Can AI Agencies Deduct OpenAI, Claude, AWS, Pinecone, Zapier, and Make Costs? (2026 Guide)

Yes, AI agencies can deduct OpenAI, Claude, AWS, Pinecone, Zapier, Make, n8n, Vercel, Supabase, and the rest of the modern AI tool stack as ordinary business expenses. The trickier question is when you deduct them and what category the cost falls into - because OBBBA's new IRC Section 174A draws a line between operating costs (Section 162, immediately deductible) and development costs (Section 174/174A, which has its own rules). For most agencies the federal deduction is immediate either way, but the categorization controls Section 41 R&D credit eligibility and creates a real timing difference for foreign development work. This guide walks through the framework, with a sample chart of accounts and a monthly bookkeeping workflow built for high-variable AI spend.

In This Article

1. The Section 162 vs. Section 174/174A Distinction
2. Development API/Cloud Use vs. Production Delivery Use
3. Domestic vs. Foreign Development - Why It Matters
4. Category-by-Category Deduction Reference
5. Sample Chart of Accounts for AI Agencies
6. Monthly Bookkeeping Workflow for High-Variable API Spend
7. Common Mistakes That Cost Real Money
8. FAQ

The Section 162 vs. Section 174/174A Distinction

Two different sections of the Internal Revenue Code can apply to your spend:

IRC Section 162 covers ordinary and necessary business expenses incurred in carrying on a trade or business. These are immediately deductible in the year incurred. Most of an AI agency's spend - cloud bills serving live client systems, SaaS subscriptions used for daily operations, advertising, rent, utilities - is Section 162.

IRC Section 174 / 174A covers research and experimental (R&E) expenditures. Under post-OBBBA Section 174A: - Domestic R&E expenses are immediately deductible (effective for tax years beginning after December 31, 2024). - Foreign R&E expenses must still be capitalized and amortized over 15 years under the unmodified Section 174. - Software development costs are explicitly within the scope of Section 174/174A.

Why the categorization still matters even when both are immediately deductible domestically: 1. Section 41 R&D credit eligibility - costs treated as Section 174 R&E are eligible to be Qualified Research Expenditures. Section 162 operating costs are not. 2. Foreign R&E is on a 15-year amortization schedule. Misclassifying foreign development as Section 162 understates a long-term tax obligation. 3. Retroactive 2022-2024 small-business opportunity - eligible small businesses (average gross receipts ≤$31M) can amend prior returns to retroactively apply Section 174A to previously amortized 2022-2024 R&E. Mechanics are complex; CPA-review-only.

Development API/Cloud Use vs. Production Delivery Use

Same API account, different tax treatment depending on what it's doing. This is the operational distinction that drives both Section 174A categorization and R&D credit allocation.

Development / Experimental Use (Potential Section 174A R&E)

• Building a new RAG pipeline and benchmarking retrieval strategies - Training or fine-tuning a custom model and evaluating against a base - Designing multi-agent orchestration logic; testing reliability and failure modes - Prototype API calls during development of a new automation - Iterating on prompt strategies as part of a benchmarked experimentation cycle

These costs are incurred to resolve technical uncertainty about whether and how the system will work. They're R&E costs eligible for Section 174A treatment and potentially for the Section 41 R&D credit.

Production / Delivery Use (Section 162 Operating)

• Cloud bills running a live client system in production - LLM API calls actually serving end-user requests in a deployed agent - Routine maintenance, monitoring, observability, logging - Bug fixes and routine updates after a system is live - Documented configuration of an existing template (Zapier scenario, Make.com workflow)

These costs are incurred to operate an already-built system. They're Section 162 operating expenses - fully deductible, but not Section 41 QREs.

Why Most Agencies Don't Track This (and Why They Should)

Bookkeeping software won't categorize this for you. AWS doesn't tag a cost as 'development' or 'production' - you do, via project tags, account separation, or workspace allocation. Most agencies dump all cloud and API spend into a single 'Software & Tools' line and miss the categorization entirely. For agencies under $50K of total cloud/API spend with no R&D credit ambition, that's fine. For agencies above $100K of spend with material development work, the missed categorization is real money.

Practical structure: separate development workspaces, separate AWS accounts, project-tagged Pinecone indexes, project-tagged OpenAI organizations. The cleaner the separation, the easier the categorization at year-end.

Domestic vs. Foreign Development - Why It Matters

Section 174A's immediate-expensing benefit applies only to domestic R&E. Foreign R&E is still on a 15-year amortization schedule under the unmodified Section 174.

