The post explores the factors influencing the decision of where to run models—whether on a frontier API or on local hardware—emphasizing the organizational rather than purely engineering considerations such as regulation, procurement, and accountability.