Summary of Key Points
Three years ago, starting a business in the field of large-scale models was like competing for train tickets during the Spring Festival travel rush—everyone focused on parameters, financing, and the team, without caring about the financial logic behind the projects. Now, the once-popular “Six Little Tigers” have diverged in their approaches (going public, focusing on specialized medical applications or enterprise solutions, betting on open-source models, etc.). This shift reflects a transformation in the industry from relying on speculative spending to calculating actual business benefits. Investors no longer solely look at technological gimmicks but are more concerned with revenue quality, profitability, and real customer needs, leading to a stratified market landscape.
1. Post-IPO: Financial Reports Are a Must
Companies like Zhipu and MiniMax saw significant stock price increases after going public (Zhipu’s stock price rose by over eight times), but the volatility was also alarming (with a single-day drop of 28%). More importantly, their financial reports revealed stark realities: Zhipu generated over 700 million yuan in revenue in 2025 but spent 3.1 billion yuan on research and development (4.4 times its revenue), resulting in a net loss of 310 million yuan; MiniMax earned less than 80 million US dollars in revenue but invested 250 million yuan in R&D (3.2 times its revenue), also incurring a loss of 250 million yuan.
In simple terms: Before going public, companies could rely on the narrative of being the “Chinese version of OpenAI” to attract funding, but once they are listed, their financial performance must be made public. Losses cannot be concealed for long. The market may offer high valuations for AI startups, but it won’t continue to support them indefinitely—stock prices have already risen based on speculation, and now they need to prove themselves with actual revenue (not just by spending money).
2. Shifting to Niche Areas: Not Giving Up, but Getting Closer to Customers’ Needs
Companies like Baichuan and LingyiWanwu are focusing on specialized fields rather than general-purpose models. For example, hospitals buy AI solutions not because of the complexity of the models but because they can help with more accurate diagnoses; factories purchase AI systems not for the number of model calls but because they can reduce production downtime (which can save hundreds of thousands of dollars).
In simple terms: In the past, startups claimed their technology was the best, but when customers asked why they should pay, they couldn’t provide a convincing answer. Now, by focusing on niche areas, companies avoid competing with giants like Baidu and Alibaba for the largest models. Instead, they target smaller markets where customers are willing to continue spending—such as AI-powered personal assistants or enterprise intelligence systems that become essential parts of daily operations, leading to recurring revenue.
3. Market Stratification: Three Layers of Players
The large-scale model market will be divided into three tiers:
1. Lower Layer: Cloud providers (like Alibaba Cloud and Tencent Cloud) and a few leading model companies, responsible for providing computing power and basic models (the foundation of the industry). These companies need to spend heavily, and only a few will survive in the long run.
2. Middle Layer: Players from industries like healthcare and finance, who don’t need the largest models but can integrate AI into their data (such as medical records) and compliance requirements (financial regulations). It’s difficult for customers to switch to other providers due to high migration costs.
3. Upper Layer: Intelligent systems designed for individuals or specific tasks (e.g., writing sales scripts or code for programmers), charged based on usage or effectiveness (e.g., charging 5 yuan for writing a marketing copy that saves an enterprise 10 hours of work).
In simple terms: Small businesses don’t need to build their own power plants (like the lower layer); they just need to know which products will attract customers and ensure they provide value. Startups should focus on niche markets to thrive.
4. Smarter Capital: From Storytelling to Financial Performance
Previously, investors focused on things like the number of model parameters and rankings. Now, they ask: Is revenue growing quickly? Is the gross margin high? Is R&D spending worthwhile? Will customers renew their subscriptions? Do we have enough cash to sustain operations?
In simple terms: Stock prices may rise based on speculation about AI’s future dominance, but quarterly financial reports provide a reality check. Zhipu’s 28% stock price drop indicates market concerns about its profitability. The next round of competition won’t be about who has the biggest story; it’ll be about who can turn model usage into profit, demos into actual sales, and users into regular customers. The game has moved from the lab to the cash register—real revenue is essential for survival.
Final Conclusion
AI isn’t dead; what’s fading is the illusion that great technology automatically leads to success. The current large-scale model industry is starting to behave like a normal business, focusing on customer needs, potential profits, and sufficient funding to achieve profitability. That’s the key to long-term success.