Beneath the surface of its success, OpenAI faces four major challenges.

Written by Zhao Ying , Wall Street Insights

Benedict Evans, former partner at a16z and a renowned technology analyst, recently published an in-depth analysis pointing out four fundamental strategic dilemmas facing OpenAI behind its apparent prosperity. He argues that despite OpenAI's large user base and ample capital, its lack of a technological moat, insufficient user stickiness, rapid competition from rivals, and product strategy constrained by laboratory research directions are threatening its long-term competitiveness.

Evans points out that OpenAI's current business model lacks a clear competitive advantage. The company has neither unique technology nor network effects; only 5% of its 900 million weekly active users are paying customers, and 80% of users will send fewer than 1,000 messages by 2025—equivalent to less than three notifications per day on average. This "mile wide, inch deep" usage pattern indicates that ChatGPT has not yet become a daily habit for users.

Meanwhile, tech giants like Google and Meta have caught up with OpenAI technologically and are leveraging their distribution advantages to seize market share. Evans believes that the real value in AI will come from new experiences and applications that have yet to be invented, and OpenAI cannot create all of these innovations on its own. This forces the company to fight on multiple fronts simultaneously, with a comprehensive layout from infrastructure to the application layer.

Evans' analysis reveals a core contradiction: OpenAI's attempt to build competitive barriers through massive capital investment and a full-stack platform strategy raises questions about the effectiveness of this strategy in the absence of network effects and user lock-in mechanisms. For investors, this means a need to reassess OpenAI's long-term value proposition and its true position in the AI ​​competitive landscape.

Technological advantage disappears: Model homogenization intensifies

In his analysis, Evans points out that there are currently about six companies capable of launching competitive, cutting-edge models with roughly equivalent performance. These companies surpass each other every few weeks, but none have established a technological lead that no other company can match. This contrasts sharply with platforms like Windows, Google Search, or Instagram—which achieve self-reinforcing market share through network effects, making it difficult for competitors to break their monopolies regardless of their investment of resources.

This situation of technological equality may change due to certain breakthroughs, most notably the realization of continuous learning capabilities, but Evans believes OpenAI cannot currently plan for this. Another potential differentiating factor is the scale effect of proprietary data, including user data or vertical industry data, but existing platform companies also have an advantage in this area.

With model performance converging, competition is shifting towards branding and distribution channels. The rapid market share growth of Gemini and Meta AI exemplifies this trend—to the average user, these products appear largely identical, while Google and Meta possess powerful distribution capabilities. In contrast, Anthropic's Claude model, while frequently ranking highly in benchmarks, suffers from near-zero consumer awareness due to a lack of consumer strategy and product offerings.

Evans draws a parallel between ChatGPT and Netscape, which initially held an advantage in the browser market but was ultimately defeated by Microsoft using its distribution advantages. He believes that chatbots and browsers face the same differentiation challenge: they are essentially just input and output boxes, with extremely limited room for product innovation.

Fragile user base: Scale cannot mask insufficient user stickiness

Despite OpenAI's significant lead with 800-900 million weekly active users, Evans points out that this figure masks a serious user engagement problem. The vast majority of users who are familiar with and know how to use ChatGPT have not made it a daily habit.

Data shows that only 5% of ChatGPT users pay for content, and even among American teenagers, the percentage using it a few times a week or less is far higher than the percentage using it multiple times a day. OpenAI revealed in its "2025 Year-End Summary" event that 80% of users sent fewer than 1,000 messages in 2025, equivalent to less than three notifications per day on average, and even fewer actual chat sessions.

This superficial usage means that most users don't see the differences in personality and focus between the various models, nor do they benefit from features like "memory" designed to build stickiness. Evans emphasizes that the memory feature only brings stickiness, not network effects. Meanwhile, usage data from a larger user base might be an advantage, but when 80% of users use it only a few times a week at most, the extent of this advantage is questionable.

OpenAI itself acknowledges the existence of problems, pointing out a "capability gap" between model capabilities and actual user usage. Evans believes this is an attempt to avoid the fact that product-market fit is unclear. If users can't think of what to do with it in their daily lives, it means it hasn't changed their lives yet.

The company launched the advertising program partly to cover the service costs for over 90% of non-paying users, but more strategically, it allows the company to offer these users the latest, most powerful (and most expensive) models, hoping to deepen user engagement. However, Evans questions whether giving users a better model will change the situation if they can't think of what to do with ChatGPT today or this week.

The platform strategy is questionable: it lacks a true flywheel effect.

