Beneath the Surface Glory: OpenAI’s “Four Major Dilemmas”
TechFlow Selected TechFlow Selected
Beneath the Surface Glory: OpenAI’s “Four Major Dilemmas”
Benedict Evans stated that OpenAI’s long-term competitiveness is threatened by a lack of technological moats, insufficient user stickiness, a platform strategy lacking flywheel effects, and a product strategy constrained by lab-driven R&D directions.
Navigating Web3 tides with focused insightsBy Zhao Ying
Source: WallStreetCN
Benedict Evans, former partner at a16z and renowned tech analyst, recently published an in-depth analysis directly identifying four fundamental strategic challenges facing OpenAI beneath its surface-level success. He argues that despite OpenAI’s massive user base and ample capital, the company lacks a technological moat, suffers from low user stickiness, faces rapid追赶 by competitors, and has its product strategy constrained by lab-driven R&D priorities—all of which threaten its long-term competitiveness.
Evans points out that OpenAI’s current business model does not offer clear competitive advantages. The company possesses neither unique technology nor network effects; only 5% of its 900 million weekly active users pay, and 80% sent fewer than 1,000 messages in 2025—equivalent to fewer than three prompts per day on average. This “one mile wide, one inch deep” usage pattern indicates ChatGPT has yet to become a daily habit for most users.
Meanwhile, tech giants like Google and Meta have already caught up technologically and are leveraging their distribution advantages to capture market share. Evans believes the real value in AI will come from new experiences and use cases yet to be invented—and OpenAI cannot single-handedly create all of them. This forces the company to fight on multiple fronts, building across the stack from infrastructure to applications.
Evans’ analysis reveals a core contradiction: OpenAI attempts to erect competitive barriers through massive capital investment and a full-stack platform strategy, yet without network effects or user lock-in mechanisms, it remains uncertain whether this approach can succeed. For investors, this implies the need to reassess OpenAI’s long-term value proposition and its true standing within the AI competitive landscape.
Erosion of Technical Advantage: Accelerating Model Homogenization
Evans notes that approximately six organizations today can release frontier models with broadly comparable performance. Each firm overtakes the others every few weeks—but none has established a technical lead that rivals cannot match. This contrasts sharply with platforms like Windows, Google Search, or Instagram, whose network effects enabled self-reinforcing market dominance, making it nearly impossible for competitors to break through regardless of how much money or effort they invested.
This state of technical parity could shift due to breakthroughs—most notably the realization of continual learning—but Evans argues OpenAI cannot currently plan for such developments. Another potential differentiator is the scale effect of proprietary data—including user or vertical-specific data—but incumbent platform companies hold similar advantages here.
As model performance converges, competition is shifting toward branding and distribution channels. The rapid market-share gains of Gemini and Meta AI confirm this trend: to ordinary users, these products appear largely indistinguishable, while Google and Meta possess formidable distribution power. In contrast, Anthropic’s Claude model frequently tops benchmark tests but suffers near-zero consumer awareness due to its lack of consumer-facing strategy and product execution.
Evans draws a parallel between ChatGPT and Netscape, which held early dominance in browsers but was ultimately defeated by Microsoft’s distribution advantage. He contends chatbots face the same differentiation challenge as browsers: fundamentally, they consist of little more than an input box and an output box, leaving minimal room for product innovation.
Fragile User Base: Scale Masks Low Stickiness
Although OpenAI enjoys a clear lead with 800–900 million weekly active users, Evans stresses this figure obscures serious engagement issues. The vast majority of users who know and understand how to use ChatGPT have not adopted it as a daily habit.
Data shows only 5% of ChatGPT users pay, and even among U.S. teenagers, those using the service a few times per week—or less—far outnumber those using it multiple times per day. At OpenAI’s “2025 Year-in-Review” event, the company revealed that 80% of users sent fewer than 1,000 messages in 2025—roughly fewer than three prompts per day on average, with actual conversational interactions even fewer.
This shallow usage means most users fail to perceive differences in personality or focus across models and cannot benefit from stickiness-oriented features like “memory.” Evans emphasizes memory may foster stickiness but does not generate network effects. Meanwhile, usage data from a larger user base might confer an advantage—but when 80% of users engage only a few times per week, the magnitude of that advantage is questionable.
OpenAI itself acknowledges this problem, citing a “capability gap” between model capabilities and actual user adoption. Evans interprets this as sidestepping the reality of unclear product-market fit: if users cannot think of what to do with it on an ordinary day, it hasn’t yet changed their lives.
The company launched advertising initiatives partly to cover service costs for its over-90% non-paying users—but more strategically, to expose these users to its latest, most powerful (and expensive) models in hopes of deepening engagement. Yet Evans questions whether giving users better models will help if they cannot conceive of a use case for ChatGPT today—or even this week.
