When Running Your Own AI Models Beats Paying for APIs - Steven Latré, imec
High token consumption is driving enterprises from cloud APIs to self-hosted open models.
“Model routing in this sense is perhaps just the first low-hanging fruit in achieving good optimization of selecting the right model for the right task.”
Steven Latré of imec describes a three-stage enterprise AI maturity curve: starting with closed APIs, moving to cost optimization via model routing, and now evaluating self-hosted open-scale models with owned infrastructure. The talk frames model routing as only the first step, with full stack ownership as the logical endpoint for high-consumption organizations. This practitioner account from a non-typical AI player (semiconductor research) validates a broader industry shift away from API dependency as token costs scale.