What counts as foreign R&E: - Payments to foreign contractors performing development work outside the U.S. - Cloud compute purchased from foreign providers, where the underlying infrastructure is offshore - Foreign-located employees performing development work

What's domestic: - U.S.-based employees and contractors - U.S.-domiciled cloud regions used for development (AWS us-east-1, GCP us-central1, etc.) - U.S.-based SaaS provider development tools

Practical impact for AI agencies using offshore developers: if a foreign contractor in Eastern Europe spends 6 months building a new RAG pipeline for your agency, those payments are foreign R&E. They're capitalized and amortized over 15 years - 1/15 of the cost is deductible in year one, with 14 years of subsequent amortization. Same exact work performed by a U.S. contractor would be 100% deductible in year one under Section 174A. The differential can be material when offshore development is a meaningful share of your build cost.

Mitigation: for projects with mixed teams, track domestic vs. foreign development separately in your project records. Foreign R&E payments should hit a dedicated chart-of-accounts line, not a generic 'Contractor Payments' bucket.

Category-by-Category Deduction Reference

Every cost listed below is deductible. The table identifies the typical Section 162 vs. Section 174A treatment - which controls R&D credit eligibility, not whether the deduction is allowed.

LLM APIs

Cloud Compute and Hosting

Vector Databases and Memory

No-Code / Workflow Automation Platforms

Note: SaaS subscription fees generally do not qualify as Section 41 R&D credit cloud QREs even when used in development. The cloud QRE rule applies to computer use - rented compute power - not to bundled SaaS pricing.

Development Tools and Productivity

Marketing, Sales, and Operations

Hardware and Equipment

Sample Chart of Accounts for AI Agencies

A clean chart of accounts is the foundation. For QuickBooks Online, Xero, or Wave, structure expenses into clearly labeled accounts that already encode the Section 162 vs. potential 174A distinction:

Income - Client Services Revenue - Recurring Retainer Revenue - Affiliate / Reseller Revenue - Other Revenue (templates, courses)

Cost of Services / COGS - LLM API - Production (Section 162) - Cloud Hosting - Production (Section 162) - Vector DB - Production (Section 162) - Pass-through Software Resold to Clients - Subcontractor Payments - Production (US) - Subcontractor Payments - Production (Foreign)

Operating Expenses - SaaS Subscriptions - Tools (Zapier, Make, n8n, Airtable, etc.) - Development Tools (Cursor, Copilot, Linear, GitHub) - CRM and Sales Tools - Advertising and Marketing - Professional Services (CPA, Legal, Insurance) - Office and Admin - Travel and Meals - Education and Training - Bank and Payment Processing Fees

Research and Experimental (Potential Section 174A) - LLM API - Development / Experimentation (US) - LLM API - Development / Experimentation (Foreign) - Cloud Compute - Development / Training (US) - Cloud Compute - Development / Training (Foreign) - Vector DB - Development - Subcontractor Payments - Development (US) - Subcontractor Payments - Development (Foreign) - Internal R&D Wages (allocated portion of W-2)

Note on the structure: the development/production split inside each major spend category (LLM API, Cloud, Vector DB, Subcontractors) is what enables both Section 174A categorization AND Section 41 R&D credit allocation later. If your accountant resists, push back - the marginal effort at categorization time is much lower than the year-end reconciliation cost.

Monthly Bookkeeping Workflow for High-Variable API Spend

AI agencies have notoriously volatile monthly spend. A $40K month followed by an $8K month is normal. The workflow that keeps this clean:

Step 1: Provider-Side Tagging (Set Up Once)

• OpenAI: create separate organizations or projects for development vs. production. The dashboard shows per-project usage; export monthly. - Anthropic: workspaces and API keys per environment. Generate monthly usage reports per workspace. - AWS: mandatory cost allocation tags (Environment=dev|prod, Project=client-x). Cost Explorer breakdown by tag. - GCP / Azure: project-level segmentation; monthly billing exports. - Pinecone / vector DBs: separate dev and production indexes; bill segregation by index name.

Step 2: Monthly Reconciliation

• Download usage CSVs from each provider (first week of the following month while data is fresh) - Reconcile against bank/credit-card statement - Allocate by account: production -> COGS line; development -> R&E line - Tag domestic vs. foreign portions if relevant

Step 3: Quarterly Roll-up

• Aggregate dev vs. production spend for QBR - Cross-check against project P&L and time tracking - Flag any miscategorized line items before they pile up

Step 4: Year-End Close

• Final dev vs. production allocation - Foreign vs. domestic R&E split - Hand to CPA along with project documentation for Section 174A treatment and any Section 41 R&D credit calculation

Tooling shortcut: for agencies with material spend, services like Vantage, Vector, or Anyscale automate cloud cost allocation and tagging. They pay for themselves at >$50K/yr cloud spend.

Common Mistakes That Cost Real Money

Lumping everything into 'Software & Tools.' Easy bookkeeping; impossible to claim R&D credit later. The categorization burden falls on you at year-end, often after the data has aged out of provider dashboards.

