Who Captures the Value When AI Becomes Ubiquitous: What Enterprise Leaders Need to Model Now

Deploying AI and capturing value from AI are not the same activity, and confusing the two is one of the more expensive strategic errors an enterprise can make right now. The history of general-purpose technologies suggests that when a capability becomes widely available, productivity gains tend to accrue to whoever controls the bottleneck, not necessarily whoever deploys the technology first. Enterprise leaders committing further capital to AI should be modelling where that bottleneck sits before assuming the returns flow back to them.

The Rent-Seeking Problem Most AI Strategies Ignore

Economic rent is the return that accrues to a resource or position that cannot be easily replicated. In infrastructure-heavy technology cycles, rent tends to concentrate at the layer that everyone else depends on. Cloud computing is the clearest recent precedent: enterprises captured operational efficiency, but the majority of margin expansion went to the hyperscalers who owned the compute layer beneath them.

AI is following a structurally similar pattern. Foundation model providers, GPU manufacturers, and the handful of companies that own proprietary training data at scale are all positioned to extract rent from the enterprises deploying on top of them. Efficiency gains from AI adoption at the application layer are real, but they are partially offset by the cost of accessing the infrastructure that makes those gains possible.

The strategic implication is that an enterprise's net value capture from AI depends not just on what productivity it extracts, but on how much of that productivity it pays back to the layers below. Most AI ROI models we see do not account for this dynamic explicitly.

Why Adoption Measurement Understates the Problem

The standard way enterprises measure AI adoption is through deployment metrics: number of tools rolled out, percentage of employees with access, volume of API calls. These metrics tell you about activity, not about whether economic value is being retained.

A more useful measurement frame distinguishes between productivity gains that reduce input costs, productivity gains that improve output quality in ways customers will pay for, and productivity gains that competitors are capturing at the same rate. Only the second category is likely to create durable competitive advantage. The first and third tend to become table stakes, with the margin benefit competed away or absorbed by input price increases over time.

This gap between adoption measurement and value capture measurement is significant because it shapes capital allocation decisions. Enterprises that optimise for deployment velocity without modelling the retention question are likely to over-invest in AI tooling relative to the returns they will actually see on their income statement.

Where Incumbents Have a Structural Advantage

Incumbents with proprietary data, established customer relationships, and existing distribution have a genuine structural advantage in the current AI cycle. The advantage is not that they can build better models. It is that they can apply AI to problems where their proprietary context is the differentiating input, and where switching costs protect the resulting value from being competed away immediately.

A financial institution applying AI to its own historical transaction data to improve credit decisioning is doing something a new entrant cannot easily replicate. The model architecture is commoditising quickly, but the data and the regulatory relationship are not. This is where incumbents should be concentrating their AI investment, not in deploying generic tooling that every competitor can access on the same terms.

The risk for incumbents is complacency. Proprietary data and distribution protect value today, but they do not protect it indefinitely if a new entrant builds a sufficiently differentiated data asset through a different route. Incumbents need to be actively compounding their data advantages, not simply assuming they persist.

The Infrastructure and Policy Levers That Redirect Returns

Two external forces will materially affect where AI returns flow over the next several years, and neither is fully within an enterprise's control. The first is infrastructure pricing. As open-weight models improve and self-hosting becomes more viable for a wider range of use cases, the rent extracted by API-based model providers faces genuine competitive pressure. Enterprises that have retained the engineering capability to evaluate and deploy open-weight alternatives are better positioned to benefit from that pressure than those that have standardised entirely on managed API access.

Companion piece to our broader work on enterprise AI cost strategy. See The Token Bill Is the New Server Bill for a detailed analysis of open-weight model economics and self-hosting trade-offs.

The second force is regulatory structure. Jurisdictions that impose data localisation requirements, model audit obligations, or sector-specific AI certification standards are effectively raising the cost of participation for external providers while creating defensible positions for local incumbents who can meet those requirements. Enterprises operating in regulated sectors should be treating regulatory compliance capability as a value capture asset, not just a compliance cost.

What to Model Before Committing Further Capital

Before increasing AI investment, enterprise leaders should be able to answer three questions with specificity. First, at which layer of the AI stack does your organisation sit, and what is your negotiating position relative to the layers above and below you? Second, which of your current AI productivity gains are genuinely proprietary, and which are available to your competitors on identical terms? Third, what is your exposure to input cost inflation from model providers, and what is your realistic optionality to switch?

These questions are not abstract. They determine whether your AI investment compounds into a durable cost or capability advantage, or whether it funds someone else's margin. The enterprises that will capture disproportionate value from AI ubiquity are those that have answered these questions honestly and structured their investments accordingly.