Ask HN: If the models are so good, why doesn't OpenAI just start 100 companies?
Points and comments are a snapshot, not live.
If frontier AI models are so capable, why sell tokens instead of building businesses?
The post argues that OpenAI and Anthropic could make more money by founding companies that leverage their own models directly, rather than selling API tokens. With token commodification and declining prices, application-side value may exceed utility-side margins. The author suggests that building products might be more profitable than selling the underlying model access.
What commenters are saying
Top commenters note that OpenAI is already moving into apps and consulting, buying a firm to that end. The gold rush shovel analogy is popular: selling infrastructure is proven revenue, while running many businesses requires diverse human skills. Multiple comments emphasize OpenAI's short runway and need for fast revenue, plus Jevons paradox expectations where cheaper AI drives more usage, not less. Some argue models aren't yet good enough to autonomously run complex businesses.