Google Cloud CUD vs AWS Reserved Instances vs Azure Reservations: Enterprise Comparison

Why This Comparison Matters for Enterprise Buyers

Enterprise multi-cloud environments almost always involve commitment discount programmes from Google Cloud, AWS, and Azure simultaneously. The temptation is to treat all three as equivalent — commit to usage in exchange for discounts — but the structural differences between programmes create material risk for organisations that apply the same decision logic across clouds.

Getting the commitment mix wrong carries real financial consequences. Over-committing to Standard Reserved Instances when workloads migrate to different instance families wastes sunk cost. Under-purchasing Committed Use Discounts for stable Google Cloud compute means paying full on-demand rates for workloads that could be covered for 37 to 55 percent less. Azure Reservations tied to specific VM series become stranded if the engineering team changes instance types during an optimisation project.

This analysis compares the three programmes on the dimensions that matter for enterprise procurement: discount depth, commitment flexibility, payment mechanics, coverage breadth, and the negotiation layer that sits above each programme for large-spend organisations.

AWS: Reserved Instances and Savings Plans

AWS offers two primary commitment discount mechanisms: Reserved Instances (RIs) and Savings Plans. Understanding the difference between them is one of the most consequential decisions in AWS cost management. The programmes are not interchangeable, and the optimal mix depends on workload stability and engineering team velocity.

Reserved Instances: Maximum Discount, Minimum Flexibility

Standard Reserved Instances commit to a specific EC2 instance type, operating system, tenancy, and region for one or three years. In exchange, AWS offers discounts of up to 72 percent versus on-demand pricing for three-year all-upfront commitments. Regional Standard RIs offer up to 40 to 60 percent for one-year terms depending on instance family. Convertible RIs — which allow exchange to different instance families — offer up to 66 percent discount but sacrifice some savings for the flexibility to change configurations.

The critical limitation of Standard RIs is their specificity. A Standard RI purchased for an m5.xlarge in us-east-1 running Linux provides no discount if the engineering team migrates that workload to an m6i.xlarge or moves to a different region. Zonal RIs (as opposed to Regional RIs) guarantee capacity in a specific Availability Zone, which is valuable for mission-critical workloads, but locks usage even more tightly.

Payment options affect the total discount: all-upfront provides the highest discount, partial-upfront provides moderate discounts, and no-upfront provides the lowest discount rate within the committed category. For enterprises optimising working capital, no-upfront RIs still deliver meaningful savings versus on-demand while preserving cash flow.

Savings Plans: Flexibility Over Maximum Discount

Savings Plans, introduced by AWS in 2019, commit to a dollar-per-hour spend level rather than specific instance configurations. Compute Savings Plans (the most flexible type) apply to any EC2 instance family, region, size, operating system, or tenancy, and also cover AWS Fargate and Lambda. EC2 Instance Savings Plans commit to a specific instance family in a region but allow flexibility on size, operating system, and tenancy.

Compute Savings Plans deliver up to 66 percent savings versus on-demand — slightly less than the maximum Standard RI discount — but apply automatically across changing compute configurations. EC2 Instance Savings Plans deliver up to 72 percent savings for the committed family, matching Standard RI discount levels with more sizing flexibility. For most enterprise teams, Compute Savings Plans have replaced Standard RIs as the default commitment tool because engineering agility has increased while the operational overhead of managing RI exchanges has become a real cost.

The practical guidance: use Standard RIs only for workloads that are genuinely stable at the instance family and regional level for the full commitment term, and use Compute Savings Plans for everything else. Combining both is common — baseline stable workloads on Standard RIs, variable workloads on Savings Plans.

The Data Egress Problem AWS Doesn't Emphasise

Data egress is the most common surprise cost in AWS environments and is conspicuously absent from most commitment discount conversations. AWS charges $0.09 per GB for data transferred out to the internet from most regions, with inter-region transfer charged at $0.02 per GB. For data-intensive workloads, egress costs routinely exceed compute commitment savings — a reality that enterprise buyers only discover after commitments are locked in.

Egress costs are not covered by Reserved Instances or Savings Plans. They are metered separately on on-demand rates. Any multi-cloud or hybrid architecture that routes significant data through AWS egress paths must model these costs independently before evaluating the net benefit of compute commitments. Organisations moving data from AWS to Google Cloud or Azure pay AWS egress rates that can erode the discount advantage of Reserved Instances within 12 months.

AWS EDP: The Layer Above Commitment Programmes

The AWS Enterprise Discount Program (EDP), now technically termed a Private Pricing Agreement (PPA), sits above RI and Savings Plans discounts. EDP provides a percentage discount on total AWS billing in exchange for a multi-year spend commitment. Meaningful EDP discounts start at approximately $2 million in annual AWS commitment — organisations below that threshold typically receive little or no EDP benefit. At the $10 million-plus annual spend level, EDP discounts of 10 to 25 percent on gross billing become achievable, stacked on top of RI and Savings Plan savings.

The EDP negotiation covers which services are in-scope, the discount percentage, commit escalation schedules, and what happens when spend falls short of commitment. Poorly structured EDP agreements lock enterprises into spend commitments that become difficult to exit if cloud strategy changes, so the EDP terms require as much scrutiny as the discount percentage.

