Post

Actually, my judgment of OpenAI last year was quite simple: it essentially made empty promises, relying on round after round of funding and stories to support its valuation, and most likely ended up being acquired by a super platform like Apple, trading user base for a dignified exit. Now I don't think that way anymore. It's not that it suddenly built a moat, but rather that its series of actions over the past six months have made me feel the situation is much more complicated than I had originally expected. I don't want to split OpenAI's future into several mutually exclusive outcomes to calculate probabilities, because in reality, it's not about choosing one of four, but about betting on it simultaneously. On the consumer side, products, enterprise markets, and self-built ecosystems, it pursues all three paths, and none have been abandoned. What is truly worth watching is not which path will ultimately win, but whether these three can complement each other and ultimately form a genuine business closed loop. The first is the consumer side. ChatGPT remains the most influential AI product globally and OpenAI's most important traffic gateway. But the consumer side has a natural problem: the switching cost is too low. Use ChatGPT today, Claude tomorrow, Gemini the day after—as long as the experience is better, users can migrate in minutes. Historical dialogues and memory functions do create some stickiness, but this stickiness is far less than operating systems, office software, or social platforms. So I believe the consumer side is more like OpenAI's cash flow source than a long-term moat. The second is about the enterprise and the agent. This is also what I believe is OpenAI's most important step. Many people only see Claude's growing popularity in the enterprise market, but don't see the reasons behind it. What companies truly buy is never the model, but the workflow. When a company integrates AI into CRM, ERP, R&D processes, knowledge bases, customer service systems, or even the entire agent ecosystem, replacing models means redeveloping, revalidating, and retraining employees, with migration costs far higher than on the consumer side. This is why the value of the enterprise market far exceeds that of the consumer end. Whoever enters the corporate workflow first is more likely to build long-term revenue. Here, OpenAI faces not only Anthropic, but also platform companies like Microsoft, Google, Salesforce, and Palantir. The third point is something many people tend to overlook. What OpenAI truly deserves attention is not computing power, but scale. Many say models are becoming increasingly homogeneous, so OpenAI cannot become the AWS of the AI era. I used to think so too. But later I realized that AWS's moat came not just from migration costs, but from scale. The larger the scale, the stronger the procurement capacity, the higher the utilization of infrastructure, and the lower the unit cost. OpenAI is now investing heavily in long-term computing power contracts, self-developed chips, and inference optimization. If requests continue to grow in the future, it may not become an ecosystem moat like CUDA, but it has the potential to create a cost advantage brought by scale. These advantages are invisible to users but directly reflected in profit margins. The fourth is the entrance. Many people see browsers and AI hardware as OpenAI's future. On the contrary, I feel that the real entrance might not be these. But the developers. Today, a large number of applications are built on the OpenAI API. The migration cost for each application is not high, but when thousands of developers and millions of apps are built within the same ecosystem, this stickiness gradually accumulates. It's not as obvious as a corporate contract, but it could be OpenAI's most easily overlooked card. As for becoming Netscape in the AI era, I place it as a tail risk that requires ongoing monitoring. I don't think it will fail just because the model is outdated. What really needs to be worried is that if future models become cheaper and more homogeneous, and OpenAI still hasn't built its own platform ecosystem, it might end up like Netscape, driving the entire era without taking the most profits from the industry chain. There are many such examples in history. Browsers have changed the internet, but the ones who really make money are Microsoft and Google. Xerox invented graphical interfaces, but the ones who truly made money were Apple and Microsoft. Will AI repeat the same story? I think it's entirely possible. So now I increasingly agree with one viewpoint. OpenAI's biggest competitor may not be Claude. The real competitors are all platform companies with entry points. Apple has iOS. Google owns Android, Search, and YouTube. Microsoft owns Windows, Office, and Azure. Meta has the world's largest social ecosystem. Even if these companies don't have the best models in the industry, they can seamlessly integrate AI into products used daily by billions of users. In contrast, OpenAI's biggest asset remains ChatGPT. But if, in the future, AI no longer needs to be opened separately and is integrated into all products like search boxes or input methods, how much value can ChatGPT retain as an entry point? This is a question I've been pondering all along. So, my judgment now is different from last year. Last year, I doubted OpenAI could survive. Now I believe it will survive and remain one of the most important companies in the AI era. But my biggest question has shifted from technology to business model. OpenAI has proven itself to possess world-leading R&D capabilities and strong product capabilities. I believe a truly personal platform moat has not yet been established. What determines OpenAI's fate is no longer whose model is smarter, but who can turn the model's advantages into ecosystem advantages. Technology determines the product. The business model determines profit. And the ecosystem determines a company's ceiling.