Artificial intelligence has become an increasingly important component of software development, quantitative research, modeling, and technology innovation across the financial services industry. Hedge funds, proprietary trading firms, asset managers, and fintech companies are integrating AI into various core R&D processes. These include coding acceleration, model enhancement, large-volume data analysis, test automation, and software improvement. Rather than replacing traditional research efforts, AI has become a tool that allows technical teams to evaluate more alternatives and experiment at a speed and scale that was previously impractical.
This trend has coincided with the widespread availability of cloud-based AI platforms that provide access to substantial computing resources on a consumption basis. Instead of investing in dedicated infrastructure, companies increasingly purchase access to externally hosted computing power through AI platforms and cloud service providers. As a result, AI consumption, cloud computing charges, model training costs, and related usage-based technology expenditures have become a meaningful component of modern research and development spending. When incurred in connection with qualified research activities, these expenditures may constitute qualified research expenses (“QREs”) under Internal Revenue Code (“IRC”) Section 41 as amounts paid for the right to use computers in the conduct of qualified research. See IRC § 41(b)(2)(A)(iii).
Statutory and Regulatory Framework
IRC Section 41 expressly includes amounts paid for the use of computers in the conduct of qualified research. In modern technology environments, cloud computing and AI token usage reflect direct payment for access to computational capacity used in R&D activities. Where these resources are used to design, test, and refine technology, the expenditures fall within the statutory definition of qualified research expenses.
Section 41 specifically contemplates the inclusion of certain computer usage costs within the concept of QREs. Treasury Regulation § 1.41-2(b)(4) provides that amounts paid to another person for the right to use computers in the conduct of qualified research may qualify where the computers are owned and operated by someone other than the taxpayer, are located off the taxpayer’s premises, and the taxpayer is not the primary user of the computer. These rules were created well before AI was even conceptualized and have been historically applied to time-sharing computer arrangements. However, their intent remains highly relevant in modern cloud computing environments.
Modern AI platforms generally operate in a manner consistent with the original intent of the legal requirements. The underlying computing resources are typically owned and operated by third-party providers, housed in remote data centers, and made available to numerous customers through shared infrastructure. Users purchase access to computational capacity rather than the underlying hardware itself. Consequently, cloud-based AI usage often exhibits the same economic characteristics as the computer rental arrangements contemplated by Section 41 and Treasury Regulation § 1.41-2(b)(4).
Treasury regulations also confirm that payments for external computing resources qualify when used directly in research activities, and these principles apply equally in cloud environments. Newer regulations issued in 2025 further clarify that cloud transactions involve access to computing resources provided as a service. AI token consumption represents measured use of that computational capacity and is therefore consistent with the treatment of computer usage under Section 41.
AI Development and Direct Nexus to Qualified Research
AI and data-driven development efforts in fintech consistently involve technical uncertainty and iterative testing. Model development requires evaluation, adjustment, and validation across changing datasets and conditions. Applications such as algorithmic trading, fraud detection, risk modeling, transaction optimization, and compliance analytics require continuous refinement and reflect a process of experimentation.
AI-assisted development activities frequently occur within the types of iterative and technically uncertain environments that Section 41 is intended to incentivize. Technical teams routinely use AI tools to evaluate alternative software architectures, generate and test code, optimize algorithms, validate model outputs, develop new analytical techniques, and improve system performance. To the extent these activities are undertaken to eliminate technical uncertainty through a process of experimentation and satisfy the requirements of IRC § 41(d), the associated AI computing expenditures may possess a direct nexus to qualified research activity. Treasury Regulation § 1.41-4 recognizes that experimentation may include modeling, simulation, and systematic evaluation of alternatives, all of which frequently occur within AI enabled development environments.
The connection between computing costs and qualified research is most direct when these resources are used within development and testing workflows. In these environments, usage is tied to model building, evaluation, and improvement rather than operational execution. Segregating development activity from production use allows companies to isolate the portion of computing spend that qualifies for inclusion as qualified research expenses.
AI Token Pricing Reflects Measured Use of Computing Resources
The structure of modern AI pricing models further supports characterization of these expenditures as payments for computational resources. Leading AI providers generally charge customers based upon actual consumption of computing capacity rather than through the purchase of software or hardware. For example, Anthropic’s Claude platform prices usage based on measured token consumption, including separate charges for input tokens, output tokens, prompt caching, batch processing, and related computational functions. Anthropic also permits billing through marketplace structures that convert token usage into standardized consumption units while maintaining a direct relationship between charges incurred and computing resources consumed.
This pricing methodology is significant because it demonstrates that taxpayers are paying for access to processing capacity hosted on remote computing infrastructure. Each request submitted to an AI model consumes a quantifiable amount of computational resources, which the provider measures and bills. As a result, AI token expenditures closely resemble traditional computer usage costs and may be viewed as a modern extension of the computer rental and time-sharing arrangements specifically contemplated by Section 41 and the accompanying Treasury Regulations.
Reporting Requirements and Practical Implementation
Recent changes to Form 6765 require reporting at the business component level and emphasize clear identification of research activities and associated costs. This framework supports the inclusion of AI and cloud computing expenditures by encouraging alignment between computing usage and specific models or systems.
In practice, companies can support these costs through system-generated data such as usage logs and billing records that quantify development-related activity. Documentation of testing, iteration, and model refinement provides evidence of the experimental process and establishes a direct connection between the costs incurred and the underlying research activities.
Key Distinction: Development vs Production
Computing usage incurred in development and testing activities is properly included as a qualified research expense, while production or operational usage falls outside this scope. In a typical technology environment, development activity involves iterative testing, model refinement, and performance evaluation, whereas production systems are used to deliver finalized outputs. A clear distinction between these environments allows for precise identification of includible costs and strengthens the overall position.
Takeaway for Fintech and Trading Organizations
AI token consumption and cloud computing costs represent a logical evolution of the computer usage expenditures expressly recognized by Section 41. Although the statutory and regulatory framework predates modern artificial intelligence platforms, the fundamental economic arrangement remains the same: taxpayers are purchasing access to externally provided computational resources to support research and development activities. Section 41 explicitly includes amounts paid for the right to use computers in the conduct of qualified research, and Treasury Regulation § 1.41-2(b)(4) provides a framework that is often consistent with modern cloud computing and AI infrastructure arrangements.
As AI continues to transform software development, quantitative research, trading technology, and data analytics, AI-related computing expenditures are likely to represent an increasingly significant category of research costs. When appropriately linked to qualified research activities, supported by contemporaneous documentation, and segregated from production usage, AI token consumption and cloud computing expenditures present a well-supported basis for inclusion as qualified research expenses under Section 41.