Last year, OpenAI CEO Sam Altman attempted to integrate the company's various initiatives into a coherent strategy, presenting a chart and quoting Bill Gates: "A platform is defined as one that creates more value for its partners than for itself." Simultaneously, the CFO released another chart illustrating the "flywheel effect."

Evans believes the flywheel effect is a sophisticated and coherent strategy: capital expenditure itself creates a virtuous cycle, becoming the foundation for building full-stack platform companies. Starting with chips and infrastructure, each layer of the technology stack is built upwards, and the higher you go, the more you help others use your tools to create their own products. Everyone uses your cloud, chips, and models, and then at higher levels, the layers of the technology stack reinforce each other, forming network effects and an ecosystem.

However, Evans stated that he believes this is not a correct analogy. OpenAI does not possess the platform and ecosystem dynamics that Microsoft or Apple once had, and that the flywheel diagram does not actually show the real flywheel effect.

In terms of capital expenditure, the four major cloud computing companies invested approximately $400 billion in infrastructure last year and announced plans to invest at least $650 billion this year. OpenAI claimed a few months ago that it has a future commitment of $1.4 trillion and 30 gigawatts of computing power (without a specific timeline), while actual usage is only 1.9 gigawatts by the end of 2025. Lacking substantial cash flow from its existing business, the company is achieving these goals through financing and utilizing the balance sheets of others (partly involving "revolving revenue").

Evans argues that massive capital investment may only secure a place, not a competitive advantage. He draws a parallel between AI infrastructure costs and the aircraft manufacturing or semiconductor industries: there are no network effects, but each generation of products becomes more difficult and expensive to manufacture, ultimately leaving only a few companies able to sustain the investments required to stay ahead. However, while TSMC holds a de facto monopoly in cutting-edge chips, this hasn't translated into leverage or value capture capabilities in the upstream technology stack.

Evans points out that developers must build apps for Windows because it has almost all users, and users must buy Windows PCs because it has almost all developers—this is a network effect. But if you invent a great new app or product using generative AI, you only need to call the underlying model running in the cloud via API; users don't know or care what model you used.

Lack of product control: Strategy constrained by the laboratory

At the beginning of the article, Evans quotes Fidji Simo, product lead at OpenAI, saying in 2026: "Jakub and Mark set long-term research directions. After months of work, amazing results emerged, and then the researchers would contact me and say, 'I have some really cool stuff. How do you plan to use it in chat? How can it be used in our enterprise products?'"

This statement contrasts sharply with Steve Jobs' famous 1997 quote: "You have to start with the customer experience and work backward to the technology. You can't start with the technology and try to figure out where to sell it."

Evans argues that when you're the product lead in an AI lab, you have little control over your roadmap and very limited ability to set product strategy. You open your email in the morning to find out what the lab has researched, and your job is simply to turn that into a button. Strategy happens elsewhere, but where?

This issue highlights a fundamental challenge facing OpenAI: unlike Google in the 2000s or Apple in the 2010s, OpenAI's bright and ambitious employees don't have a truly effective product that others can't replicate. Evans believes one interpretation of OpenAI's activities over the past 12 months is that Sam Altman is keenly aware of this and is trying to translate the company's valuation into a more sustainable strategic position before the music stops.

For much of last year, OpenAI's answer seemed to be "everything, all at once, immediately." This included application platforms, browsers, social video apps, collaborations with Jony Ive, medical research, advertising, and more. Evans argues that some of this appeared to be an "all-out attack," or simply the result of rapidly hiring large numbers of ambitious people. At times, it gave the impression that people were replicating the form of previously successful platforms without fully understanding their purpose or dynamics.

Evans repeatedly uses terms like platform, ecosystem, leverage, and network effects, but he acknowledges that while these terms are widely used in the tech industry, their meanings are rather vague. He quotes Roger Lovatt, his medieval history professor from his university days: "Power is the ability to get people to do what they don't want to do." That's the real question: Does OpenAI have the ability to get consumers, developers, and businesses to use its systems more, regardless of what the systems actually do? Microsoft, Apple, and Facebook once had that ability, as has Amazon.

Evans believes a good way to interpret Bill Gates's statement is that platforms truly leverage the creativity of the entire tech industry, allowing you to build more things at scale without having to invent everything yourself, all within your system and under your control. The underlying model is indeed a multiplier; a vast amount of new things will be built using them. But is there any reason for everyone to have to use your product, even if competitors have already built the same thing? Is there any reason for your product to always be superior to the competition, regardless of how much money and effort they invest?

Evans concludes that without these advantages, all you have is daily execution. Executing better than everyone else is certainly a wish, and some companies have achieved this over a long period, even convincing themselves they've institutionalized it, but that's not a strategy.