Abstract: Adaptive experiment techniques, such as Bayesian optimization or active learning approaches, are becoming commonly-used tools across scientific domains and, increasingly, being deployed on HPC. In this presentation, we discuss the tradeoffs in ensuring efficient use of massively-parallel systems for adaptive experimental techniques through the lens of molecular design at scale. Part of the presentation will share our progress in building tools to design new battery electrolytes with by combining machine learning and quantum chemistry. The rest will focus on the general-purpose toolkit, Colmena, we are building to support adaptive experimental approaches on HPC.