Sedai Launches Autonomous GPU Optimization to Cut AI Costs

Sedai’s breakthrough GPU utilization model sees what other tools miss — and autonomously acts to cut costs across Kubernetes environments without sacrificing performance

Sedai, the self-driving cloud, announced the general availability of GPU Optimization. This new capability helps engineering teams safely reduce AI infrastructure costs, without ever compromising performance or availability. Sedai dramatically reduces the cost of AI infrastructure by identifying unused GPU allocations, right-sizing GPU workload requests, and packing GPU capacity more efficiently across Kubernetes clusters.

The AI boom has made GPU spend one of the fastest-growing and least-controlled line items in enterprise cloud budgets. According to IDC, AI infrastructure spending grew 166% year-over-year in 2025, yet studies show that one-third of all GPUs run at less than 15% utilization. The result is a compounding problem: high costs, hidden waste, and capacity shortages are slowing down AI teams.

“Engineering leaders know that AI infrastructure is expensive, but until now, they didn’t have a way to lower the bill without risking performance,” said Suresh Mathew, CEO of Sedai. “We built Sedai to remove that fear. We model true GPU utilization, identify exactly where the waste is, and execute the fixes safely and autonomously, so teams can optimize with confidence.”

Also Read: AiThority Interview with Glenn Jocher, Founder & CEO, Ultralytics

A Smarter Signal: How Sedai Measures True GPU Utilization

At the core of Sedai GPU Optimization is a proprietary GPU utilization model. It works by inferring a resource’s true GPU usage from multiple telemetry signals — beyond the surface-level metrics that most tools rely on. Standard utilization metrics, such as those reported by NVIDIA System Management Interface (nvidia-smi), measure only whether a GPU is active, not whether it is doing productive work. Sedai’s approach models real utilization at the workload level, enabling accurate identification of waste that would otherwise go undetected.

What Sedai GPU Optimization Does

Sedai GPU Optimization delivers three core capabilities:

• Idle GPU Deallocation: Detects Kubernetes workloads with GPU resources allocated but not actively used, and automatically removes those requests, with clear before-and-after cost projections surfaced directly in the UI.
  
  
• MIG Enablement and Packing: Identifies NVIDIA GPU instances where MIG is not enabled, enables it, and assigns workloads to appropriately sized slices — maximizing the number of workloads that can run efficiently on each physical GPU. MIG optimization is executed via Dynamic Resource Allocation (DRA) integration, providing a standardized, Kubernetes-native approach to GPU resource sharing.
  
  
• GPU Node Pool Optimization: Analyzes how workloads are distributed across GPU devices and recommends repacking them to consolidate onto fewer nodes, freeing entire GPU devices and reducing node spend. Recommendations are surfaced via Datapilot, with clear before-and-after cost projections at the node pool level.

Autonomous Execution, From Day One

Unlike tools that stop at recommendations, Sedai GPU Optimization is built to act. Every optimization is executed with safety checks and guardrails enforced at every step — so teams can move from insight to action without the risk of disrupting production AI workloads. Teams can start with guided recommendations and progress toward full autonomy at their own pace, all within the same trusted Sedai decision engine.

Also Read: ​​The Infrastructure War Behind the AI Boom