Sequoia Capital's annual State of AI report, released on Wednesday morning, advances a thesis that will prove uncomfortable for some of the most expensively valued companies in the artificial intelligence sector: that foundation model development has ceased to be a sustainable source of durable competitive advantage and has become, instead, a category of infrastructure — essential, widely available, and increasingly commoditised.
The argument is developed with the rigour one expects from the firm that has been making this bet, in various forms, for three years. Foundation model training, the report notes, has become progressively more capital-intensive whilst simultaneously becoming more widely replicable. The gap between the leading proprietary models and the best open-weight alternatives — which was once several generations of capability — has narrowed to, in many task categories, a degree of parity that makes the proprietary premium difficult to justify for cost-sensitive enterprise buyers.
The value migration, Sequoia argues, has moved decisively to three areas. The first is orchestration: the infrastructure layer that connects foundation models to enterprise data, tools, and workflows. The companies building this infrastructure — agent frameworks, evaluation platforms, fine-tuning pipelines — are accruing switching costs that model providers alone cannot command. The second is vertical application: the businesses that have combined AI capability with deep domain expertise to build products that would require years of specialist knowledge to replicate. The third, and the one Sequoia is perhaps most emphatic about, is the enterprise itself: the large organisations that successfully restructure their operations around AI are accruing compounding advantages in cost structure, speed, and customer experience that constitute a new form of competitive moat.
The report's reception in the venture community has been mixed, largely along lines predictable from portfolio composition. Firms with substantial foundation model positions have found reasons to dispute the commoditisation thesis. Those with primarily application-layer portfolios have endorsed it with perhaps excessive enthusiasm.