AI labs’ push up the stack risks enterprise lock-in and reduced competition, essay argues

The authors argue that current AI business models centered on selling inference are structurally weak because models are undifferentiated, switching costs are low, and prices can be adjusted freely, pushing margins toward marginal cost.

They contend that AI labs’ most plausible path to durable profitability is not through commoditized model layers but by migrating up the stack—embracing vertical integration, embedded enterprise deployments, and deliberate construction of switching costs and other moats.

Drawing on historical analysis of railroads, electricity, telecom, cloud computing, semiconductor manufacturing, and commercial aviation, the authors find that infrastructure providers rarely capture the value they create, with thin margins or outright destruction of capital common in these sectors.

By contrast, enterprise software has sustained gross margins of 75% or more for decades by combining zero marginal cost of reproduction, deep switching costs, and non-ephemeral value that allows fixed buildout costs to be amortized over long periods.

The essay highlights cloud computing and chip fabrication as partial exceptions that have escaped the commodity trap by adopting software-like lock-in mechanisms such as managed-services lock-in, egress fees, and committed-spend agreements.

The authors warn that if AI labs successfully move up the stack, concerns about monopolistic concentration and reduced competition are worth taking seriously now, rather than after lock-in effects materialize.