Two competing stories are now running on the same clock. OpenAI and Anthropic are closing in on trillion-dollar valuations as both labs prepare to test public markets. At the same time, enterprise customers are reporting, at scale, that the productivity returns on AI spending have not arrived. Bloomberg Opinion, writing June 4, frames this collision as the next defining test for the AI boom.
The numbers on the supply side are not in dispute. Anthropic closed a late-May round at a $965 billion valuation. OpenAI’s last disclosed round priced at $852 billion. Both companies are building toward IPOs that would price the AI infrastructure buildout as a permanent, compounding feature of the global economy.
The demand side tells a different story. A Bain survey published this week found that roughly 40 percent of enterprise AI spending is not generating a measurable return. Companies are continuing to invest, but the justification is company-specific: a productivity win in one legal department, a cost reduction in one call center. Sector-wide productivity gains, the kind that would justify trillion-dollar platform valuations, have not materialized in the data.
Bloomberg Opinion does not predict a collapse. The piece’s argument is more precise: public markets will be asked to hold two contradictory price signals at once. Investors will need to accept that the supplier valuation reflects expected future gains while the buyer data reflects current non-performance. That gap has to be priced somewhere.
The dotcom parallel is instructive, and Bloomberg draws it. In the late 1990s, markets priced future productivity that took roughly a decade to show up in GDP statistics. Robert Solow’s famous observation, that computers were visible everywhere except in the productivity statistics, applied to a technology that did eventually transform the economy. The question Bloomberg is raising is whether AI’s productivity payoff is delayed or simply distributed unevenly, which is a different problem.
For institutional investors evaluating an OpenAI or Anthropic offering, the ROI debate is not abstract. A trillion-dollar IPO implies that the platform monetizes at scale, which requires enterprise customers to find the spend worthwhile. The Bain data introduces a real pricing risk: if 40 percent of current enterprise AI spend is not generating a return, the path to sustained platform revenue depends on either improving product ROI or replacing the current cohort of buyers with a more productive one.
The labs’ responses over the past 48 hours are consistent with awareness of this problem. Anthropic launched a Partner Network, adding distribution channels to increase the probability that customers land successful deployments. Microsoft framed its AI positioning around intelligence-per-dollar, a signal that the industry has started to compete on efficiency rather than capability alone. Neither move resolves the ROI question, but both reflect that the question is now commercially active.
What the Bloomberg piece adds, beyond the customer-side critique covered yesterday, is the investor-framing layer. The cost debate is no longer a product conversation. It is a public-market conversation. Any analyst building a model for an OpenAI or Anthropic IPO now has to take a position on when enterprise ROI normalizes, at what rate, and for how many customers. That is a harder underwriting question than valuing a platform with a clear revenue growth curve.
Institutional investors planning to participate in either offering should treat the Bain survey data and the Bloomberg analysis as a paired document. The supply-side story and the demand-side story are both true right now, and the public markets will be the venue where they resolve.
Bloomberg Opinion (bloomberg.com), published June 4, 2026.
