Post

Actually, my judgment of OpenAI last year was very simple: I thought it was essentially writing blank checks, propping up its valuation through round after round of financing and storytelling, with the most likely outcome being acquisition by a super platform like Apple, exchanging user scale for a decent exit. Now I don't think that way anymore. Not because it suddenly built a moat, but because a series of moves in the past six months have made me feel the situation is much more complex than I originally thought. I don't want to break down OpenAI's future into several mutually exclusive outcomes to calculate probabilities, because in reality it is not choosing one out of four, but betting on all simultaneously. Consumer products, enterprise market, and self-built ecosystem—OpenAI is pursuing all three paths without abandoning any. What’s truly worth observing is not which path will ultimately win, but whether these three can complement each other and eventually form a real commercial closed loop. The first is the consumer side. ChatGPT remains the most influential AI product globally and OpenAI’s most important traffic entry point. But the consumer side has a natural problem: switching costs are too low. Today you use ChatGPT, tomorrow Claude, the day after Gemini—users can migrate within minutes if the experience is better. Historical conversations and memory functions do create some stickiness, but this stickiness is far less than that of operating systems, office software, or social platforms. So I believe the consumer side is more like OpenAI’s cash flow source rather than a long-term moat. The second is enterprise and agents. This is also what I consider OpenAI’s most important step. Many only see Claude becoming more popular in the enterprise market but don’t see the underlying reason. What enterprises truly purchase is never just 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 system, replacing the model means redeveloping, revalidating, retraining employees—the migration cost is much higher than on the consumer side. This is why the enterprise market’s value far exceeds that of the consumer side. Whoever first enters enterprise workflows is more likely to establish long-term revenue. Here, OpenAI faces not only Anthropic but also platform companies like Microsoft, Google, Salesforce, Palantir, and others. The third point, which many tend to overlook, is scale rather than computing power. Many say models are becoming more homogeneous, so OpenAI can’t become the AWS of the AI era. I used to think so too. But later I realized AWS’s moat comes not only from migration costs but also from scale. The larger the scale, the stronger the purchasing power, the higher the infrastructure utilization, and the lower the unit cost. OpenAI is now investing heavily in long-term computing contracts, self-developed chips, and inference optimization. If request volume continues to grow, it may not form an ecosystem moat like CUDA, but it has the chance to create a cost advantage brought by scale. This advantage is invisible to users but will directly reflect in profit margins. The fourth is the entry point. Many consider browsers and AI hardware projects as OpenAI’s future. I rather think its real entry point may not be these. It’s developers. Today, a large number of applications are built on OpenAI’s API. The migration cost for each application is not high, but when tens of thousands of developers and millions of applications are built on the same ecosystem, this stickiness will slowly accumulate. It’s not as obvious as enterprise contracts but could become OpenAI’s most easily overlooked card. As for becoming the Netscape of the AI era, I place that as a tail risk that requires continuous observation. I don’t think it will fail because its models fall behind. What really needs worrying is if models become cheaper and more homogeneous in the future, and OpenAI still hasn’t built its own platform ecosystem, then it might be like Netscape—pushing the entire era forward but not capturing the most profit in the industry chain. There are many such examples in history. Browsers changed the internet, but Microsoft and Google made the real money. Xerox invented the graphical interface, but Apple and Microsoft were the ones who profited. Will AI repeat the same story? I think it’s entirely possible. So now I increasingly agree with one view. OpenAI’s biggest competitor may not be Claude. The real competitors are all platform companies that own entry points. Apple owns iOS. Google owns Android, Search, and YouTube. Microsoft owns Windows, Office, and Azure. Meta owns the world’s largest social ecosystem. Even if their models aren’t industry-leading, these companies can seamlessly integrate AI into products billions of users use daily. In comparison, OpenAI’s biggest asset remains ChatGPT. But if AI no longer needs to be opened separately and instead integrates like a search box or input method into all products, how much value will the ChatGPT entry point retain? This is a question I’ve been pondering. So my judgment now is different from last year. Last year I doubted whether OpenAI could survive. Now I believe it will definitely 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 it has world-leading R&D capabilities and strong product capabilities. I think it has not yet truly established its own platform moat. What will decide OpenAI’s fate is no longer whose model is smarter, but who can turn model advantages into ecosystem advantages. Technology determines products. Business models determine profits. And ecosystems determine a company’s ceiling.