2026 Rate Optimization Insights: Azure Compute

TL;DR

• Taking into account all rate optimization strategies (e.g., custom pricing only, RIs/SPs only, no commitment strategy), all-discounts median ESR rose from 27% to 30% (previous years vs. H1 2026), with more organizations achieving positive ESRs.
• Specifically, combining custom pricing with RIs/SPs yielded a 36.6% median ESR, +17.5 percentage points (pp) compared to using RIs/SPs alone (19.1%).
• The smallest segment with <$3M annualized compute usage had the largest median ESR increase (19.1% → 23.8%); larger segments saw minimal change.
• ESR improvement was attributable to increased commitment adoption. Percent of billing scopes using commitments increased from 76% to 84%; those using both RIs/SPs rose from 36% to 41%.
• RIs dominate. 3-year VM RI adoption rose from 54% to 66%; on average, their commitment portfolio allocation also grew.
• Commitment lock-in risk increased. Median CLR rose from 15.8 to 21 months; its distribution shifted right.
• Coverage increased faster than ESR. Median coverage rose 8.2 pp (48.5% → 56.7%) while median ESR rose only 3 pp (27% → 30%) over the same period.
• The smallest and middle segments drove most of the coverage growth, while coverage for the largest segment slightly decreased.
• The impact of expiring RIs on July 1st 2026 was significant. 88% of billing scopes had spend on retiring RI series, averaging 20% of their total compute usage.

Introduction

Compute optimization delivers outsized impact on Azure costs. Based on analysis from September 2024 to June 2026, compute services, including Virtual Machines (VMs) and Azure Kubernetes Service (AKS), accounted for 36% of capacity-reservable Azure aggregated spend (actual cost paid after discounts have been applied), with VMs as the single largest driver. Storage and databases were the next largest spend areas.

AI spend on Azure grew but remains small. Only 2% of capacity-reservable spend came from Foundry Models (Microsoft’s hub for deploying LLMs such as ChatGPT, Claude, Llama, and DeepSeek). Azure Databricks and Fabric together accounted for 8% of spend.

Establishing rate optimization practices now, before AI-attributable compute represents a meaningfully larger share, is easier than retrofitting them later.

Key Statistics in H1 2026

Methodology

This survey report analyzed Azure compute services (e.g., VMs, AKS, App Service, etc.) eligible for Reservations (RIs) and Savings Plans for Compute (SPs) using anonymized data from Azure Cost and Billing exports.

While we analyzed H1 2026 (January 1, 2026 – June 30, 2026) as the current snapshot, we also examined data from previous years using a 24-month lookback period (January 1, 2024 – December 31, 2025) to ensure statistical reliability and identify meaningful trends. This approach increased sample size, reduced short-term noise, and highlighted directional shifts in rate optimization behavior over time. The dataset covered hundreds of anonymized billing scopes representing over $1 billion in annualized Azure compute usage.

Analysis was limited to data collected before ProsperOps began optimizing commitments to avoid skewing results.

Azure billing data does not make it easy to separate RI/SP savings from custom pricing discounts. The Effective Savings Rates in this report reflected the impact of both, but excluded savings from using Spot.

Billing scopes were segmented into three annualized compute usage bands based on list on-demand rates:

• Less than $3M in annual compute usage (<$250K monthly usage)
• $3M to $12M in annual compute usage ($250K–$1M monthly usage)
• More than $12M in annual compute usage (>$1M monthly usage)

The $3M boundary separated organizations likely without EA/MCA access from those that had negotiated terms. The $12M boundary separated mid-tier spenders from large enterprises with dedicated FinOps programs.

Effective Savings Rate

Effective Savings Rate (ESR) measures the net return on rate optimization efforts as a single, comparable percentage, rather than relying on input metrics like coverage, utilization, or discount rate that do not tell the full story:

ESR can be scoped to measure the impact of any discount type, including custom pricing and Spot VMs. Because Azure’s cost and usage data doesn’t separate custom pricing from commitments, this report calculated all-discounts ESR as total savings from both relative to the on-demand rate, excluding Spot. Custom pricing is also usage-dependent and can materially skew ESR, so we segmented the dataset by annualized compute usage (List On-Demand Rate) to allow for comparison.

