The future of domestic independent large model companies is estimated to be overseas expansion + vertical To B

The financial reports of Hong Kong stock technology companies have been released, and the market's focus is undoubtedly on two large model listed companies—MiniMax and KNOWLEDGE ATLAS. From the financial reports, it is not difficult to find that the revenue distribution of the two companies is almost completely opposite: Almost all of KNOWLEDGE ATLAS's revenue comes from the domestic market, while 73% of MiniMax's revenue comes from overseas; The vast majority of KNOWLEDGE ATLAS's revenue can be defined as To B, while about two-thirds of MiniMax's revenue comes from To C (Note: AI native product revenue is not entirely, but mostly To C). 73.7% of KNOWLEDGE ATLAS's revenue comes from localized large model deployments, which are often highly customized based on industry and enterprise needs. It is said that Moonlight's Dark Side is also planning to go public. We have not yet seen the prospectus, but according to media reports, its overseas revenue growth in the first quarter of this year is very strong. Perhaps we will see Moonlight's Dark Side, similar to MiniMax, with most of its revenue coming from overseas? In any case, it will not primarily rely on the domestic market like KNOWLEDGE ATLAS. I believe this undoubtedly represents two viable development paths for domestic independent large model companies (this independence refers to being independent of large internet companies): going overseas and domestic vertical To B. It does not necessarily have to be an either-or situation; both can be pursued, but in reality, there will always be a focus. As for other businesses, including domestic To C and "general-purpose" To B, they will certainly still be pursued, but I do not believe independent large model companies have much competitive advantage in these areas. Why? Because domestic To C is clearly the territory of large internet companies. In terms of general AI applications, ByteDance's Doubao, Alibaba's Quark and Qianwen are very powerful, and Tencent's Yuanbao has not yet given up hope of entering the first tier. In terms of vertical AI applications, early success cases like Ant Group's Aifu also demonstrate that as long as large internet companies want to engage, their resource advantages and synergy effects will be very obvious. Large companies not only have advantages in technical infrastructure and traffic but are also willing to provide ample free services, making it very difficult for any independent large model vendor to generate significant revenue from the domestic C-end. Take DeepSeek as an example: it is still one of the top four or five AI applications in the country, but because it is open-source, competitors are willing to provide "large quantities and unlimited" free services to C-end users, so the possibility of DeepSeek charging domestic C-end users is extremely low, and it can only focus mainly on the B-end. For other independent large model vendors, unless they can launch advanced large models that are significantly superior to those of large companies and are closed-source, it is unlikely they will earn high revenue from the domestic C-end—this possibility, while not zero, is very slim. As for the domestic "general AI To B" business, it is even more a battleground for large companies, the lifeline of Alibaba Cloud, Volcano Cloud, Baidu Cloud, and Tencent Cloud. Independent large model companies can certainly leverage their flexibility to achieve results in some localized battles; for example, the recent "National Lobster Farming" trend has greatly increased the token demand for many large model vendors. However, do not forget that in today's overall rising computing power prices, independent large model vendors are price takers, while large companies with vast computing power infrastructure and cloud computing departments are price makers In the long term, independent large model vendors find it difficult to compete with major companies in terms of Token costs. Even if they develop technologies to reduce inference costs, major companies will clearly implement imitation and iteration in their next versions. Historically, many reshuffles in the domestic internet industry originated from "dominant companies not paying attention or being careless," which can be attributed to the rise of ByteDance, Pinduoduo, and even Kuaishou. However, in the field of AI large models, there is no such time window; major companies generally pay high attention from the very beginning, with the only disagreement being how much to invest and how to invest. In the domestic C-end applications, the emergence of a viral DeepSeek was already an unexpected surprise. Other independent vendors' applications, no matter how strong, can probably only settle for a leading position in the second tier, making it difficult to compete with the heavily armed applications of major companies. The To B market is another story: this market is too complex, with industry verticals being very important and customer relationships also being crucial. Many vertical customers are not lucrative enough for major companies, and the difficulty of acquiring them is too high. Although major companies generally want to provide MaaS (Model as a Service) and AI-based SaaS, regardless of who provides these services, most of the underlying computing power will come from major companies, and their IaaS business will still benefit. From both the customer perspective and the major companies' perspective, it is only natural for some B-end, especially vertical B-end businesses to be completed by independent large model vendors, and there is no need to seek change. As for going overseas, I need not say more: one of the hottest entrepreneurial tracks in the past three years has been AI application going overseas. The overseas market is very large, with significant internal differentiation, including developed markets like North America and Western Europe, as well as Southeast Asia and South Asia markets that Chinese tech companies are very familiar with, along with the increasingly important Middle East and Latin America markets, etc. Whether in the C-end or B-end, Chinese AI vendors (including large model and application developers) have two distinct comparative advantages: First is the application development and iteration capability. If domestic vendors claim to be second, probably no one would dare to claim first. As early as the initial stage of the mobile internet era, the fundamental motivation for many domestic vendors to go overseas was that the domestic market was too competitive; those who couldn't compete domestically often found themselves very competitive overseas. Moreover, with numerous overseas markets, differing user preferences and regulatory levels, there are always places where they can compete. Second is the cost-effectiveness of Token costs. Recently, many media reports have indicated that several domestic large model vendors have made a lot of money selling Tokens overseas, especially in developing markets, due to low inference costs and high cost-effectiveness, making them quite popular in certain countries. This is indeed a fact (although specific numbers are difficult to verify). So, what is the underlying reason? Some have mentioned "infrastructure" reasons like power supply—but the real reasons are likely more complex The low-priced tokens provided by domestic manufacturers mainly come from activating models with smaller parameter scales. Their upper limits are certainly limited, but they are more than sufficient to meet the demands of many developing countries. This "cost-performance competition" strategy is effective, and leading companies in Silicon Valley may not have the capability or willingness to imitate it, just as they find it difficult to compete with Chinese manufacturers on "cost performance" in many other application areas. Localized operations and promotions are, of course, very important, and the localization capabilities of Chinese manufacturers are well recognized, having been repeatedly proven even before the AI era. I believe that in the foreseeable future, the development path of domestic independent large model companies will simply involve choosing between "going overseas" and "domestic vertical To B," or possibly both. Personally, I prefer the overseas business, as domestic To B is a very challenging business with high demands on suppliers, which I deeply felt when researching the domestic software industry years ago. However, the To B market does exist, and enterprises across various industries in China do have a genuine need for AI transformation. Perhaps there are indeed independent large model companies that can achieve sustained success in this field? This article has not received any funding or endorsement from any of the large model companies or their competitors mentioned in the text. The author of this article currently does not hold shares in the large model companies mentioned, but cannot guarantee that they do not indirectly hold shares through funds