Carbon-aware application management offers promising opportunities to reduce emissions by shifting workloads across time or geographical locations where cleaner energy is available. However, existing solutions often assume access to globally distributed infrastructures or focus on workloads that can tolerate execution delays. These assumptions limit their applicability to organizations with latency-critical applications, limited geographical reach, and financial constraints, making it challenging to expand to greener regions. Yet, sustainable operations are becoming an increasingly important priority for organizations of all sizes. This raises a fundamental question: How can we make sustainable computing practical and accessible under such constraints? 

In this talk, I will demonstrate how intelligent, carbon-aware cloud orchestration can enable sustainable computing without sacrificing performance or increasing costs. Through an in-depth characterization of cloud application behavior, I will highlight overlooked opportunities for reducing carbon emissions and operational costs while preserving application performance. Building on these insights, I will introduce “Aceso”, a system that dynamically distributes application components across geographical regions, adapting to changes in carbon intensity and user traffic. Finally, I will touch upon broader challenges in sustainable cloud resource management, particularly those associated with ML-driven solutions.