This metric has been widely adopted by FinOps Foundation and cloud providers as a core KPI. Because the ProsperOps algorithm is built to maximize ESR, our console tracks it, enabling customers to benchmark against pre-optimization baselines and peers.

What is Commitment Lock-In Risk

ESR alone does not tell the full story of rate optimization performance. RIs and SPs carry inherent lock-in risk: you commit to paying for a resource for 1 or 3 years, and if your usage declines or shifts, you may end up paying for commitments you cannot fully utilize, pushing ESR down and, in extreme cases, into negative territory. 

Commitment Lock-In Risk (CLR) is a companion metric to ESR that quantifies the time dimension risk of committing to a cloud provider in exchange for a discounted rate, expressed in months. Higher CLR means greater risk exposure to usage volatilities.

This risk is especially acute on Azure, where neither RIs nor SPs allow term shortening. A RI exchange merely cancels the remaining term and binds the customer to a new full-term commitment.

How Azure Custom Pricing Works for Compute

Azure compute discounts come from three levers: Reservations, Savings Plans for Compute, and negotiated custom pricing. This section focuses on custom pricing and how it interacts with RIs/SPs. 

Custom Pricing: Enterprise Agreement (EA) and Microsoft Customer Agreement (MCA)

Organizations with significant Azure spend can negotiate reduced rates through an Enterprise Agreement (EA) or Microsoft Customer Agreement (MCA). EA is Microsoft’s traditional volume licensing contract for larger enterprises; MCA is the newer, more flexible standard for direct and partner transactions. Both can include a Microsoft Azure Consumption Commitment (MACC), trading minimum spend for custom pricing that applies to on-demand rates and RIs/SPs.

How RI/SP Discounts and Custom Pricing Interact on Azure

On Azure, unlike AWS, RI and SP discounts are not cleanly separable from custom pricing in billing data. Generally, they are blended into a single effective rate.

Custom pricing is provided via a rate card with negotiated prices for each SKU. Some customers receive standard list RI/SP rates regardless of EA/MCA status; others benefit from both a lower on-demand rate and better-than-list reserved capacity rates. One implication: 3-year RIs sometimes beat custom pricing on compute resources alone, making them attractive even for organizations with EA/MCA discounts.

The ESR in this report reflects an all-discounts view (inclusive of custom pricing and RIs/SPs, but exclusive of Spot VMs), because billing data cannot separate the two effects easily.

---

H1 2026 Observations

Here are several observations from our survey analysis on Azure compute usage and their rate optimization results.

The median all-discounts ESR was 30%, an increase of 3 percentage points from previous years.

Median ESR was 30% in H1 2026, up from 27% in 2024-2025. The most notable change was the minimum ESR increasing from -26.9% to -0.6%, indicating that more organizations generated positive ESRs in H1 2026.

The +3 pp improvement in H1 2026 was driven by the smallest compute usage segment (<$3M annually), where the median ESR rose by 4.7 percentage points (pp). The other larger segments improved marginally, having less room to improve.

ESR increased with annualized compute usage in H1 2026 as shown in the segmented box plots, consistent with the pattern that larger organizations invest more in FinOps resources and tools.

A dual-pronged strategy with both custom pricing and RIs/SPs achieved the highest ESR outcomes.

This yielded a median ESR of 36.6%, which was 17.5 percentage points above using RI/SPs alone without custom pricing (19.1%).

Higher ESR was attributed to greater commitment adoption (84% of billing scopes).

Organizations were less likely to have no commitment strategy; 16% of billing scopes had no commitments in H1 2026, which meant 84% used RIs and/or SPs. 

Organizations using both RIs and SPs grew from 36% to 41%; RI-only edged up from 34% to 36%; SP-only marginally increased from 6% to 7%.

Organizations increased their adoption of RIs and longer-term commitments to achieve greater savings.

3-year and 1-year VM RIs saw the greatest increase in adoption, followed by 3-year SPs. Other shorter-term commitments, 1-year SPs and 1-year App Service RIs, declined in the same period.

3-year RIs and SPs represented a greater share of their overall commitment portfolio in H1 2026.

Average 3-year RI allocation rose from 44% to 49% of total commitment spend; 3-year SP allocation grew from 15% to 20%.

