Fortune review: OpenAI and Anthropic IPO valuations ignore the ROI gap, Sam Altman admits

The coming mega IPO roadshow is selling a fantasy with a pricing problem. Fortune’s look at OpenAI, Anthropic, and SpaceX points to $852 billion for OpenAI and $965 billion for Anthropic, with both companies reportedly looking to raise $60 billion as they approach public markets. The thesis is blunt: these valuations assume an AI economy that does not exist yet, at least not at the scale and ROI enterprise buyers are currently experiencing.

OpenAI CEO Sam Altman effectively waved a red flag on that gap when he admitted, according to Fortune, that corporate concern about excessive AI costs is “fair criticism.” That matters because it hits the center of the IPO pitch. If the biggest buyers in the lab’s preferred segment are struggling with ROI, the market is paying today for returns that are not landing quickly enough.

So what are these frontier labs optimizing for? Fortune says the race is heavily tilted toward the top 15% of the global AI market. Anthropic is framed as even more optimized for enterprises with fast networks, deep talent, and generous compute budgets, where CEOs and employees can test models and look for productivity wins. This is the segment that produces the most impressive demos for copilots and frontier models, and it is exactly the slice of the world that public-market investors can picture.

But Fortune argues this is also the place where most of the money is not. The reason is not that enterprise AI is useless. It is that corporate buyers are “struggling to find the ROI,” and cheaper open-source alternatives can do “just as well,” at least for many near-term use cases. In other words: the frontier-lab story is built around outcomes that require sustained spending and proprietary advantage. The buyer side is increasingly asking whether it should pay frontier pricing at all.

Fortune then widens the lens and makes a familiar but under-discussed business point: historically, the “real money” shows up where there is unmet need, not where technology demos look coolest. It claims this unmet need exists in both highly digitally evolved economies and in developing economies where AI revenue can scale. On the developed-economy side, it points to modernization in banks, insurers, and ministries. A concrete example in the piece: Fortune notes that 43% of core banking systems and 95% of ATM transactions still run on COBOL, a program that predates the year the Beatles got together.

The market reaction cited in the story reinforces its theme about ROI and compute discipline. Fortune says that after Anthropic argued Claude could automate modernization of that legacy stack, IBM dropped 13.2% in what it calls its worst session since 2000. Whether or not that one move proves the thesis, it illustrates how the market can quickly pressure incumbents when AI promises collide with cost, integration risk, and timelines.

Then Fortune shifts to “Break Out” economies, listing India, Brazil, Indonesia, Kenya, Vietnam, and others, where digital momentum is accelerating fast but credit access often lags. The argued “killer app” is AI for credit scoring trained on payment data, plus identity authentication and fraud detection. Fortune ties this to existing transaction rails already operating at scale, including India’s UPI processing 22.6 billion transactions in March 2026 alone and mobile money moving more than $2 trillion worldwide in 2025. The point for decision-makers is straightforward: if demand is already there, AI can attach to it and monetize sooner, without waiting for a fully formed “global AI economy” of augmented knowledge workers.

Finally, Fortune points to “Watch Out” economies, mostly across Sub-Saharan Africa and South Asia, and claims its research estimates that AI crop-disease detection across just seven African countries could unlock $6.1 billion for 14 million smallholder farmers. In a detail that reads like a gut-check to the IPO chatter, Fortune says those populations report “the highest trust in AI” of any cohort measured anywhere, higher than the Silicon Valley executives whose enthusiasm is priced into these IPOs.

If you zoom out, Fortune frames a history lesson. At the dot-com peak, capital poured into Pets.com and Webvan, but durable internet revenue went to infrastructure providers like Cisco (routers) and Akamai (content delivery), and later to AWS. The mobile era had a similar script: handset makers mostly lost out, while tower companies like American Tower and Crown Castle profited because carriers had to rent the infrastructure no matter which phone won. Fortune argues that the more transformative the technology gets, the more durable value migrates to the layer everyone must pay for over and over.

The piece applies that logic to today’s AI stack. It claims strategic acquirers are already behaving as if they know where the long-term payment will land: it cites IBM buying DataStax, ServiceNow acquiring Data.world, and Salesforce paying $8 billion for Informatica in a depressed 2025 deal market. The subtext is important for boards and CFOs: acquirers are not betting on which model wins. They are buying the pipelines AI models run on, and the infrastructure that customers will pay for repeatedly.

Fortune also adds an “arithmetic” warning about the cost buildout. It cites Bain & Company’s warning that AI will need $2 trillion in annual revenue by 2030 to justify compute spending, describing an $800 billion shortfall. It also notes Oracle disclosed $248 billion in data-center leases running 15 to 19 years, against customer contracts that often run five. And it claims open-weight models are compressing inference prices by an estimated 30% to 50% a year, capping the margins any model layer can defend.

None of this, Fortune says, guarantees the mega IPOs are doomed. It acknowledges possibilities: OpenAI could eventually hit revenue targets it has been missing; Anthropic could win enough enterprise customers; SpaceX’s launch economics might justify its price. But the core contention remains: the race to be first is also a race to sell a story about AI diffusing frictionlessly across the global economy before ROI numbers catch up. For executives watching from adjacent sectors, the takeaway is not “don’t buy AI.” It is that the market might be paying for the narrative before it pays for the unit economics, and the survivable winners could be the companies that monetize utilities, rails, and modernization in places where need is already active, not imaginary.