Stacklet launches Cloud AI FinOps Benchmark for cloud costs

Stacklet has launched the Cloud AI FinOps Benchmark, which sets tested controls for cloud AI cost governance across Amazon Web Services, Google Cloud, and Microsoft Azure.

The benchmark is designed to help teams assess cloud AI spending and reduce waste through Stacklet's control plane. It covers GPUs, foundation models, custom models, storage, and token-usage thresholds, with controls that can be adjusted to fit internal workflows.

AI cost scrutiny

Cloud AI spending is under growing scrutiny as inference workloads run continuously, development environments stay active, and training jobs and related artefacts accumulate. Token consumption adds another layer of cost management, particularly for teams using managed AI services across multiple cloud providers.

Many organisations can see spending rising but lack a consistent framework to govern it. Stacklet says the new benchmark addresses that gap by defining baseline controls and linking them to remediation steps such as retiring idle endpoints, pausing stalled training jobs, blocking unapproved models, and generating alerts when token usage exceeds a threshold.

The benchmark follows Stacklet research into AI services offered by the three largest cloud providers. The company said its team examined provider services at the application programming interface level to identify where costs emerge, how spending accumulates, and which configurations tend to create waste.

That work was then translated into policy controls. Coverage has expanded across services including AWS Bedrock and SageMaker, Google Vertex AI, and Azure AI, and Stacklet said the benchmark will continue to add controls as cloud providers release new AI services.

Policy controls

One part of the offering centres on pre-built policy packs. These are delivered as ready-to-run sets of tested policies that customers can use without building controls from scratch, while still allowing them to tailor remediation and notification settings to their operating model.

Another focuses on both live environments and development workflows. The controls can be applied to running resources and can also inspect Terraform and other infrastructure-as-code configurations before deployment, giving teams a way to identify waste earlier in the lifecycle.

Travis Stanfield, Chief Executive Officer of Stacklet, said the company developed the benchmark in response to a gap in visibility and action around AI-related cloud costs.

"AI is a real, fast-growing number on the cloud bill now, and most teams are still figuring out how to see it, let alone control it. But watching a dashboard doesn't save anyone money," said Travis Stanfield, Chief Executive Officer, Stacklet.

"You need a standard for what good looks like and a way to act on it, and that's the gap we built the Cloud AI FinOps Benchmark to close," added Stanfield.

Cloud governance

The launch points to a broader shift in cloud management as companies move from tracking conventional infrastructure spending to governing AI-specific usage. Unlike traditional compute and storage, AI workloads can involve specialist hardware, variable token costs, and a mix of experimentation, training, and production activity that does not always fit existing financial controls.

For finance and engineering teams, that creates a governance problem as much as a visibility problem. A benchmark approach gives companies a reference point for acceptable configurations and usage patterns, while policy-based controls allow those standards to be enforced or checked automatically.

Stacklet's product is available immediately, the company said. Users can assess their environments against the benchmark and act on findings through the same control plane used for policy enforcement.

The announcement also included support from an early user. Avalara's cloud and AI optimisation lead said cost control needs to start before AI workloads reach production.

"AI cost doesn't wait for production; it builds up in experimentation and development long before a workload ships. A benchmark that defines what good cloud AI governance looks like, and can act on it to optimize spend, is exactly what teams scaling AI need," said Lindbergh Matillano, Director of Cloud & AI Optimization, Avalara.

"We're looking forward to trying Stacklet's Cloud AI FinOps Benchmark as we continue to grow our AI footprint," added Matillano.