Google Cloud: Committed Use Discounts

Google Cloud's commitment programme operates under the Committed Use Discount (CUD) framework, which encompasses two distinct mechanisms: resource-based CUDs and spend-based CUDs. Google also applies automatic Sustained Use Discounts (SUDs) for Compute Engine resources that run for more than 25 percent of a month — a passive saving that requires no commitment but applies only to Compute Engine.

Resource-Based CUDs: Deepest Discounts for Compute

Resource-based CUDs commit to specific Compute Engine resources — vCPU, memory, GPU, local SSD — in a specific region for one or three years. The discount structure is straightforward: one-year resource CUDs deliver 37 percent off on-demand pricing, and three-year CUDs deliver 55 percent. For memory-optimised machine types, three-year discounts reach 70 percent. These discounts apply to the committed resource type across any machine type that uses those resources within the committed region, giving more flexibility than AWS Standard RIs without sacrificing all the discount depth.

A key advantage of Google's CUD structure over AWS Standard RIs: resource-based CUDs are not tied to a specific VM instance configuration. A commitment for 100 vCPUs in us-central1 covers any Compute Engine VM that uses those vCPUs in that region, regardless of VM series or configuration changes. This provides more natural coverage for engineering teams that regularly resize or change VM families without requiring exchanges or conversion transactions.

Spend-Based CUDs: Flexibility Across Services

Spend-based CUDs commit to a minimum hourly spend amount across a broader set of Google Cloud services. One-year spend-based CUDs deliver 28 percent off eligible services; three-year commitments deliver 46 percent. The eligible service list for spend-based CUDs has expanded significantly through 2022 and 2023 to include Cloud SQL, Cloud Bigtable, Firestore, and other managed services, making them relevant to workloads beyond pure Compute Engine.

Cloud SQL CUDs are worth special attention: one-year delivers 25 percent, three-year delivers 52 percent — meaningful savings for database-heavy environments. Bigtable and Firestore CUDs deliver 20 percent (one-year) and 40 percent (three-year) respectively. Spend-based CUDs provide the coverage flexibility of AWS Savings Plans but apply to a broader Google Cloud service portfolio than the equivalent AWS mechanism.

Google's Multiprice CUD Transition

As of July 2025, Google Cloud completed a shift to a multiprice CUD consumption model. Under the old model, a single CUD price applied uniformly. Under the new model, CUDs consume at tiered rates depending on the on-demand price of the resource being covered. This change matters most for enterprises with diverse compute profiles: workloads running on higher-priced machine types consume CUD value faster, requiring enterprise buyers to model their specific workload mix rather than applying a flat discount rate assumption.

Negotiated Discounts Above CUDs

For enterprises committing $10 million or more annually to Google Cloud, private negotiated discounts — distinct from CUDs — become available. These negotiated savings stack on top of CUD discounts and typically require multi-year contractual commitments similar to AWS EDP. For eight-figure Google Cloud commitments, negotiated discounts of 15 to 25 percent on eligible services are achievable, representing a procurement layer that most enterprises access through direct Google account teams or specialist advisors rather than the standard CUD purchase flow.

Azure: Reservations and Savings Plans

Azure's commitment programme mirrors AWS in offering both Reservations (equivalent to AWS RIs) and Azure Savings Plans (equivalent to AWS Compute Savings Plans, introduced in 2022). Azure Reservations are one of the most mature cloud commitment mechanisms and offer genuine flexibility advantages in specific dimensions.

Azure Reservations: Depth and Scope

Azure Reserved VM Instances commit to a specific VM series in a specific region for one or three years. One-year reservations typically deliver 36 to 40 percent off pay-as-you-go pricing; three-year reservations deliver 55 to 72 percent depending on VM series and region. Azure's reservation pricing is among the most aggressive for VM discounts at the three-year level — a function of Azure's pricing strategy for multi-year enterprise commitments.

Azure Reservations cover a broader service portfolio than AWS Standard RIs or Google resource-based CUDs: SQL Database, SQL Managed Instance, Cosmos DB, App Service, Azure Databricks, Azure VMware Solution, and other managed services all have reservation pricing. For enterprises with significant Azure-native managed service consumption, the breadth of reservation coverage provides more optimisation surface than equivalent programmes on other clouds.

Azure's reservation exchange and refund policy is notably more flexible than AWS Standard RIs: enterprise buyers can exchange reservations for equivalent or greater value within the same VM family, and partial refunds are available (subject to a 12 percent early termination fee) for reservations that no longer align with workload requirements. This flexibility reduces the risk of over-committing to specific Azure configurations compared to AWS Standard RI lock-in.

Azure Savings Plans: Commitment Without Specificity

Azure Savings Plans, launched in late 2022, commit to an hourly spend amount in exchange for discounts across Azure compute services including VMs, Azure Kubernetes Service, and Azure App Service, regardless of VM size, region, or operating system. Savings Plan discounts typically fall between Reservation rates and pay-as-you-go rates — delivering roughly 15 to 35 percent savings depending on the commitment level and term. For workloads that change regions or VM families regularly, Azure Savings Plans provide meaningful savings without reservation rigidity.

