Flexera Data: 54% of Enterprise Workloads Now in Public Cloud, GenAI at 72% Adoption

Public cloud crosses the 50% workload threshold

Flexera's 2026 State of the Cloud report, published in the past week, shows public cloud now runs 54% of enterprise workloads, up from 52% a year ago. SMBs moved faster, jumping from 55% to 63% of workloads in public clouds year over year. The shift matters because once more than half your infrastructure sits in metered, consumption-based environments, cost control moves from optional to structural.

The data also confirms AWS and Azure remain neck and neck in market share, which creates meaningful procurement leverage. A credible multi-cloud or dual-vendor posture—backed by actual workload distribution—can extract better committed-spend discounts or custom pricing. Google Cloud and others remain secondary options for most large buyers.

GenAI adoption at 72% drives new cost pressure

Flexera found 72% of organizations already use GenAI extensively or sparingly, with another 26% experimenting. That adoption rate, combined with MILL5's separate research showing enterprise AI budgets projected to reach $85,521 per month in 2025—a 36% increase from 2024—means AI infrastructure moves from pilot line item to material OPEX.

The implication: CFOs should now treat a $1 million-plus annual AI infrastructure run-rate as normal for large programs. That threshold changes when you introduce dedicated FinOps for AI, when you move from per-team to shared GPU infrastructure, and when you require vendors to demonstrate GPU cost visibility in RFPs.

Concrete savings benchmarks for AI infrastructure

MILL5's "Hidden Cost of AI" framework provides rare, quantified savings ranges enterprises can use to evaluate vendor claims. The firm outlines quick wins achievable in weeks 5-8 of an optimization program: shutting down idle development resources yields 15-25% savings on those workloads, implementing auto-scaling for inference workloads yields 20-40% savings, right-sizing over-provisioned instances yields 25-35% savings, and moving appropriate training workloads to spot instances yields 60-90% savings on those workloads. Overall target: 30-40% reduction in AI infrastructure costs without impacting performance.

These numbers matter because they create a benchmark. If a cloud-optimization vendor or platform cannot approach 20-40% savings via basic autoscaling and right-sizing, that is a red flag. Enterprises can now push vendors to show how their tooling achieves these ranges rather than accepting vague efficiency claims.

FinOps adoption accelerates as workload growth continues

Flexera notes more enterprises are adopting FinOps practices to manage costs as cloud growth continues. The combination of 54% of workloads in public cloud, 72% GenAI adoption, and projected 36% year-over-year AI budget growth means cost optimization becomes a first-class requirement, not a side project.

RFPs will increasingly require native and third-party FinOps integration, granular cost allocation and showback/chargeback features, and proven support for GPU and AI cost visibility. The market for third-party optimization vendors—Apptio Cloudability, CloudHealth, CloudZero, CAST AI, ProsperOps, Zesty, Run:ai, Granulate, Turbonomic, Kubecost—grows more competitive as the addressable market expands with cloud and GenAI workloads.

What to watch

The 54% public cloud workload penetration and 72% GenAI adoption rates establish a new baseline for 2026 IT cost-optimization planning. These are no longer early-adopter concerns. Watch for vendors to respond with more transparent GPU cost controls and published savings benchmarks, and for enterprises to use the MILL5 ranges as a negotiating tool in cloud and AI infrastructure contracts. The AWS-Azure near-parity also creates an opening for buyers to play vendors against each other more effectively than when AWS held dominant share.