Up the Stack: How AI’s Escape From the Commodity Trap Risks Enterprise Lock-in
AI Snake Oil · Arvind Narayanan · 2026-07-09
Princeton researchers Arvind Narayanan and Akash Kapur argue that frontier AI model inference is structurally exposed to a commodity trap driven by Bertrand competition, and that labs are migrating up the product stack into enterprise software and digital workers to build durable moats, raising concerns about customer lock-in and reduced competition.
Extraction
Topics: ai-business-modelsai-industry-economicsenterprise-aimarket-competitionai-policy
Claims
- Frontier model inference meets the near-ideal conditions for Bertrand competition—undifferentiated products, similar capital costs, low switching costs, and free price adjustment—making durable high margins at the model layer structurally implausible.
- Historical capital-intensive infrastructure industries (railroads, telecoms, airlines) consistently failed to capture the value they created, while enterprise software maintained gross margins above 75% through switching costs and zero marginal cost of reproduction.
- AI labs' most likely path to durable profitability runs through vertical integration, embedded enterprise deployments, and deliberate switching cost construction rather than through model inference revenue.
- Labs are already migrating up the stack with products like ChatGPT and Claude Code, enterprise knowledge-base features, and early digital worker products, with the total addressable market at the top layer being the economy's entire labor spend.
- Moving up the stack raises serious concerns about customer lock-in, reduced competition, and concentration of economic and political power that antitrust regulators should address before moats harden.
Key quotes
The labs' most likely path to durable profitability runs not through the foundation layers (chips, datacenters, models) that have thus far accounted for the bulk of investments, but higher up the stack, through a mix of vertical integration, embedded enterprise deployments, and the deliberate construction of switching costs and other 'moats'.
A digital worker that is embedded with every team in a company quickly becomes a store of tacit knowledge for the whole organization and becomes essential to workflows and business processes across different teams. Unless portability is explicitly built in, it becomes a digital employee that effectively cannot be fired.
During the telecom and fiber buildout of the late 1990s, capacity exploded 186,000-fold in seven years, prices crashed, and roughly $2 trillion in market capitalization was erased. The value generated by the infrastructure primarily accrued to industries and applications built on top of it.