CloudBolt Adds GPU Cost Optimization to StormForge

The general release lets teams allocate GPU spending to namespaces, workloads and labels that consumed the capacity.

Updated on Nov. 10, 2026 in Data Centers

Isometric editorial illustration of stacked unmarked server trays divided into coloured compartments by a thin vertical bronze divider, evoking allocation of GPU capacity. — CloudBolt Software made GPU cost optimization in StormForge generally available at KubeCon + CloudNativeCon North America 2026, letting teams allocate accelerator spending to namespaces and workloads. AI Illustration. Upload story photo >

CloudBolt Software announced general availability of GPU optimization in StormForge at KubeCon + CloudNativeCon North America 2026, after the feature had been in early access since September. The release allocates GPU costs using figures drawn from the cloud provider's bill rather than list-price estimates.

Why it matters

GPU spending is growing faster than many teams can explain it, making per-workload attribution a practical need for organizations running accelerated computing in Kubernetes. The tool also adds node pool cost and idle reporting, along with reliability recommendations.

GPU cost allocation covers shared and time-sliced GPUs and is sourced from the cloud provider's bill, with node pool views showing cost, idle capacity, Spot usage and commitment coverage. The announcement does not report measured reductions in GPU waste.

The players

CloudBolt Software

CloudBolt Software is a company that develops cloud cost and resource optimization software, with StormForge among its products.

StormForge

StormForge is a Kubernetes resource optimization platform that CloudBolt Software has extended with GPU cost allocation features.

The details

Teams can drill down from each cluster to a single node pool and instance type to see cost, idle capacity, Spot usage and commitment coverage. New reliability recommendations account for CPU contention and read namespace LimitRange policies, and Java memory recommendations were improved to reduce out-of-memory errors.

September: GPU optimization entered early access.
KubeCon + CloudNativeCon North America 2026: CloudBolt announced general availability.

The Tech Race

The release extends Kubernetes cost visibility from CPU and memory into GPUs, which the container scheduler increasingly manages for AI workloads. It positions CloudBolt against other tools competing to attribute accelerator spend inside cloud-native clusters.

Engineering and finance teams running GPU workloads can use namespace and label attribution to tie accelerator spending to specific applications. The release also surfaces idle capacity and commitment coverage that can inform rightsizing decisions.

The takeaway

Tracking GPU spend at the workload level is becoming a basic requirement as accelerator costs climb. Teams that attribute usage by namespace and label can identify waste before it compounds.