Commitment adoption and longer terms increased CLR.

The CLR distribution shifted right as shown in the histogram, with more organizations having higher commitment risk. The median CLR in months rose from 15.8 to 21.

Coverage also increased in H1 2026, which put organizations at risk of overcommitment.

Median coverage rose from 48.5% to 56.7% (8.2 pp increase). Some organizations were still overcommitted. Analysis of organizations’ usage patterns indicated that overcommitment was more likely due to mismanagement of commitments and higher coverage, rather than sudden declines in usage.

Compute Commitment Coverage Percentiles	2024-2025	H1 2026	Change (pp)
98th	103.8%	102.4%	-1.4
75th	76.5%	77.4%	+0.9
50th	48.5%	56.7%	+8.2
25th	6.1%	22.8%	+16.7
Min	0.0%	0.0%	No change

Both the smallest and middle compute usage segments contributed to the increase. Coverage decreased by 3.7 pp with the largest segment, suggesting that this group may be nearing a coverage ceiling with median coverage already above 70%. 

Further increasing coverage offers diminishing ESR upside while adding overcommitment risk if usage dips. Rather than chase incremental coverage gains, the largest segment took a more conservative stance to protect against usage volatility.

Coverage, like ESR, increased with annualized compute usage in H1 2026. Bigger spenders tend to have greater FinOps maturity and cover highly.

Organizations were still loading up on RIs despite the July 1 deadline.

On July 1, 2026, RI purchases and renewals for several older VM series became unavailable, as shown in the table below. Some non-expiring VMs (Dv3/Dsv3/Ev3/Esv3) also had their RIs discontinued (see Microsoft announcement).

Organizations continued to increase their RI allocation, though, even after the official announcement in March 2026. 88% of billing scopes analyzed in 2026 had retiring RIs, representing 20% of their total compute usage on average. Organizations that delay modernization risk higher costs and ESR degradation.

---

ProsperOps Perspectives

ProsperOps is an intelligent FinOps automation platform that autonomously manages Azure compute commitments to maximize ESR while minimizing commitment lock-in risk.

Reservations provide better ROI, but are hard to manage manually.

RIs deliver substantially higher discounts than SPs; a 1-year RI averages 43% versus 29% for a 1-year SP. Azure workloads typically stay on the same compute type, reducing the need for SPs’ cross-resource flexibility. Since RIs are exchangeable in commitment amount, they provide sufficient financial flexibility for most scenarios.

But Azure RIs are scoped to a specific region, machine series, and resource type. They don’t float like SPs and must be actively managed. There’s no marketplace for offloading unused RIs; exchanges only change the commitment amount while resetting to a full 1- or 3-year term.

These challenges compound across multiple subscriptions, each requiring independent assessment of usage, coverage, and exchange decisions. Without a centralized view, the operational burden grows fast.

Gathering cost and usage data is also difficult on Azure. Data latency means you may be forecasting rather than optimizing against the current state, and delays lead to missed savings.

Most organizations have no formal visibility into their ESR or CLR. The ProsperOps console tracks both automatically.

Batch-purchased RIs yield diminishing ESR returns.

A simple strategy might be buying traditional full-term RIs in large batches to cover usage, but increasing coverage does not guarantee better ESR, which is dependent on multiple factors: usage patterns, overall discount rate, RI portfolio allocation and terms, flexibility, and utilization. 

Batch purchasing commitments to increase coverage generates only marginal proportional ESR gains, as seen in the data. While median coverage increased by 8.2 pp in H1 2026, median ESR improved by 3 pp.

Sophisticated techniques increase savings but require automation.

Most Azure organizations batch-purchase full-term RIs and SPs with limited automation, producing suboptimal outcomes as usage fluctuates. Automation enables vertical flexibility (commitment amounts change with usage) and horizontal flexibility (commitments that shift across resource types and regions).

Adaptive Laddering distributes commitments in small, staggered increments, increasing flexibility and reducing CLR. Executing it manually is error-prone and time-intensive: each resource type-region combination requires data gathering, rung calculations, coordination, and execution continuously.

Manual processes cannot keep pace. ProsperOps applies Adaptive Laddering and other techniques automatically, ingesting hourly data and executing via API in near real-time. As AI workloads grow, optimizing their compute infrastructure becomes critical.

