Stacklet Ships Cross-Cloud AI Cost Controls as Enterprise Waste Hits 29%

Stacklet ships automated AI cost governance across three clouds

Stacklet released its Cloud AI FinOps Benchmark on September 24, providing ready-made governance controls for AI spending across AWS, Microsoft Azure, and Google Cloud. The product enforces policies on GPU utilization, model selection, storage allocation, and token consumption, and can automatically retire idle inference endpoints. The release addresses a gap in most FinOps platforms: the ability to measure and constrain AI workloads without building separate controls for each cloud provider.

Stacklet did not disclose pricing, quantified savings, customer counts, or benchmark data. This limits the commercial case to a governance capability rather than a demonstrated cost-reduction tool. Enterprises evaluating the product should treat it as infrastructure for policy enforcement, not a turnkey savings program. The competitive set includes Flexera, Cloudability, CloudHealth, Kion, and native AWS, Azure, and Google Cloud cost-management tools. Stacklet's differentiation is its focus on AI-specific controls rather than general cloud visibility.

Cloud waste reversed five years of progress

Flexera's 2026 State of the Cloud research estimates that 29% of IaaS and PaaS spending is wasted, reversing a five-year downward trend. The same research shows FinOps teams' responsibility for governing IaaS and PaaS usage increased from 38% to 45% year over year, while business-unit participation rose from 20% to 25% and software-asset-management participation from 6% to 15%. The 29% figure is a survey estimate, not a controlled benchmark. Buyers should validate it against their own billing data before setting savings targets or funding new optimization programs.

The rising waste estimate expands the addressable market for FinOps platforms, cloud managed-service providers, and native optimization tools. It also shifts competition from simple cost dashboards toward governance spanning cloud, SaaS, data centers, and AI. For procurement teams, the implication is that rightsizing, commitment management, workload placement, and automated policy enforcement are now fundable categories—if the business case ties directly to measured waste in the enterprise's own environment.

AI spending is pushing FinOps beyond public cloud

The FinOps Foundation reports that 98% of respondents now manage AI spending, compared with 31% in 2024. The same organization says 48% of respondents manage data-center cost and usage within their FinOps practice. This movement favors platforms that normalize costs across public cloud, private infrastructure, SaaS, and on-premises environments, rather than tools limited to AWS, Azure, or Google Cloud billing exports.

Enterprises should evaluate whether a vendor supports common cost-allocation and usage schemas—particularly the FOCUS specification—before signing a platform contract. A public-cloud-only tool may leave material infrastructure spending outside the optimization program. For AI workloads, buyers should require allocation at the level of model, application, team, token, GPU, and environment, rather than relying only on monthly cloud-account totals. They should also test whether optimization policies can prevent cost overruns without disrupting production inference or training workloads.

Hybrid deployment is emerging as a cost-management strategy

Several Indian technology companies reported material cost reductions through hybrid AI deployment. NoBroker cut AI-usage costs by 80% in less than 24 months by combining frontier models with open-weight models across cloud and local infrastructure. Zippee reduced cloud spend per shipment by 30%–35% over the previous year through instance rightsizing, spot capacity, and tighter data-retention policies. MobiKwik reduced AWS costs by 12% using a third-party provider. These figures are company- or provider-reported rather than independently audited, so they should not be treated as universal benchmarks.

The examples strengthen the case for alternatives to uniform public-cloud deployment: open-weight models, local infrastructure, spot instances, and managed FinOps providers compete with simply scaling reserved public-cloud capacity. For enterprises with predictable batch workloads or high-volume inference, the relevant procurement question is no longer only "which cloud offers the lowest unit price?" It is whether workload placement, model selection, retention policy, and interruption tolerance can be optimized together. Spot capacity can lower compute costs but introduces availability and engineering risks. Local infrastructure can improve economics but increases capital, operations, and refresh-cycle commitments.

What to watch

The strongest near-term procurement signal is the movement toward automated, cross-cloud AI cost controls. Stacklet's release is concrete but lacks disclosed commercial or savings data. Flexera and FinOps Foundation figures show why enterprises are likely to fund this category: substantial estimated waste, rapidly expanding AI spend, and FinOps responsibility spreading beyond public-cloud billing. Buyers should ask whether a FinOps platform can measure GPU utilization, model selection, token consumption, and idle inference endpoints across all three major clouds. If the vendor cannot answer with specific policy examples and measurable enforcement mechanisms, the product is a dashboard, not a control system.