When it comes to large models, Robin Li made a choice that was different from his counterparts in big manufacturers: open source the model. In the eyes of many people, this path is tantamount to handing over the most core things to others, with little return in sight in the short term.
His logic is exactly the opposite. Closed source models can receive money, but only if the model is strong enough and users are willing to pay. What open source can gain is ecology and usage scale. When enough developers build things on your technical base, long-term value will come back in another way.
The search engine gave him the foundation
To understand Robin Li’s choice, we have to go back to his old profession. A search engine is essentially a large-scale information processing system, backed by massive data, complex algorithms and continuous iterative engineering capabilities. What has been accumulated over the years has new uses in the AI era.
He has made a judgment many times: the value of AI lies in its application, not in how high the parameters of the model are stacked. This view is not mainstream in the technical circle, because the focus of public opinion at that time was whose model was stronger and ranked higher on the list.
Why does he insist on talking about applications?
Robin Li’s judgment is based on a very practical observation: no matter how advanced the technology is, if it fails to fall into specific scenarios, it will not generate revenue and cannot support the next round of investment. The model is the cost center and the application is the source of revenue.
This judgment determines how the company allocates resources. When his peers were comparing model sizes, he focused more on how to connect the model to search, office, and various specific businesses. This is a path that doesn’t look sexy enough, but has better business logic.
price of bet
The price of choosing application priority is falling behind in terms of technical sound. In the public evaluation of model capabilities, if a company’s performance is not outstanding, it can easily be labeled as having insufficient technical capabilities. For a founder who has a technical background and made achievements in algorithms in his early years, this kind of evaluation is not pleasant.
Another cost is that application first means bearing the cost first. The investment in model training is front-loaded, and application monetization is paid back. During this period, the financial statements will be ugly. To withstand this pressure requires management’s patience and ability to explain.
How to play the open source card
Open-sourcing the model is a natural extension of his logic. Open source lowers the threshold for use, allows more developers to use this technology at low cost, and can also identify problems faster and accelerate iteration. For a company looking to build a presence at the app level, scale is more important than short-term licensing revenue.
Of course, open source also has risks. Competitors can also use these technologies, and it is difficult to form a technical exclusive advantage. Therefore, the real barrier lies not in the model itself, but in engineering capabilities, scenario understanding, and user scale. This choice is tantamount to placing the company’s winning hand on what you are better at.
What will time test?
The success or failure of this strategy depends on a key issue: whether the revenue generated by the application can cover the continued investment in model research and development and form a positive cycle. If it does, the seemingly slow path will show its advantage; if not, the early vocal disadvantage will be magnified.
Robin Li’s bet is that the final form of AI is an infrastructure layer, and the application ecosystem that grows on it is where the value lies. It will take a long time to see whether this judgment is correct.
organizational adjustments
The choice of technical route will directly affect the company’s organizational style. Focusing on applications means that the importance of product, engineering, and implementation teams will increase, and the relative weight of pure research teams will change. This kind of adjustment is something that requires determination for a company that started with technology.
Another change is the pace. Doing research can follow the cycle of papers and reviews, but making applications must follow the rhythm of users and the market. For an organization that is accustomed to the former to adapt to the latter, there will be a lot of friction in the process.
The thing he talks about most
Robin Li has repeatedly said one thing in public over the years, which is not to get caught up in an arms race in model capabilities, but to focus on what problems technology solves. This expression sounds simple, but in an industry where everyone is comparing parameters, speaking out requires a stance.
Behind this expression was his business intuition. Search engine experience told him that technological leadership does not equal commercial success. What users really pay is whether the problem can be solved. This judgment may not be applicable to all scenarios, but at least it is self-consistent in its logic.
How to understand this bet
The choice of an enterprise’s technical route is never just a technical issue, but also a resource allocation issue. Choosing a path means choosing where to invest limited resources, and also choosing what price to bear. The one chosen by Robin Li has a high cost in terms of volume and a clearer path to closed-loop business.
Related Reading
- US road rage killer’s sentence quashed because AI video of victim was shown in court2026-10-03
- What we’re learning about the FlyDubai co-pilot who allegedly stabbed the captain and tried to crash the plane2026-10-03
- Profits shrunk by 90%, Yingshi expanded while cutting costs2026-10-03
- New forces reshuffle in September: Xiaopeng and Xiaomi sell for 40,000, Hongmeng Ideal retreats2026-10-03
- OpenAI fires workers for mishandling ‘sensitive information’2026-10-02
