Steve Eisman flags a concentration risk in AI demand, pointing to OpenAI and Anthropic’s outsized role in big cloud and software revenue

A well-known value investor, Steve Eisman, is warning that the AI economy may be more concentrated than it looks. Speaking on CNBC’s Fast Money on August 13, Eisman argued that OpenAI and Anthropic are capturing the bulk of AI-related revenue at several major companies that are building and selling AI-enabled products, including Microsoft, Amazon, Google, and Oracle.

According to the reporting, Eisman placed OpenAI and Anthropic at roughly 70% of AI-related revenue for those large customers. He further suggested the share could be as high as 25% to 35% of the companies’ total revenue, depending on how the revenue is defined and measured. The core point, as framed in the segment, was not just that AI is driving growth, but that it is being fueled by a narrow set of model providers.

Eisman’s concentration warning matters to the broader market because it links end-demand for AI capabilities to upstream model supply. If the monetization of AI products depends heavily on one or two proprietary model ecosystems, any shift in pricing power, licensing terms, performance, or distribution could ripple quickly across the companies trying to operationalize those models for customers.

Even though the CNBC segment focused on the biggest AI deployment buyers, the concern naturally extends to the technology stack that surrounds them. For example, chip makers that power data centers, including NVIDIA, and application-layer software providers, such as Palantir, sit downstream of the infrastructure buildout. Their revenue trajectories depend not only on whether AI spending rises, but also on how that spending is allocated among model providers, cloud platforms, and enterprise customers.

In practice, major buyers typically aggregate multiple components to ship AI features, including cloud services, inference and training hardware, and developer tools. Concentration risk therefore shows up as a form of dependency: customers may be tied to the performance and business terms of particular model providers even as they diversify hardware and software vendors.

The market implication is that investors may begin to look beyond headline “AI growth” and ask a more specific question: how much of a company’s AI monetization is attributable to a limited set of third-party model providers. That framing can affect expectations for margins, the durability of growth rates, and how quickly buyers can switch models or redistribute demand if a dominant provider stumbles or changes commercial terms.

Still, important specifics are not disclosed in the available account of Eisman’s remarks. The reporting does not provide the underlying methodology for the revenue shares, nor does it break down whether the figures refer to direct licensing, platform usage, or the portion of AI revenue that is causally linked to OpenAI and Anthropic outputs. Without those details, it is not possible to verify exactly how stable or comparable the percentages are across firms and time.

For now, the takeaway is that AI is consolidating around key model ecosystems, even as the surrounding hardware and enterprise software markets remain highly competitive. What to watch next is whether companies that monetize AI features start emphasizing model diversification in their disclosures, whether their customers demand more portability across model vendors, and whether alternative models gain measurable share in enterprise workloads. Any shift in those indicators would directly address the dependency concern Eisman highlighted.