Enterprise Azure Agreement customers can use Azure Prepayment (Monetary Commitment) to purchase reservations and Savings Plans, avoiding invoice-cycle billing and integrating cloud commitment spend into existing EA financial structures — an important cash management consideration for organisations with established Azure EA agreements.

Side-by-Side Comparison

The following comparison addresses the dimensions enterprise buyers consistently prioritise when structuring multi-cloud commitment portfolios:

• Maximum discount depth: AWS Standard RI and Azure Reservations both reach 72 percent at three-year all-upfront terms. Google Cloud resource CUDs reach 55 percent (70 percent for memory-optimised). AWS Savings Plans and Azure Savings Plans deliver 66 to 72 percent and 15 to 35 percent respectively.
• Flexibility mechanism: AWS provides two distinct programmes (RIs for stability, Savings Plans for flexibility). Google's CUDs inherently provide more flexibility than AWS Standard RIs through resource-type coverage rather than instance-type specificity. Azure offers exchanges and refunds on Reservations plus Savings Plans for maximum flexibility.
• Coverage breadth: Azure Reservations cover the widest range of managed services. Google's spend-based CUDs cover a growing managed service portfolio. AWS Savings Plans cover compute broadly but exclude many managed services covered only by specific RI types.
• Payment options: AWS offers all-upfront, partial-upfront, and no-upfront across most commitment types. Google's CUDs require no upfront payment — discounts apply to ongoing billing automatically. Azure Reservations support upfront and monthly payment options.
• Capacity guarantees: AWS Zonal RIs guarantee capacity in specific Availability Zones. Google CUDs and Azure Reservations do not provide capacity guarantees — discounts only.
• Egress exposure: AWS has the most significant egress pricing exposure, relevant to multi-cloud architectures where data moves between providers. Google and Azure have lower base egress rates, though neither is free for production-scale data movement.
• Negotiation layer: All three providers offer private pricing agreements above commitment programmes at $2M+ annual spend (AWS EDP), $10M+ (Google negotiated discounts), and within EA/MCA structures (Azure).

Common Mistakes in Multi-Cloud Commitment Management

Treating RI and CUD decisions as equivalent: The lock-in risk of an AWS Standard RI for a specific instance type in a specific region is materially higher than the equivalent Google resource CUD. Enterprise buyers who apply the same coverage ratio approach across both programmes without adjusting for flexibility differences systematically over-commit on AWS.

Ignoring the egress offset: Discount analysis that counts only compute savings without modelling data egress costs understates the true cost of AWS-anchored architectures. Before locking in a three-year AWS EDP, model the egress costs of the expected multi-cloud data flows. Egress costs of $1 to $2 million annually can erode the value of an EDP discount that appeared significant in isolation.

Purchasing commitments before engineering visibility: Commitment programmes require 12 to 36 months of covered usage to deliver their projected savings. Purchasing commitments before the engineering team has validated the architecture, instance types, and regional footprint creates stranded reservation risk. The best time to purchase Standard RIs or three-year CUDs is after 6 to 12 months of on-demand usage demonstrates stable consumption patterns.

Not using the negotiation lever: At $2M or more in AWS annual spend, $5M or more in Azure spend, or $10M or more in Google Cloud spend, the commitment discount programme discounts are not the best available pricing. Private pricing negotiations deliver additional percentage points that require proactive engagement — they do not appear automatically. Many enterprises at these spend thresholds pay commitment programme rates without ever accessing the negotiated tier.

Recommendations for Enterprise Multi-Cloud Buyers

The first recommendation is straightforward: model your specific workload profile before purchasing any commitment at scale. The discount percentages published by all three providers represent maximums under optimal conditions. Your actual blended discount depends on your mix of stable versus dynamic workloads, service portfolio, regional spread, and payment preference.

For AWS, default to Compute Savings Plans for variable workloads and add Standard RIs only for workloads you can genuinely guarantee at the instance family level for the full term. Use AWS Cost Explorer recommendation data as input, not instruction — the recommendations optimise for coverage coverage, not cost efficiency.

For Google Cloud, resource-based CUDs are the most efficient commitment mechanism for compute-heavy environments. Spend-based CUDs make sense for mixed environments where Compute Engine represents less than 60 percent of total Google Cloud spend. Review the multiprice CUD transition impact on your specific workload mix before modelling 2023 and beyond savings.

For Azure, leverage the exchange mechanism. Purchase Reservations for stable workloads but plan reservation reviews at quarterly intervals to exchange configurations that have drifted from the original commitment. Use Azure Savings Plans as the baseline commitment for development and test environments and any workloads with significant architectural change expected in the commitment window.

Across all three clouds, engage your account teams at 60 to 90 days before major commitment renewals. The private pricing negotiation window is at commitment renewal, not mid-term. Leverage competitive positioning — demonstrating that your multi-cloud footprint creates optionality between providers — is the most effective lever for extracting negotiated discounts above standard commitment programme rates. Contact Redress Compliance to benchmark your current commitment portfolio against peer organisations and identify the gaps between what you are paying and what is achievable.