Microsoft RI updates demand immediate action.

The July 1, 2026 deadline for RIs on expiring and several non-expiring VM series has passed. Organizations that have not migrated their workloads to modern VM series now face that transition without the runway to build a replacement commitment strategy and are likely experiencing ESR decline.

ProsperOps is helping customers assess their on-demand exposure and build optimization strategies across three dimensions:

• Commitment eligibility — which RIs and SPs are supported on a given VM series
• Performance fit — matching usage against what the workload actually needs
• Feature continuity — ensuring capabilities on current VMs carry over to those that customers migrate to

Acting now preserves the savings already built into a customer’s commitment portfolio and positions them to capture new rate optimization opportunities as Azure continues to retire older VM series. See how a customer was able to maintain a high all-discounts ESR by switching to a new VM series in this blog post.

---

What I Should Do Based on My ESR

Do you know your Azure Compute ESR? If not, you can calculate your ESR using the FinOps Foundation ESR Playbook or start a free Savings Analysis with ProsperOps.

The scatterplot below shows the full all-discounts ESR distribution against annualized usage from January 1, 2024 to June 30, 2026. Find your segment and ESR to see how you rank.

Smallest Segment (< $3M annualized compute usage)

If you’re not using commitments, you’re leaving savings on the table. A negative ESR means unutilized commitments are costing you more than on-demand.

Advice:

• Identify your top compute resources by spend.
• Purchase RIs on the most consistent-usage resources to capture savings with limited downside risk. 
• Audit for underutilized RIs and exchange them for better-utilized ones. 
• Determine the amount of compute usage that can only be covered by Savings Plans, and maintain at least that Savings Plan coverage amount.

Middle Segment ($3M–$12M annualized compute usage)

You likely have commitments but are leaving meaningful savings on the table. You might not be taking advantage of custom pricing on RIs/SPs or covering conservatively to avoid risk of overcommitment.

Advice:

• If you qualify for EA/MCA, prioritize RIs/SPs where custom pricing delivers greater discounts than on-demand. 
• Apply those commitments to compute services that would maximize savings. 
• If you have fluctuating usage, build flexibility into your rate optimization strategy by distributing commitments in small increments with staggered expirations. 

Largest Segment (> $12M annualized compute usage)

You have a mature FinOps practice, but maintaining high ESR across multiple subscriptions is difficult. High coverage means overcommitment risk if usage declines; incremental ESR gains require more risk.

Advice:

• Automate commitment management to increase flexibility and reduce CLR. 
• Develop a system that can monitor usage, identify opportunities, implement Adaptive Laddering, and execute RI exchanges continuously across multiple subscriptions.

Bottom Line

Automation is non-negotiable for top-percentile ESR results. As non-AI and AI-attributable compute grows, manual processes can’t keep pace with the scale and volatility.

ProsperOps automation is purpose-built to implement the strategies outlined here, including Adaptive Laddering, plus more sophisticated techniques not covered in this report. We can help you maximize and grow your savings while minimizing commitment risk.

Want a deeper dive? ProsperOps offers a free savings analysis. This is a quick assessment of your current ESR, how it benchmarks against peers, and the incremental savings potential with automation.

Get a free savings analysis today.

Note About Reading Box Plots

Box plots visualize the percentiles for Azure compute ESR and coverage performance. The horizontal lines in the colored boxes represent the bottom, middle, and top quartiles (25th, 50th, and 75th percentiles, respectively), and the middle line represents the median value. The whiskers extending vertically from the colored boxes indicate the minimum and maximum values within 1.5 times the interquartile range (IQR).

---

About ProsperOps

ProsperOps is the leading FinOps automation platform for cloud cost optimization on Microsoft Azure, Amazon Web Services (AWS), and Google Cloud. Eliminating waste and achieving cost-saving goals are challenging when cloud usage is elastic, but commitments are inelastic. Founded in 2018, ProsperOps reduces costs by synchronizing rate optimization with workload optimization, eliminating waste and boosting cross-team efficiency for FinOps. Our platform drives world-class Effective Savings Rates and mitigates Commitment Lock-In Risk for our customers.