minimax completed a Series A financing of over $250 million, with public information showing that Tencent was the sole investor in this round. At that time, there were less than three months left before the official release of Tencent's self-developed Hunyuan large model application. In other words, while developing its own large model, Tencent has also been seeking external investment targets.
From this perspective, Tencent and Alibaba have similar strategic layouts in large models—whether through internal development or external investments, they seize every promising opportunity. In the field of large model investments, Baidu's 'wait-and-see' stance contrasts with the proactive approach of the pioneering 'AT' (Alibaba and Tencent). However, even within the 'wait-and-see' position, Baidu and ByteDance have different strategies.
Absent from the investor lists of any major large model unicorns, Baidu was initially perceived by outsiders as going all-in on in-house research in the large model domain, channeling all its resources and hopes into Wenxin Yiyan (Ernie Bot). However, in reality, Baidu is another investor in Shengshu Technology's angel round. While Alibaba shifted to higher-valuation projects, Baidu consecutively participated in three rounds of funding for Shengshu Technology.
In terms of project quantity alone, Baidu has been quite active. Since 2023, Baidu has invested in several large model-related projects, including Shengshu Technology, Xihu Xinchen, Baiying Technology, and WuWen XinQiong. From the comparison with Alibaba and Tencent, it's evident that Baidu follows two principles in large model investments: it avoids expensive projects and general-purpose large language models (LLMs). The five unicorns mentioned earlier are all currently valued at over $1 billion, and Baidu has not participated in any of their investments.
Baidu's four chosen investment targets differ from its self-developed Wenxin series products in their focus areas. Shengshu Technology primarily focuses on multimodal generative large models, Xinhu Xinchen mainly develops applications based on large models (such as AI writing and AI painting), Baiying Technology and Wu Wen Xin Qiong both operate in the B2B sector. The former develops AI dialogue services and marketing systems for enterprises and government departments, while the latter provides integrated hardware and software solutions for vertical domain large models—essentially building infrastructure for large models.
Investing in applications, multimodal, and infrastructure but avoiding LLM—this, from the results, appears to be Baidu's large model investment strategy based on the development status of its self-developed projects. At the very least, the decision-makers behind Baidu's investments believe that in terms of technical capabilities for general large models, there are no external frontrunners that significantly surpass the Wenxin large models to warrant heavy financial bets. Compared to Baidu's "invest a little" wait-and-see approach, ByteDance's attitude toward large model investments is more "intense." According to public information, ByteDance has not invested in any external large model company. However, this "intensity" can be interpreted as either conservatism or another form of radicalism—ByteDance is essentially betting all its chips on internal ventures.
When the targets in front of you become the objects of pursuit across the industry, explaining why not to invest might be harder than justifying why to invest. Zhu Xiaohu represents a pure investor's perspective, focusing solely on the commercial returns of individual projects, which naturally allows for simpler and clearer logic. But for large corporations with complex business interests, there are far more considerations at play.
What matters is not just 'belief,' but also position and stance. ByteDance did have moments of temptation. According to LatePost: "In the first half of 2023, ByteDance once considered investing in AI model companies MiniMax and StepFun, but ultimately refrained." This was precisely when Tencent and Alibaba began investing in large model companies through strategic investments. In other words, ByteDance once stood at the same crossroads as these two companies but ultimately chose a different direction.
ByteDance's different choice may stem from two reasons. First is its confidence in its in-house R&D capabilities, especially in application-layer product development. Over the past decade, ByteDance has created products with profound impacts on users nationwide and even globally. The speed of product development and the agility of product iterations once earned it the nickname "App Factory."
With ample cash reserves, a strong talent pool, and rich experience in application-layer R&D, these three advantages may be among the reasons why ByteDance decided to heavily invest in its in-house R&D path starting last year. Since January 2023, ByteDance has formed multiple teams to develop large language models (LLMs) such as Cloud Lark and multimodal models like BuboGPT at the model layer. At the application layer, the company has invested in the development of more than 10 products, some of which have already been launched, while others remain confidential.
To better integrate resources, ByteDance established a new AI department called 'flow' at the end of last year. The technical leader is Hong Dingkun, ByteDance's Vice President of Technology, while Zhu Wenjia, the head of the large model team, also serves as the business leader of flow. The AI conversational product Doubao, which began testing in August 2023, operates as an independent business line under flow. The other three business lines are AI education, internationalization, and community. For example, the recently launched AI character interaction app Hualu belongs to the community business line under flow.
According to Hedgehog Commune, the flow team mainly consists of two groups. One group comprises employees who transferred internally from ByteDance, including some former members of the Pico team, Douyin team, and Xigua Video team. The other group comes from external recruitment, with a focus on hiring employees who have experience with large model products like Baidu's ERNIE Bot. One reason ByteDance hasn't invested in large language models through external investments might be due to considerations about its existing business needs.
