OpenAI has no durable competitive moat. Roughly six frontier model labs, including Anthropic, Google, and Meta, ship equivalent-capability models today. They leapfrog each other every few weeks. ChatGPT has a large user base but narrow engagement, no network effects, and no consumer product beyond the model itself that has demonstrated product-market fit. Sam Altman appears to know this, and the last 12 months of OpenAI's moves look like an attempt to trade paper valuation for harder strategic positions before the window closes.
The product strategy problem is structural, not personnel. Fidji Simo, OpenAI's head of Product, described opening her email to discover what the lab built and then figuring out how to ship it as a chat feature. Mike Krieger and Kevin Weil flagged the same dynamic in 2025. When research sets the roadmap and product follows, you cannot execute a Jobs-style customer-backward strategy. Meanwhile, well-capitalized incumbents with existing distribution, Google chief among them, no longer look like they cannot execute on AI. The 'Google can't do AI' thesis is dead.
The deeper question Benedict Evans is working through in the full piece is whether any of OpenAI's moves, on models, compute, distribution, or enterprise, can create a winner-takes-all dynamic before foundation models get commoditized into margin-free infrastructure. The model benchmark chart alone, showing six labs in a dead heat as of January 2026, is worth the read. The analysis of what a viable moat could even look like from here is sharper.
[READ ORIGINAL →]