OpenAI IPO Path: Costs, Competition, and Token Scrutiny

The IPO momentum around OpenAI is running into a sharper question: whether the economics of frontier AI are compatible with public-market expectations at all. A Financial Times analysis published today frames the core tension directly: the costs of staying at the frontier are punishing, and the penalties for falling behind may be worse.

Key points

• The FT argues that both OpenAI and Anthropic face structural obstacles to a successful float, centered on the relentless capital demands of frontier model development.
• A separate report notes that OpenAI and Anthropic are accelerating toward IPOs even as enterprise customers grow skeptical of pay-per-use token pricing models, with some shifting to cheaper alternatives.
• Chinese AI startup Z.ai is being positioned as a direct challenger to both OpenAI and Anthropic following the release of its flagship GLM model.
• On the commercial side, JD Sports Fashion has gone live as the first enterprise retailer to let customers purchase directly through ChatGPT, alongside Microsoft Copilot and Google Gemini, via a new integration with commercetools and Stripe.
• Sam Altman has reportedly called any IPO valuation below $1 trillion a “nonstarter,” according to aggregated coverage of the FT piece, though the FT’s own framing suggests public markets may view that figure skeptically.

What does the FT’s skepticism actually rest on?

The FT piece (paywalled) does not provide granular financials in its summary, so the precise arguments require a subscription to verify in full. What the summary does make clear is the dual-pressure problem: OpenAI must keep spending at scale to remain competitive, and any slowdown in that spending risks ceding ground to rivals. That is not a comfortable dynamic to present to public shareholders who expect a path to margin expansion.

The token pricing concern adds another layer. Enterprise customers are reportedly questioning the cost burden of pay-per-use consumption and exploring cheaper model alternatives. If OpenAI’s revenue growth depends on token volume from large enterprise accounts, and those accounts are actively shopping for substitutes, that complicates the revenue quality story that underwriters would need to tell.

None of this is new as a theoretical concern. What is notable is that the FT is running this analysis now, as OpenAI’s confidential IPO filing has been widely reported as imminent.

Is Z.ai a credible threat, or just timing?

Z.ai’s emergence as a named OpenAI competitor is worth tracking for investors, even if the Times of India summary is thin on specifics beyond the GLM flagship model. The pattern is consistent with what has played out before: a Chinese lab releases a model that benchmarks competitively, Western frontier labs face a fresh round of questions about pricing power and moat depth.

For OpenAI’s IPO narrative, the competitive framing matters more than any single model release. Underwriters pricing a $1 trillion valuation need to argue that OpenAI holds a durable position. Each credible new entrant, whether from China or elsewhere, is a data point that works against that argument in public markets, even if it does not immediately shift enterprise contracts.

ChatGPT as a commerce channel: a concrete revenue signal

Away from the IPO debate, the JD Sports Fashion deployment offers a tangible example of how OpenAI’s platform is monetizing beyond direct subscriptions. JD Sports customers in the US can now complete purchases inside ChatGPT, with Stripe handling payments and commercetools providing the commerce infrastructure. JD is the first enterprise retailer to use Stripe’s Agentic Commerce Suite in this configuration.

This matters for the IPO story because it points toward a commerce-layer revenue stream that is structurally different from token consumption. If OpenAI can sit inside purchase flows, rather than just answering questions, the monetization model diversifies. Carrefour separately put its entire French grocery catalogue inside ChatGPT in March, giving 26 million ChatGPT users access to basket-building and delivery slot booking without leaving the chat interface.

These are still early-stage deployments. But they are live, named, and verifiable, which is more than can be said for many of the projections embedded in a $1 trillion valuation thesis. Whether they scale fast enough to satisfy public-market investors before the IPO window opens is the question that today’s coverage, taken together, does not yet answer.

This update is published by an independent site tracking OpenAI. Nothing here is investment advice.