Treating SaaS subscriptions as cloud QREs. No-code platforms (Zapier, Make, Airtable) are subscription services. They aren't 'computer use' QREs for Section 41 even when used in development. They're Section 162 operating expenses.

Missing the foreign vs. domestic R&E distinction. A foreign developer's payments aren't deductible immediately under 174A - they're amortized over 15 years. Lumping them with domestic contractor payments understates a real long-term tax obligation.

Not allocating LLM API costs between dev and production. Single ChatGPT Team workspace; single OpenAI org; one giant bill. By tax time the data is gone or so commingled it can't be reasonably allocated. Set up workspace separation before you need it.

Capitalizing software in the wrong category. Some bookkeepers capitalize all software development to be 'conservative.' Post-OBBBA, this gives away the Section 174A immediate deduction. Domestic R&E is generally better expensed unless you have a specific reason to capitalize.

Treating personal-use API spend as fully deductible. If you use ChatGPT Plus, Claude Pro, or Cursor for personal projects in addition to agency work, deduct only the business-use percentage. Mixed-use is normal; documenting the split protects the deduction.

FAQ

Can I just deduct my entire monthly OpenAI bill as a business expense?

Yes, the entire bill is deductible if it's used for business. The question is which line on Schedule C / your P&L. Production use of the API serving live client systems is Section 162 operating expense. Development use - building/testing new automations, fine-tuning experiments, RAG benchmarking - is potentially Section 174A R&E and may also be a Section 41 R&D credit QRE. Same total deduction in either case; different categorization controls credit eligibility.

Are Zapier, Make.com, n8n, and ActivePieces subscriptions deductible?

Yes. All no-code/automation platform subscriptions used in your agency are deductible Section 162 operating expenses. Note that they're SaaS subscriptions, not cloud-compute purchases - so they generally don't qualify as Section 41 R&D credit cloud QREs even when used during development.

Do I need to separate development and production cloud usage?

If you want to claim the R&D credit, yes. If you don't, the Section 162 vs. Section 174A categorization still matters but is less impactful. For agencies under $50K total cloud/API spend with no R&D credit ambition, lumping everything into one 'Software & Tools' line is workable. Above $100K, the cleaner separation pays off.

Can I deduct ChatGPT Plus and Claude Pro?

Yes - if used for agency work. ChatGPT Plus ($20/month), Claude Pro ($20/month), Cursor Pro ($20/month), GitHub Copilot ($10-39/month), and similar developer-productivity subscriptions are fully deductible as Section 162 operating expenses. If you also use these for personal purposes, deduct only the business-use percentage and document the split.

What's the deal with 100% bonus depreciation on hardware?

OBBBA permanently restored 100% bonus depreciation effective for property acquired after January 19, 2025. AI agency hardware (laptops, monitors, GPUs, dedicated servers) qualifies. The federal deduction is immediate. NJ caps Section 179 at $25,000 and does not allow federal bonus depreciation - so material hardware purchases create a NJ-federal timing difference that requires the GIT-DEP worksheet.

What if my offshore developer paid us in crypto?

Same Section 174 / 174A foreign R&E treatment applies. The form of payment doesn't change the categorization; place of performance does. A foreign contractor performing development work outside the U.S. generates foreign R&E regardless of whether they're paid in USD wire, USDC, or BTC. Document the FMV at receipt and the place of performance separately. See the foreign contractor W-8BEN guide.

Are LLM API costs eligible for the R&D credit?

Likely yes, when used for qualified research, but the IRS hasn't issued formal guidance specifically on LLM API costs. Two paths: (1) treat as cloud/computer-use QREs under Reg. 1.41-2(b)(4) - token-based API calls are functionally renting compute; (2) treat as supplies for prototype development under Section 41(b)(2)(A)(ii). Best practice: classify as cloud QREs where possible, document the development-vs-production allocation, note the open guidance issue, and maintain defensible records. See the R&D credit deep-dive.

Ready to Set Up Bookkeeping That Actually Survives Tax Season?

AI agency bookkeeping has unique pain points: variable monthly API spend, mixed development and production use of the same tools, foreign vs. domestic development tracking, and Section 174A categorization that QuickBooks won't do for you. I'm Greg Monaco, a NJ-licensed CPA (License #20CC04711400). I help AI automation agencies set up the chart of accounts, monthly reconciliation workflow, and year-end documentation that supports the deductions you actually want to claim - including the R&D credit if you're doing real experimental work.

Schedule a free 30-minute consultation

Circular 230 Disclosure: This post provides general tax information and is not a substitute for personalized tax advice. Consult a qualified tax professional for advice specific to your situation.

Related Articles

AI agency cluster: AI Automation Agency Hub | R&D Tax Credit | NJ Sales Tax on Custom AI Chatbots | Foreign Contractor W-8BEN | S-Corp Salary for AI Agency Owners