Enterprise AI: The Margin Thesis Nobody Is Underwriting
Deployment spending is rising faster than measurable productivity, but the distribution of returns is extremely narrow. Nine percent of enterprise programs account for the majority of realized value.
Boards are approving AI budgets at a pace that outstrips their ability to measure returns. That is not irrational — it is optionality purchasing — but it obscures a sharp bimodal outcome distribution.
Our analysis of deployment programs across sectors finds that returns cluster around a narrow set of characteristics: high transaction volume, low process variance, and clear ground truth for evaluation.
The strategic implication is uncomfortable for software vendors. As inference costs collapse, defensibility migrates from the model layer to the integration and data-governance layer, where switching costs are structural.
We expect the next twelve months to reward operators who treat AI as a process-redesign program with a technology component, rather than a technology program with a process component.
In practice that means the integration layer — data plumbing, evaluation harnesses, workflow redesign — is where the return is won or lost, long before a model is chosen.
- Value concentrates in workflows with high volume and low variance, not in knowledge work generally.
- Model cost curves fall faster than integration cost curves — integration is the moat.
- Firms that restructured process before deploying capture 3.4x the return.
- Expect a consolidation wave in vertical AI applications within four quarters.
The Autonomous Freight Inflection Has Already Happened
Driver-out operations on southern US corridors crossed commercial viability in Q2. The constraint is no longer autonomy — it is terminal capacity, insurance structure and freight contract design.
The Quiet Rewrite of Banking Infrastructure
Settlement layers are being replaced without a headline event. The competitive consequence is a re-rating of deposit franchises and a repricing of correspondent banking relationships.
Smart Cities and the Next Capital Cycle
Municipal technology spending is shifting from surveillance and sensing toward energy and mobility systems, which changes which developers and vendors benefit.