Can Cloud Computing Costs Qualify for the R&D Tax Credit?

Last reviewed October 14, 2026

Yes. Cloud infrastructure costs can qualify as supply expenses under IRC Section 41(b)(2) when they are used in the conduct of qualified research. As more companies move their development, testing, and experimentation workloads to AWS, Google Cloud, and Microsoft Azure, cloud costs have become one of the most significant — and most commonly overlooked — categories of qualified research expenses (QREs).

This page explains which cloud costs qualify, which do not, how to separate R&D spending from production spending, and how to document everything for an IRS-ready credit claim.

Cloud costs as qualified supply expenses

Under IRC Section 41(b)(2), supply expenses qualify for the R&D tax credit when they are used in the conduct of qualified research. The IRS defines supplies as tangible property other than land or depreciable property that is used or consumed during the research process.

Cloud computing costs fit this definition because they represent compute resources, storage, and networking capacity that are consumed during R&D activities. Unlike physical servers (which are depreciable capital equipment), cloud resources are expensed as they are used — making them eligible as supply expenses rather than capital expenditures.

The key requirement is straightforward: the cloud resources must be directly tied to qualified research activities. Cloud costs incurred for production hosting, general business operations, or activities that do not meet the four-part test do not qualify.

Which cloud costs qualify

The following categories of cloud spending commonly qualify as R&D supply expenses when they are used for qualified research:

• Compute instances used for R&D. AWS EC2 instances, Google Cloud Compute Engine VMs, and Azure Virtual Machines used to run development workloads, experiments, prototypes, and proof-of-concept builds. This includes any compute time consumed while engineers are evaluating technical alternatives and resolving uncertainty.
• GPU and TPU costs for machine-learning model training. Training ML models typically involves significant compute expense. GPU instances (such as AWS P4d, GCP A2, or Azure NDv4) and TPU resources consumed during model experimentation, hyperparameter tuning, and architecture evaluation are strong candidates for the credit. See also AI Development R&D Credit.
• Database services used during development and testing. Managed database instances (RDS, Cloud SQL, Cosmos DB) that support R&D workloads — including development databases, test data stores, and experimental data pipelines — qualify when they are consumed during the research process.
• CI/CD pipeline compute costs. Build servers, automated test runners, and deployment pipeline compute used to evaluate and validate R&D work. When CI/CD infrastructure runs tests that are part of the process of experimentation (evaluating whether an approach works), those compute costs are tied to qualified research.
• Development and staging environments. Cloud resources provisioned specifically for development and staging — where engineers build, test, and iterate on uncertain technical approaches — qualify as R&D supply expenses. These environments exist solely to support the research process.
• Storage used for R&D data. S3 buckets, Cloud Storage, and Blob Storage used to store training datasets, experimental results, development artifacts, and test fixtures consumed during qualified research. Storage that holds production data serving end users does not qualify.

Which cloud costs do NOT qualify

Not all cloud spending is tied to R&D. The following costs typically fall outside the scope of qualified supply expenses:

• Production hosting. Compute, storage, and networking resources used to serve end users in production do not qualify. Once a product or feature has moved past the R&D phase and into commercial production, the associated cloud costs are no longer tied to qualified research.
• General business SaaS subscriptions. Email services, CRM platforms, project management tools (Jira, Asana), communication tools (Slack, Teams), and other business software are not supply expenses consumed during research. They support business operations broadly, not the research process specifically.
• Storage for non-R&D data. Cloud storage used for marketing assets, customer data in production, backups of production systems, and other non-research purposes does not qualify.
• Costs incurred after commercial production begins. Under Section 41(d)(4)(A), research expenses incurred after the beginning of commercial production are excluded. Once a product is released to customers, ongoing cloud costs for that product are production expenses, not R&D expenses — even if the engineers continue to make improvements. (However, genuinely new R&D work on the next version or a substantially new capability can start a new qualifying activity.)

How to separate R&D cloud costs from production

The biggest challenge with claiming cloud costs is separating R&D spending from production spending. The IRS expects clear, defensible allocation methods. Here are the four most common approaches, ranked from strongest to weakest from a documentation standpoint:

1. Separate cloud accounts or projects

The cleanest approach is to run R&D workloads in dedicated cloud accounts. AWS Organizations lets you create separate accounts for R&D and production. Google Cloud projects and Azure subscriptions serve the same purpose. When all resources in an account exist solely for R&D, the entire account bill can be claimed as a supply expense — no allocation math required.

2. Resource tagging

If separate accounts are not practical, tag all cloud resources with R&D project identifiers. Most cloud providers support cost allocation tags (AWS), labels (GCP), or tags (Azure) that flow through to billing reports. Create a consistent tagging taxonomy — for example, rd-project: search-engine-v2 — and enforce it through policy. Then pull cost allocation reports filtered by R&D tags to calculate qualifying expenses.

3. Cost allocation reports

Cloud providers offer detailed cost-and-usage reports that break down spending by service, resource, time period, and tag. AWS Cost and Usage Reports, GCP Billing Export, and Azure Cost Management all provide this data. Use these reports to identify and total the resources consumed by R&D activities. Keep the reports as part of your documentation.