Questionable Platform Strategy: Absence of a Genuine Flywheel
Last year, OpenAI CEO Sam Altman attempted to unify the company’s various initiatives into a coherent strategy, presenting a chart and quoting Bill Gates: “A platform is defined as creating more value for others than it creates for itself.” Simultaneously, the CFO unveiled another chart illustrating a “flywheel effect.”
Evans describes the flywheel effect as an elegant, cohesive strategy: capital expenditure itself forms a virtuous cycle, serving as the foundation for building a full-stack platform company. Starting with chips and infrastructure, each successive layer of the tech stack is built upward—increasingly enabling others to build their own products using your tools. Everyone uses your cloud, chips, and models; then, at higher layers, the stack components reinforce each other, generating network effects and ecosystems.
Yet Evans bluntly states this analogy is incorrect: OpenAI lacks the platform and ecosystem dynamics once possessed by Microsoft or Apple, and the so-called flywheel chart does not depict a genuine flywheel effect.
On capital expenditure, the four major cloud providers spent roughly $400 billion on infrastructure last year and announced plans to invest at least $650 billion this year. OpenAI claimed months ago a commitment to $1.4 trillion in compute capacity and 30 gigawatts—though no timeline was specified—while its actual utilization stood at just 1.9 gigawatts by end-2025. Lacking large-scale cash flow from existing businesses, the company relies on fundraising and leveraging others’ balance sheets (some involving “circular revenue”) to achieve these goals.
Evans argues massive capex may merely buy a seat at the table—not a competitive advantage. He compares AI infrastructure costs to aircraft manufacturing or semiconductor fabrication: no network effects exist, yet each generation’s process becomes harder and costlier, eventually limiting participation to only a few firms capable of sustaining frontier-level investment. Yet TSMC, despite its de facto monopoly in cutting-edge chips, derives no leverage or value-capture ability upstream in the tech stack.
Developers must build apps for Windows because it commands nearly all users; users must buy Windows PCs because it hosts nearly all developers—that’s network effects. But if you invent a brilliant new application or product using generative AI, you simply call foundational models hosted in the cloud via APIs—users neither know nor care which model you used.
Loss of Product Leadership: Strategy Subordinated to the Lab
Evans opens this section with a 2026 quote from OpenAI’s Head of Product, Fidji Simo: “Jakub and Mark set the long-term research direction. After months of work, something amazing emerges—and researchers contact me saying, ‘I’ve got something cool. How will you use it in chat? How will it work in our enterprise product?’”
This stands in stark contrast to Steve Jobs’ 1997 dictum: “You have to start with the customer experience and work backwards to the technology. You can’t start with the technology and try to figure out where to sell it.”
Evans contends that when you’re the product leader of an AI lab, you cannot control your own roadmap—you have extremely limited ability to set product strategy. You open your email each morning to discover what the lab produced overnight, and your job is to turn it into a button. Strategy happens elsewhere—but where?
This highlights OpenAI’s fundamental challenge: unlike Google in the 2000s or Apple in the 2010s, OpenAI’s smart and ambitious employees do not collectively possess a truly effective product—one that others cannot replicate. Evans suggests one interpretation of OpenAI’s activities over the past 12 months is that Sam Altman deeply recognizes this reality and is racing to convert the company’s valuation into a more durable strategic position before the music stops.
For much of last year, OpenAI’s answer appeared to be “everything, simultaneously, immediately”: application platforms, browsers, social video apps, collaboration with Jony Ive, medical research, advertising, and more. Evans views some of these initiatives as resembling “full-spectrum attacks,” or simply the result of rapidly hiring large numbers of aggressive talent. Sometimes, it feels as though teams are replicating the form of previously successful platforms without fully grasping their purpose or underlying dynamics.
Evans repeatedly invokes terms like platform, ecosystem, leverage, and network effects—but admits these concepts are widely used yet notoriously vague in the tech industry. He quotes his college medieval history professor, Roger Lovatt: “Power is the ability to make people do things they don’t want to do.” That is the real question: Does OpenAI possess the ability to compel consumers, developers, and enterprises to use its systems more—even if those systems themselves deliver no inherent advantage? Microsoft, Apple, Facebook, and Amazon once did.
A useful way to interpret Bill Gates’ statement, Evans suggests, is that a true platform harnesses the entire tech industry’s creativity—so you don’t need to invent everything yourself, enabling you to build more things at scale, all within your system and under your control. Foundation models are indeed multipliers: vast amounts of new innovation will be built atop them. But do you have a reason to compel everyone to use your product—even when competitors have built identical ones? Do you have a reason your product will remain superior regardless of how much capital and effort rivals deploy?
Evans concludes that without such advantages, all you possess is daily execution. Executing better than everyone else is certainly desirable—and some companies have done so for extended periods, even convincing themselves they’ve institutionalized it—but execution alone is not a strategy.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
