OpenAI CEO Sam Altman said at Stripe Sessions that the company is not trying to become a high-margin software business. His description was closer to a utility-scale platform: a large AI infrastructure provider with permanently low margins, charging customers based on usage and supplying what he called usable intelligence.
OpenAI frames its business as core infrastructure
Altman compared OpenAI to an AI version of Stripe, where developers and companies can buy AI capacity as needed and plug it into their own products, workflows, and services. In his framing, the company is not focused on selling a single finished application. It wants to offer a broad capability layer that others can build on.
He said data centers, models, and the harness layer used to schedule and orchestrate AI should be delivered as one integrated system. That is why he described OpenAI less like a traditional product company and more like a provider of something essential, similar to power or water.
On the cost side, Altman said OpenAI has signed 20-year power and land agreements, signaling a long-term commitment to building out supply. The model only works, in his view, if the company is optimized for scale rather than high margins.
Low switching costs make high margins hard to sustain
Altman was direct about the weakness of moats in AI. He said users have recently moved from competing coding tools into Codex, which now has more than 3 million weekly active users. Still, he argued that smarter AI makes switching easier, not harder, because users can simply instruct an agent to move tasks elsewhere.
That point sits at the center of his business argument. If switching costs stay low, then high profitability is not a realistic baseline for the sector. OpenAI’s answer is to accept thin margins and compete on network size, growth, and infrastructure reach.
Altman rejects the idea of owning the whole stack
When asked by Stripe co-founder Collison whether AI companies would move up and absorb the full value chain, Altman said some may try, but OpenAI does not want to. His view is that outside companies should build products on top of OpenAI’s systems, and those customers should benefit as models improve.
That leaves OpenAI in the role of a general-purpose intelligence layer rather than the owner of every application category. The company’s upside, under this approach, comes from broad usage instead of controlling each end product.
Companies built on model gaps may face pressure
Altman also offered a simple test for judging businesses in the AI sector: if a company quietly hopes models stop improving, it probably earns money by covering gaps the models still cannot handle. Once those gaps narrow, that position becomes vulnerable.
At the same time, he cautioned businesses against overstating how much AI will instantly upend existing structures. On agent payments, for example, he said money still has to move through some channel. Companies that make strong products and solve real problems will not disappear just because AI tools spread more widely.
He also pointed to Shopify CEO Toby Lütke as one of the strongest examples of AI adoption he has seen, praising both his personal use of AI and his push to have teams use it across the business rather than turning adoption into a token leaderboard exercise.