Key Takeaways:
- AI startups can offset up to $500K in payroll tax each year.
- Model training, algorithm design, and infrastructure work often qualify.
- Most AI startups see benefits equal to 6–10% of qualified spend.
- Documentation is key to maximizing savings.
- Federal and state credits can be stacked for even bigger benefits.
A Brief History of the R&D Tax Credit
The Research and Development (R&D) Tax Credit was first introduced in 1981 as a temporary incentive to encourage U.S. companies to increase their investments in innovation. Originally set to expire after a few years, it was extended multiple times before being made permanent in 2015 under the Protecting Americans from Tax Hikes (PATH) Act.
Major milestones include:
- 1981: Credit created as part of the Economic Recovery Tax Act.
- 2015: PATH Act makes the credit permanent and extends eligibility to startups, allowing them to apply credits against payroll taxes for the first time.
- 2023: Expansion to allow offsets against both the employer portion of Social Security and Medicare payroll taxes, doubling potential impact from $250,000 to $500,000.
Today, the R&D tax credit represents a multibillion-dollar annual incentive nationwide, with projections climbing above $17 billion for tax year 2024, according to the Joint Committee on Taxation. Burkland clients alone received over $23M in R&D tax credit in the past year.
Despite its scale, the R&D tax credit remains underutilized by AI startups, with many founders assuming it doesn’t apply to them.
Why the R&D Tax Credit Matters for AI Startups
AI startups face unique financial pressures. Payroll for specialized engineers, rising GPU cloud compute costs, and data infrastructure can quickly consume cash. The R&D tax credit directly reduces these burdens.
Benefits of R&D Tax Credit for AI Startups
- Early-stage boost: Startups with less than $5M in revenue and under five years of operations can apply credits against payroll taxes.
- Non-dilutive funding: Reduces burn without raising additional venture capital.
- Runway extension: Six-figure savings can buy months of additional operating time.
How Much Can AI Startups Save with R&D Tax Credit?
The savings potential is substantial. In practice, most AI startups see a net benefit equal to 6–10% of their qualified R&D spend. For startups under $5M in revenue, up to $500,000 annually can be applied against payroll taxes.
Quick Example:
- $3M annual payroll
- $1.2M of eligible R&D costs (i.e., wages, contracted research, equipment leases)
- Credit range: $72,000–$120,000 (6–10%)
- $72K–$120K applied directly against payroll taxes, applied on quarterly basis and rolling forward for up to 20 years
Estimated Savings Table
| Eligible R&D Spend | Approx. Credit (6–10%)* |
| $500,000 | $30,000–$50,000 |
| $1,000,000 | $60,000–$100,000 |
| $2,000,000 | $120,000–$200,000 |
| $5,000,000+ | $300,000–$500,000 cap |
*Cash savings, applied as a payroll tax offset
What Qualifies for R&D Tax Credit for AI Startups?
To qualify for R&D tax credit, work must pass the IRS’s “Four-Part Test”:
1. Qualifying Purpose
The activity must be aimed at developing or improving a business component—something tied to functionality, performance, reliability, or quality. For an AI startup, this might mean building a recommendation engine that returns results more accurately, training a computer vision model that runs faster on edge devices, or improving the reliability of a fraud detection algorithm.
2. Technological in Nature
The work has to rely on the hard sciences, like computer science, engineering, or math. In practice, that could mean experimenting with neural network architectures, engineering a distributed training pipeline, or applying optimization methods to reduce GPU usage. What wouldn’t qualify are projects grounded in the social sciences, such as user behavior studies or market research.
3. Elimination of Uncertainty
At the start of the project, you must be uncertain about whether something can be done, how it can be achieved, or the best design to use. For example, you may not know if a large language model can meet accuracy thresholds given your dataset, or whether a reinforcement learning approach will scale. The research is meant to resolve these unknowns.
