Idle cloud and GPU resources, termed "zombie workloads," are significantly increasing operational costs and power consumption in data centers, a problem intensified by GPU-intensive AI workloads. These workloads, including abandoned libraries, programs, services, and storage volumes, persistently consume resources in cloud and on-premises environments. IDCA research indicates zombie workloads account for up to 13 per cent of US cloud usage, while FinOps tool providers estimate overall cloud waste can reach 25 to 30 per cent or more. Cloud-native microservices can run unnoticed in the background even after an application stops, and Eric Newcomer of Intellyx noted these "headless" services can compose hundreds of components, making cleanup difficult. Graziano Castro of Akamas and a Cloud Native Computing Foundation Ambassador said GPUs are significantly more expensive than CPU cores, making idle GPUs a major financial drain, adding that in the LLM era "the cost of ignoring inefficiency went up by an order of magnitude almost overnight." Kubernetes is evolving with new primitives for dynamic resource allocation and smarter batch job scheduling, while efficiency monitoring relies on GPU health and utilisation checks, with tools like NVIDIA's Data Center GPU Manager (DCGM) used, though Castro cautioned a GPU can appear busy while waiting for other actions. OpenTelemetry standards work is vital for correlating performance signals, and IDCA's Roger Strukhoff emphasised clear policies and routine environment-wide scans to eliminate zombie workloads, particularly after internal consolidations or acquisitions.