For US businesses building or customizing AI, the most valuable funding lever in 2026 isn’t a grant — it’s the tax code. Two provisions work together to offset the cost of AI and software development. This guide explains them with official IRS sources. (dgm is an AI integration partner, not a tax firm — treat this as orientation and work with a qualified CPA on any claim.)

Two levers, working together

There are two distinct benefits, and it’s easy to confuse them:

  • A deduction for research expenses (how much of your R&D you can write off, and when).
  • A credit for increasing research activities (a dollar-for-dollar reduction in tax).

You can generally use both: deduct qualifying domestic R&D under Section 174A and claim the Section 41 credit on the qualifying portion. Together they materially lower the after-tax cost of building AI.

Section 174A: immediate expensing is back

The big 2026 change is the return of immediate expensing. The One Big Beautiful Bill Act (P.L. 119-21, signed July 4, 2025) created new Section 174A, which permanently allows full, immediate deduction of domestic research and experimental expenses for tax years beginning after December 31, 2024 — reversing the earlier rule that forced companies to capitalize and amortize them. The IRS implemented the change in Revenue Procedure 2025-28, which also lets eligible small businesses elect to apply Section 174A retroactively and amend prior returns. Foreign research expenses, by contrast, are still amortized over a longer period.

For a company spending on AI engineers and software development in the US, this means those costs are deductible now rather than spread over years — a real cash-flow improvement.

Section 41: the R&D credit

The R&D tax credit reduces tax dollar for dollar on qualifying research. To qualify, research must meet a four-part test: it must be a permitted-purpose business component, be technological in nature (the IRS lists engineering and computer science explicitly), aim to eliminate technical uncertainty, and proceed through a process of experimentation. AI and software development frequently fit this pattern.

There’s an important nuance for software: internal-use software (built primarily for your own operations) must also clear a high-threshold-of-innovation test — innovation, significant economic risk, and not commercially available without modification. Software developed to sell, lease, or license to others generally isn’t subject to that heightened test. The credit is claimed on Form 6765, and qualified expenses generally include wages for qualified services, supplies, and a portion of contract research.

The payroll-tax offset for startups

A common misconception is that the credit only helps profitable companies. Qualified small businesses can apply the R&D credit against payroll taxes instead of income tax — and that cap was raised to $500,000 per year for tax years beginning after December 31, 2025. For an early-stage AI company with little income-tax liability but real payroll, that can turn the credit into near-term cash savings. Eligibility has conditions (including gross-receipts limits), so confirm with a CPA.

What this means for an AI project

If you’re paying people to build AI agents, models, integrations, or custom software in the US, you may be sitting on two benefits at once: an immediate deduction for the spend and a credit on the qualifying portion. The keys to capturing it are scoping the work so the technical uncertainty and experimentation are clear, and documenting it well — contemporaneous records matter if the claim is ever examined.

How dgm helps

dgm implements osFoundry and other AI for US businesses; we are not a tax firm. Where we add value here is upstream: scoping and documenting your AI development so it is better positioned to qualify for R&D treatment, and building the AI itself. The credit calculation and filing should be handled by your CPA or a qualified R&D tax specialist — but knowing the lever exists often changes how a company budgets an AI build in the first place.