The Maze: OpenAI's IPO story is no longer only about model quality or ChatGPT usage. The new financial snapshot puts the company in infrastructure territory: huge demand, huge losses, huge supplier obligations, and a public-market test that will decide how much capital the AI layer can keep absorbing.
The spending number is the strategy. OpenAI's 2025 bill hit $34bn, with about $19bn tied to research and development and nearly $6bn to sales and marketing. Revenue reached $13bn, but the gap between usage growth and infrastructure cost remains the whole plot. For merchants and software teams building AI shopping assistants, customer-service agents, retail media tools, product-data workflows, or seller copilots, this is not abstract Silicon Valley finance. It is the upstream cost base of the tools now being sold into commerce.
The IPO filing turns compute into a disclosure problem. OpenAI filed confidential IPO paperwork in June 2026, giving it optionality to tap public markets while keeping timing flexible. That matters because frontier AI has started to look less like normal software and more like a funded infrastructure race. Public investors will not only ask how many people use ChatGPT. They will ask how much each unit of demand costs, how much cloud capacity is committed, how defensible the revenue is, and how long private-market patience can subsidize the gap.
The supplier chain is exposed too. The pressure does not stop at OpenAI's cap table. Nvidia, Oracle, CoreWeave, Microsoft, and other compute-linked suppliers have all benefited from the AI capacity buildout. If OpenAI has to cut prices, slow side projects, renegotiate compute, or show a clearer path to operating profit, the ripple moves through GPU orders, data-center financing, enterprise AI pricing, and the application vendors sitting on top. AI commerce operators may buy software from a neat SaaS dashboard, but underneath it sits an expensive hardware and cloud stack.
Governance risk arrives before the roadshow. A 42-state U.S. attorney general coalition opened a broad probe into OpenAI days after the IPO filing became public, focused on areas including advertising, engagement, data handling, minors, seniors, safety policies, and model behavior. That does not prove wrongdoing. It does show the public-company version of AI will carry more compliance cost, more disclosure pressure, and less room for "move fast, explain later."
Why it matters: Ecommerce leaders are being told AI will make search, service, ads, merchandising, and operations cheaper. Maybe. But the supplier economics are not cheap yet. If the model layer depends on permanent capital intensity, then AI vendors will eventually pass the bill downstream through pricing, bundling, usage caps, or tighter enterprise contracts. The useful question is not whether AI works. It is who pays when it does.
Sources: Financial Times | Axios | Business Insider | Tom's Hardware