Currently, the main investment targets in the market are still LLM projects. However, text understanding and generation don't align as closely with ByteDance's core business operations. Technologies like text-to-audio and text-to-video would have a more profound impact on ByteDance's business.
On the other hand, more aggressive investors like Tencent and Alibaba, including Baidu which tests waters with smaller projects, all possess mature cloud service businesses. Their early investments aim to secure more industry dividends as AI infrastructure providers for their cloud services. After all, Alibaba Cloud, Tencent Cloud, and Baidu Cloud are all among China's top ten cloud service providers. But ByteDance clearly hasn't completely shut the door on investments. Its determination to go all-in on in-house development might waver slightly after witnessing the impressive debut of text-to-video product Sora and the rapid growth of its cloud service platform Volcano Engine. The same report from LatePost mentioned: "Recently, ByteDance has again engaged with leading large model companies to reassess the necessity of investments." This occurred in February of this year.
Everyone knows AI is the ship sailing toward the next era, but there are countless boarding passes. At least for now, no one can determine which pass is genuine and which might lead people onto a soon-to-sink giant vessel.
Some say domestic tech giants' investments in large models stem from a kind of "FOMO" panic. That's the acronym for "fear of missing out"—though I don't understand why four English letters are needed to replace "missing out," but this industry often operates this way. Of course, regardless of what it's called, the panic about "missing out" is completely understandable. The course of history has proven that ChatGPT wasn't created by Google or Microsoft, but by OpenAI; similarly, TikTok didn't emerge from Alibaba, Tencent, or Baidu. Being in this environment, it's hard not to feel anxious, wondering if the ships sailing to the next era must inevitably depart from new docks.
The result of FOMO is that leading internet companies are placing their bets on multiple ships through investments or in-house development.
Some also say that whether to invest depends entirely on the decision-makers' "belief in large models" - a faith similar to what they once had in mobile internet, not the kind where Chen Guilin takes one shot at a time. In fact, it's not just the investment departments of internet giants that are facing choices, but also the executives and ordinary employees within these companies.
For example, Wang Huiwen, co-founder of Meituan, has already completed a round in the large model entrepreneurial track. In February 2023, it was Wang Huiwen's widely circulated "heroic post" that fired the first shot of executives from major companies entering the large model entrepreneurship. However, four months later, Wang Huiwen temporarily withdrew due to personal health reasons, and Meituan acquired Light Year Beyond for 2.065 billion yuan.
On April 2, 2024, Meituan CEO Wang Xing issued an internal email stating that Wang Huiwen is about to return and will serve as a part-time advisor to Meituan, but did not disclose whether his specific business direction is still related to AGI. Over the past year, an increasing number of executives from major tech companies have made similar career choices.
According to statistics from tech media outlet Zhidongxi, 14 executives have left their positions at big tech firms to launch AIGC startups. Three each came from Alibaba and Baidu, while the remaining eight originated from companies like Microsoft, Meta, ByteDance, Didi, and Meituan.
While corporate investments represent financial bets on potential opportunities, executives bring their own resources—even capital—to the table. For ordinary employees, however, choosing to dive into large model entrepreneurship means gambling career stability for explosive growth potential. For them, making the wrong choice carries significantly higher risks. A senior internet industry employee commented: "This is different from a regular job change. If you were working on information flow and switched to another company to work on growth, the gap wouldn't be as significant—these are within the same business system. Your accumulated experience would remain valuable for future career moves. But transitioning to large model development means entering a completely different system. If unsuccessful, returning would present far more challenges."
Thus, from investors evaluating large models to employees considering entrepreneurial opportunities in this field, everyone faces the same fundamental question—'Do you truly believe in large models?'
Within major tech companies, transferring to large model-related departments is considered a relatively safer option. It allows professionals to stay within their familiar corporate environment without adding another job change to their records. Faced with the temptation of new unicorns, many ordinary employees at large tech companies are still wavering. After completing a Series A funding round of over $1 billion, Dark Side of the Moon posted several job openings for growth product roles. A product manager specializing in growth at a major internet company once hesitated about applying but ultimately decided against it, feeling that "the scope of what PMs can do at this stage is quite limited."
However, on the other side of the hesitation, many have already chosen to charge ahead.
At the end of March, Dark Side of the Moon organized a movie screening event. In the event group chat, many participants from the internet industry expressed their desire to join the company. Meanwhile, the growth product positions have already been removed from the recruitment website, indicating that this star AI company has swiftly filled its vacancies with suitable candidates.
