The End of the Foundation Model Era: Open-Weight Models, Sovereign AI, and Inference as Infrastructure

A new academic paper by Jared James Grogan posits a fundamental structural shift in artificial intelligence, declaring the era of foundation models—characterized by massive, proprietary pre-training runs from 2020 to 2025—to be definitively over. The author argues that the competitive moat of large-scale pre-training has collapsed due to open-source models achieving frontier performance and inference costs plummeting toward zero. This technical shift is compounded by a major political catalyst: the U.S. government's formal designation of Anthropic as a supply chain risk in February 2026, which accelerated an existing transition toward national sovereignty in AI.

Grogan's analysis identifies four simultaneous, interconnected disruptions reshaping the industry. Economically, the circular financing that inflated foundation model company valuations is collapsing. Technically, the paradigm is shifting from pre-training at scale to post-training optimization and the composition of AI agents. Commercially, power is moving from foundation model providers to application-layer integrators who treat models as a commodity. Politically, governments are reasserting their historic role as gatekeepers of strategic technology. The paper's counterintuitive conclusion is that open-weight models—not closed ones—are becoming the primary instrument for sovereign control, allowing governments to command AI capability without vendor dependence.