4. Time-based allocation for shared infrastructure

When R&D and production workloads share the same infrastructure — for example, a shared Kubernetes cluster — a time-based allocation method can apportion costs. If a cluster runs R&D workloads 40% of the time and production workloads 60% of the time (supported by scheduling logs, pod metrics, or usage data), 40% of the cluster cost can be attributed to R&D. This method is acceptable but requires supporting data.

The supply expense rules in detail

Under IRC Section 41(b)(2), supplies are defined as any tangible property other than land or depreciable property that is used in the conduct of qualified research. To qualify, the supply must be either consumed or have its useful life shortened by the research activity.

Cloud computing resources satisfy this test because:

• They are expensed (not capitalized), so they are not depreciable property.
• They are consumed during the research process — once a compute instance runs an experiment, that compute time is gone.
• They are used directly in the conduct of qualified research activities that meet the four-part test.

The IRS has not issued guidance specifically addressing cloud computing as a supply expense, but the tax community widely treats cloud costs as analogous to traditional supply expenses like materials, utilities, and consumable equipment used during research. The logic is straightforward: if you would have claimed the electricity and hardware depreciation on a physical server running R&D workloads, the equivalent cloud costs serve the same purpose.

Practical example

Consider a mid-size SaaS company that spends $200,000 per year on AWS. Their cloud usage breaks down as follows:

• Production hosting (EC2, RDS, CloudFront for customer-facing application): $100,000
• Development and staging environments (dedicated EC2 instances, RDS dev databases): $30,000
• CI/CD pipeline compute (CodeBuild, test runners): $15,000
• ML model training (GPU instances for recommendation engine experiments): $25,000
• R&D data storage (S3 buckets for training data and experiment results): $10,000
• General business and ops (monitoring, backups, email via SES): $20,000

The qualifying R&D cloud costs total $80,000 — the sum of development and staging environments ($30,000), CI/CD pipeline compute ($15,000), ML model training ($25,000), and R&D data storage ($10,000). The $100,000 in production hosting and $20,000 in general business costs do not qualify.

This company uses separate AWS accounts for production and R&D workloads and tags resources with project identifiers, making the $80,000 figure easy to substantiate with billing reports.

Documentation best practices for cloud costs

Strong documentation turns a defensible claim into a bulletproof one. For cloud supply expenses, maintain the following records:

• Monthly cost allocation reports from your cloud provider, filtered by R&D tags, accounts, or projects. Download and archive these reports — do not rely solely on the provider dashboard.
• Tagging policy documentation that defines your R&D tagging taxonomy, who is responsible for tagging, and how compliance is enforced.
• Account or project structure documentation showing which cloud accounts or projects are designated for R&D versus production.
• Mapping of cloud costs to qualifying projects. Each cloud expense should be traceable to a specific R&D project that meets the four-part test. Include project descriptions, the technical uncertainties involved, and the experimentation performed.
• Allocation methodology documentation if you use time-based or usage-based allocation for shared resources. Describe the method, the data sources used, and the calculations performed.
• Screenshots or exports of resource configurations showing that specific instances, clusters, or storage buckets were provisioned for R&D purposes.

For comprehensive guidance on what records to keep, see Documentation Requirements.

How TaxCredit4U helps with cloud cost claims

TaxCredit4U is a self-service platform that guides you through identifying and documenting your R&D expenses, including cloud supply costs. Here is how it works:

• Guided project entry. You enter your R&D projects and describe the technical uncertainties involved. For each project, you specify the associated costs — including cloud infrastructure expenses — and the platform organizes them into the correct QRE categories.
• Supply expense tracking. Cloud costs are entered as supply expenses and linked to the qualifying projects they support. The platform ensures each expense is properly categorized and documented.
• Credit calculation. TaxCredit4U calculates your credit using both the Alternative Simplified Credit (ASC) and Regular Research Credit (RRC) methods so you can compare and choose the best option.
• IRS-ready documentation. The platform generates a complete study report with project narratives, expense breakdowns, and all calculations needed to support your Form 6765 filing.
• CPA review. Every study is reviewed by an R&D tax credit specialist before delivery.

Frequently asked questions

Do SaaS development tools like GitHub and Jira qualify as R&D expenses?

Generally no. SaaS tools like GitHub, Jira, and Slack are not consumed during the research process — they facilitate project management and collaboration but are not supply expenses under IRC Section 41(b)(2). However, cloud compute resources that directly process R&D workloads — such as EC2 instances running experiments or GPU clusters training machine-learning models — do qualify because they are consumed in the conduct of qualified research.

Can I claim cloud costs for a failed R&D project?

Yes. The R&D tax credit does not require that a project succeed. As long as the project met the four-part test — permitted purpose, technological in nature, technical uncertainty, and process of experimentation — the associated cloud costs qualify as supply expenses regardless of the outcome. Many legitimate R&D efforts involve experiments that do not produce the desired result, and the IRS recognizes this.

How do I separate R&D cloud costs from production cloud costs?

The most reliable method is to use separate cloud accounts or projects for R&D workloads. AWS Organizations, GCP projects, and Azure subscriptions all support this approach. If separation is not practical, use resource tagging to label R&D resources with project identifiers, then generate cost allocation reports filtered by those tags. For shared infrastructure, a time-based allocation method — supported by logs or usage data — can apportion costs between R&D and production use.