4. Process of Experimentation
You must use a systematic experimentation process to work through the uncertainty. For AI companies, this often looks like testing multiple model variations, simulating outcomes, running hyperparameter sweeps, or comparing inference speeds across different architectures to identify the best option.
Examples of Qualifying & Non-Qualifying Activities:
| ✅Qualifying Activities | ❌Non-Qualifying Activities |
|
|
|
|
|
|
|
|
🤦♂️Common Misconceptions Among Founders
- “We’re too early.” Even pre-revenue startups with payroll can benefit.
- “Model training isn’t research.” Novel architectures and experiments are classic R&D.
- “Our data purchases qualify.” Raw data doesn’t qualify, but building the systems to use it does.
These misconceptions explain why the R&D tax credit for AI startups is often overlooked, even though the work often qualifies.”
How to Claim the R&D Tax Credit
1. Identify eligible work.
Start by working with a tax professional to pinpoint which projects and expenses qualify. For AI startups, this usually includes payroll for engineers and researchers, certain contractors, and cloud compute directly tied to experimentation and model development. Purchased data or routine QA, however, won’t count.
2. Document thoroughly.
The IRS expects clear evidence that your projects involved technical uncertainty and a process of experimentation. Sprint notes, Git commit histories, Jira tickets, time-tracking records, and technical design docs all help build that paper trail. While the R&D tax credit is a powerful benefit, it’s also one of the areas the IRS looks at closely. Claims can attract additional scrutiny, and companies that can’t back up their numbers may lose credits they would otherwise qualify for. By keeping organized records from the start, you not only maximize the value of your claim but also protect it if questions ever come up.
3. Calculate the credit.
Depending on the method, most startups end up with 6–10% of their qualifying spend back as a credit. For companies with heavy R&D payroll, this often means six figures in savings.
4. File the right forms.
- Form 6765 is included with your annual tax return to elect and calculate the credit.
- Form 8974 and Form 941 are then used to apply that credit against payroll taxes for qualified businesses, often filed by your payroll provider after Form 6765 is filed
5. File on time.
The election to apply credits against payroll taxes must be made with the original tax return. You can’t amend later and retroactively apply to payroll, which makes timely, accurate filing essential.
For many founders, these steps can feel like one more thing on a very long to-do list. That’s where a partner like Burkland adds real value. Our tax team works with AI startups every day, helping them track eligible work, prepare airtight documentation, and file correctly to maximize credits. In some cases, we’ve helped companies uncover hundreds of thousands of dollars in savings they didn’t realize were available.
👉 If you’re building in AI, don’t go it alone. Burkland’s tax experts can help you identify, capture, and defend your R&D credits so you keep more cash in the business.
Retroactive Claims
While you can’t amend returns to retroactively apply credits against payroll taxes, you may be able to amend income tax returns to claim R&D credits against income tax for up to three prior years. This option is less relevant for pre-revenue startups but can be valuable for AI companies that reached profitability sooner than expected.
State-Level R&D Tax Credits
In addition to the federal program, many states offer their own R&D tax credits. These state-level incentives can sometimes be applied against income tax, and in a few cases, they are refundable or transferable, providing even more flexibility for startups. The rules vary widely by state. Some are particularly generous, while others are limited or apply only to certain types of taxpayers. For AI startups based in innovation hubs like California or Massachusetts, state credits can be claimed and carried forward for us against future income taxes owed. While New York does not offer an explicit R&D Credit program, incentives like the Qualified Emerging Technology Credit can be explored. Because the formulas and eligibility rules differ, founders are best served by working with an experienced tax advisor who can evaluate both federal and state opportunities together and ensure the company is maximizing its total credit potential.
The Bottom Line
For many AI startups, the R&D tax credit is one of the most effective non-dilutive financing tools available. With six-figure annual savings on the line, it pays to track qualifying work and file correctly the first time.
Burkland’s tax team helps AI startups across the USA capture millions in R&D credits. Don’t leave money unclaimed. Get in touch today to request an R&D tax credit analysis and see how much your startup could save.