Determining Useful Lives of AI Assets

Introduction

In the swiftly evolving domain of artificial intelligence (AI), technologies like large language models (LLMs) are advancing at an unprecedented pace. These models, foundational to numerous AI applications, require substantial investments in development and data acquisition. For companies at the forefront of AI innovation, accurately estimating the useful life of LLMs and the associated data assets is a critical accounting estimate. This determination impacts amortization schedules, financial statements and footnote disclosures, and strategic decision-making, ensuring that financial reporting aligns with the economic realities of these rapidly evolving technologies.

In this article, part of our ongoing AI Accounting Insights series discussing accounting considerations for companies developing generative AI technologies, we review the factors influencing the determination of useful livesof capitalized (1) LLM development assets and (2) data assets useful lives.Understanding these considerations helps financial leaders align accounting practices with the realities of technological innovation and market demand.

AI Assets

Generative AI refers to a type of AI that can produce new content — whether it’s text, images or code — based on the patterns it has learned from existing data. Generative AI is capable of generating content that is novel, coherent and contextually appropriate. This technology is powered by advanced machine learning techniques that enable computers to understand and mimic human creativity and language complexities.

One of the key innovations behind generative AI is the development of LLMs. LLMs are designed to process and generate human-like text. They are built using deep learning techniques and trained on vast amounts of data, including books, articles, websites and other written content. The model learns the structure of language, including grammar, context and the relationships between concepts, which enables it to generate relevant and contextually accurate responses. By leveraging LLMs through generative AI applications, businesses can automate and enhance a variety of language-intensive tasks, leading to increased efficiency and innovation.

LLMs form the backbone of generative AI applications and are increasingly integrated into various business applications. Developing LLMs and generative AI applications can entail significant costs across multiple stages, whether an entity chooses to build an LLM internally or license an existing model for fine-tuning. For internally developed LLMs, expenses begin with the conceptualization and design phase, which includes salaries for AI researchers and engineers who define the model’s purpose, architecture and algorithms. The training phase is notably resource-intensive, demanding substantial computational power through high-performance hardware or cloud-based solutions, and often represents one of the most significant cost drivers. Subsequent phases like fine-tuning the model for specific tasks, testing and validation to ensure accuracy and reliability, and deployment and maintenance further contribute to the overall costs.

Acquiring and preparing datasets for training LLMs is another significant cost, especially as freely available data is becoming less novel and the most valuable data requires purchase or license. When these acquired data assets are expected to provide future economic benefits and meet certain capitalization criteria, they are capitalized as intangible assets.

Alternatively, licensing an existing LLM reduces development time and upfront expenses but involves licensing fees and costs associated with fine-tuning the model using specific datasets, as well as similar expenses for testing, deployment and ongoing maintenance. AI-focused entities can increasingly capitalize on a significant portion of the costs associated with developing LLMs and generative AI applications for internal use.

Refer to our prior articles, which further address each of these topics in more detail:

• Accounting Considerations in the Development of AI Projects
• Accounting for Data Acquisition Costs Related to AI Projects

Understanding the useful lives of these capitalized assets is essential for proper amortization and impairment assessment.

Accounting Considerations for Determining the Useful Lives of AI Assets

The useful life reflects the period over which the asset is expected to contribute to future economic benefit to the entity, which can be influenced by technological advancements, expected usage, market dynamics, legal factors and competitive pressures. Capitalized internal-use software development costs are considered intangible assets. They are covered under ASC 350-40,Internal-Use Software, while capitalized data assets, which are considered separate intangible assets, are addressed under ASC 350-30,General Intangibles Other Than Goodwill.

Intangible assets with finite useful lives are amortized over their estimated useful lives, reflecting the pattern in which the asset’s economic benefits are consumed. Factors to consider in estimating useful life include:

• Expected Use: How the entity expects to use the asset.
• Legal or Contractual Provisions: Any legal or contractual terms that may limit the asset’s useful life.
• Obsolescence: The effects of technological advancements, competition or other economic factors.
• Maintenance Expenditures: The level of ongoing investment required to maintain the asset’s functionality.

Determining the Useful Life of Capitalized LLM Development Costs

Determining the useful life of capitalized LLM or generative AI application development costs involves assessing several key factors, including:

Technological Obsolescence. Innovations such as new algorithms, model architectures and training methodologies can emerge swiftly, rendering existing AI generative AI applications or LLMs less competitive or even obsolete. As a result, the anticipated period during which a current AI technology will remain technologically relevant may be relatively short. Management must evaluate the pace of technological change within their specific market. If emerging technologies are likely to supersede the current LLM or application within a year or two or less, the useful life should reflect this shorter horizon. Regular collaboration with technical teams can provide insights into expected technological trajectories.

Market Demand. Market dynamics play a significant role in determining an AI technology’s useful life. Shifts in customer preferences, emerging use cases, and evolving industry standards can impact the demand for products or services based on a particular LLM or application. For instance, if customers begin favoring models or applications with enhanced ethical considerations, such as bias mitigation or explainability, an existing LLM lacking these features may see reduced demand. Assessing market trends and customer feedback helps estimate how long the LLM or application will continue generating economic benefits. A proactive approach to product development and adaptation can extend the useful life but may also involve additional capital expenditures.

