Benedict Evans has written a sharp, uncomfortable analysis of OpenAI's competitive position. And it boils down to one question: what does OpenAI actually have that its competitors don't?
Not the question you'd expect for a company recently valued at over $150 billion. But Evans makes a case that's hard to ignore.
No unique technology
Start with the technology itself. OpenAI doesn't have a structural advantage in AI research. The best researchers move between labs. The compute clusters are available to anyone with enough capital. The architectures are published or reverse-engineered within months.
Google, Meta, Anthropic, and a dozen well-funded startups are all building models that rival or exceed GPT-4 in specific domains. Some are smaller, faster, cheaper. Some are better at reasoning or coding. None of this is proprietary in the way that, say, Google's search index was in 2005.
Evans' point? Model quality alone isn't a moat. Not when the gap between first and second place closes every few months.
Shallow engagement, despite scale
Then there's usage. OpenAI has a massive user base - ChatGPT hit 100 million users faster than any product in history. But how deep is that engagement?
Most people use ChatGPT sporadically. A question here. A draft there. It's not woven into workflows the way Google Search or Microsoft Office is. It's not sticky in the way that a social network is.
Compare this to Microsoft, which is embedding AI directly into Word, Excel, and Outlook - tools people use every single day. Or Google, integrating Gemini into Search, Gmail, and Android. These aren't standalone chatbots. They're embedded into habits.
Evans asks: if OpenAI doesn't own the distribution layer, how does it capture long-term value?
Platform dynamics - or ecosystem chaos?
Here's the deeper question. Will AI follow platform dynamics - where one or two generalist models dominate - or will it fragment into a diverse ecosystem of specialised models?
If it's the former, OpenAI is in a fight with Google and Microsoft - both of whom have distribution, capital, and compute at scale. If it's the latter, OpenAI becomes one option among hundreds, competing on price and performance in a commoditised market.
Evans doesn't predict which way this goes. But he's sceptical that OpenAI has the structural advantages to win either scenario outright.
What about the API business?
OpenAI's API business is substantial - developers building on GPT-4 and now o1. But this is also where commoditisation bites hardest. Developers will use whichever model is cheapest, fastest, or best for their specific use case. Switching costs are low. Model performance is converging.
Evans' point: API revenue doesn't build a moat unless you're the only game in town. And OpenAI isn't.
So what does OpenAI have?
Brand recognition. Early mover advantage. A talented team. A partnership with Microsoft that provides capital and distribution - but also creates dependency.
Evans isn't saying OpenAI is doomed. He's saying it's not obvious how they win long-term against incumbents with deeper pockets, better distribution, and equally capable models.
It's a question worth sitting with. Because if the answer isn't clear for OpenAI - the most visible AI company in the world - it's not clear for anyone.
Read the full analysis at Benedict Evans' blog.