Competitive Landscape. Competition, especially from the open-source community, can significantly influence the useful life of proprietary LLMs. Open-source LLMs can offer high-quality alternatives that can diminish the competitive advantage of proprietary models. As these open-source models improve and gain adoption, companies may need to accelerate their innovation cycles to maintain market position. This competitive pressure may shorten the useful life of an LLM and related applications, as companies must update or replace models more frequently. When estimating useful life, it’s important to consider the potential impact of open-source developments and the company’s capacity to respond effectively.

Legal and Regulatory Factors. Legal considerations, including intellectual property rights, data privacy laws and industry regulations, can affect the useful life of AI technologies. Compliance requirements may necessitate modifying existing models or limit their applicability in certain markets. For example, new regulations on data usage could restrict the deployment of an LLM and applications trained on certain datasets. Understanding the legal environment and anticipating regulatory changes are essential in estimating the useful life. Non-compliance risks impair the asset’s value and may lead to additional costs or liabilities.

Given the above factors and depending on how new models and applications are developed and how frequently technologies are replaced, the useful life of certain AI technologies may be relatively short. Even the largest, most robust models are replaced within 1-2 years or less. While older models may remain available at lower price points, their future benefits are limited as customers move to the latest models or applications.

Given how quickly the AI industry is evolving, some LLMs and generative AI applications may have useful lives of less than one year. While companies may argue against capitalizing the costs associated with these technologies due to their short lifespan, often overlooked is the fact that there may still be an impact to the income statement. If the LLMs or related generative AI applications are revenue-generating, these capitalizable costs should likely be included as part of the costs of services rather than classified as research and development. This distinction is important, as it can significantly affect gross margin and will likely become a hot topic going forward, given the substantial costs associated with building LLMs. We encourage management to consult with their auditors as this is an area where interpretations are still being developed.

Determining the Useful Life of Acquired Data Assets

Data assets often have benefits beyond the current AI model or application as the entity expects to utilize existing datasets to train future technologies. Considerations in determining the useful life of capitalized data assets include:

Nature and Quality of Data. Data assets vary in their longevity based on their content and applicability. Foundational data, like textbook or encyclopedic knowledge, tend to have enduring value. In contrast, data tied to current events, market conditions, or specific time-sensitive information may become obsolete quickly. Evaluating the type of data and its expected relevance over time helps determine its useful life. Data that can be reused across multiple models or projects may justify a longer useful life.

Technological Advancements. Advancements in AI can affect the utility of existing data assets. New training techniques may require different types or formats of data, potentially reducing the usefulness of previously acquired datasets. Additionally, improvements in data synthesis or augmentation technologies might lessen the reliance on raw data inputs.

Legal and Contractual Considerations. Data assets acquired under perpetual licenses or outright purchases generally have a longer useful life, assuming they remain relevant. However, data obtained through term-based licenses is limited to the agreement’s duration. Changes in data licensing terms or revocation of rights can affect the asset’s useful life. Moreover, evolving data privacy regulations, such as the General Data Protection Regulation (GDPR) or the California Consumer Privacy Act (CCPA), may restrict data usage.

Strategic Business Shifts. Data can become obsolete due to changes in business strategy, market focus, or external factors. For example, if a company shifts its focus to a different industry or geographic market, existing data assets may no longer be relevant. Regular strategic reviews can identify such shifts early, allowing for adjustments in amortization schedules.

Certain data assets may have an indefinite useful life — meaning there’s no foreseeable limit to the period over which they are expected to provide economic benefits. As a result, these assets would not be amortized. The indefinite life conclusion would be reevaluated each reporting period, and the assets would be subject to a more rigorous impairment test than those with definite useful lives.

Amortization Method Considerations

Once the useful lives of AI assets are determined, companies must select appropriate amortization methods. The method should reflect how the asset’s economic benefits are consumed. Given the rapid change and how quickly AI technologies may become outdated, an accelerated amortization method may be more appropriate than a straight-line approach.

Conclusion

Determining the useful lives of capitalized LLM development costs and data assets is a complex task for AI companies, requiring a thorough understanding of technological trends, market dynamics, competitive forces and legal considerations. Accurate estimation ensures that financial statements reflect the true value of these assets and that amortization aligns with the periods benefiting from them.

CFOs and controllers should proactively engage with technical teams, monitor industry developments and regularly review asset performance. This collaborative approach allows companies to adapt effectively to changes, maintain compliance with accounting standards and make informed strategic decisions.

At CBIZ ARC, we specialize in providing top-tier technical accounting and financial consulting services to growth-oriented companies, including those leading in AI innovation. Our expert teams bring a unique combination of deep accounting knowledge and a clear understanding of emerging technologies, helping companies navigate complex financial, systems and data management challenges. From addressing technical accounting issues and preparing for IPOs to unlocking deeper insights with AI-driven tools, we deliver customized solutions that ensure compliance, financial transparency, and performance optimization. Renowned for our commitment to seamless execution and responsive client service, CBIZ ARC provides the strategic support to drive your company’s sustained growth and long-term success.