Content reviewed and verified by Graham Chee, with FCPA-led practice at Local Knowledge, Mascot NSW. Continuous CPA Australia member since 1986. Prior career at Goldman Sachs, BNP Investment Management and Merrill Lynch.. Last reviewed August 2026. Next review scheduled for November 2026.
TL;DR
Unlock balance sheet value by correctly accounting for proprietary AI models and fine-tuned LLMs under AASB 138.
Introduction: Unlocking the Balance Sheet Value of Your AI Innovations
In an era where Artificial Intelligence (AI) is rapidly transforming business operations, Australian enterprises are investing significant capital into developing bespoke AI solutions, including proprietary AI model weights and fine-tuned Large Language Models (LLMs). While the operational benefits are clear, the financial reporting implications, particularly the potential for these investments to be recognised as capitalisable assets on the balance sheet, often remain opaque. This analysis, written by Graham Chee, FCPA, GRCP — Fellow of CPA Australia since November 2005, continuous CPA member since 1986, and principal of Local Knowledge — delves into the intricate accounting treatment of these advanced AI developments. We move beyond generic discussions of AI adoption to provide a deep technical dive into AASB 138 'Intangible Assets', offering a clear pathway for Australian businesses to account for their custom AI R&D expenditure. By adhering to rigorous accounting standards, businesses can accurately reflect their true asset base, enhance financial transparency, and provide a more robust valuation for stakeholders. This article will equip you with the knowledge to distinguish between research and development phases for AI, apply the recognition criteria for intangible assets, and understand the valuation methodologies pertinent to proprietary AI in the Australian context.
Understanding AASB 138: The Foundation for AI Intangible Assets
AASB 138 'Intangible Assets' provides the authoritative framework for accounting for intangible assets in Australia. It defines an intangible asset as an identifiable non-monetary asset without physical substance. For custom AI logic, such as proprietary algorithms, trained model weights, and fine-tuned LLMs, meeting this definition is crucial. Identifiability is key; an asset is identifiable if it is separable (i.e., capable of being separated or divided from the entity and sold, transferred, licensed, rented or exchanged) or arises from contractual or other legal rights. In the context of AI, this often relates to intellectual property rights, such as copyright or patent protection over the underlying code, architecture, or unique training datasets and methodologies that result in specific AI model weights. Control over the asset is also paramount, meaning the entity has the power to obtain the future economic benefits flowing from the underlying resource and can restrict others’ access to those benefits. For proprietary AI, this typically stems from legal enforceability, technical safeguards, or the confidential nature of the AI’s design and training data. Finally, future economic benefits must be expected to flow to the entity, which for AI can include revenue from product sales, cost savings, or other benefits derived from its use. The standard mandates that an intangible asset shall be recognised only if it is probable that the expected future economic benefits that are attributable to the asset will flow to the entity and the cost of the asset can be measured reliably [AASB 138, para 21].
Distinguishing AI Research from Capitalisable Development Costs
Criteria for Recognising Proprietary AI Weights and LLMs as Assets
For costs related to proprietary AI weights and fine-tuned LLMs to be capitalised, an entity must demonstrate that all six specific criteria outlined in AASB 138, paragraph 57, have been met. These criteria provide a stringent test for capitalisation:
- Technical Feasibility: The entity must demonstrate the technical feasibility of completing the intangible asset so that it will be available for use or sale. For AI, this means proving the model can function as intended and achieve its designed purpose.
- Intention to Complete: The entity must have the intention to complete the intangible asset and use or sell it. This requires clear strategic alignment and business plans.
- Ability to Use or Sell: The entity must demonstrate its ability to use or sell the intangible asset. This involves assessing market demand, internal capacity, or strategic fit.
- Future Economic Benefits: The entity must demonstrate how the intangible asset will generate probable future economic benefits. This could be through market analysis for external sales or internal cost-benefit analysis for internal use.
- Availability of Resources: Adequate technical, financial, and other resources to complete the development and to use or sell the intangible asset must be available. This includes access to computational power, skilled personnel, and funding.
- Reliable Measurement of Cost: The entity must be able to reliably measure the expenditure attributable to the intangible asset during its development. This necessitates robust cost tracking systems.
Meeting these criteria for AI development often requires detailed project management, clear documentation of milestones, and a robust system for tracking direct and indirect costs associated with the AI’s creation. Without satisfying all six, the expenditure must be expensed [AASB 138, para 57].
Valuation Methodologies for AI Intangible Assets in Sydney
Once the recognition criteria for capitalising proprietary AI weights and LLMs are met, the next step involves determining their initial measurement and subsequent valuation. AASB 138 requires an intangible asset to be measured initially at cost. For internally generated AI, this cost includes all directly attributable costs necessary to create, produce, and prepare the asset to be capable of operating in the manner intended by management. This can encompass personnel costs (salaries of AI engineers, data scientists), materials and services consumed (e.g., cloud computing resources for training, third-party data acquisition), legal fees for intellectual property protection, and amortisation of patents and licences used to develop the asset. Indirect costs, such as general administrative overheads, are typically excluded unless they are directly attributable to preparing the asset for use [AASB 138, para 66].
Subsequent measurement can be either the cost model or the revaluation model. Given the nascent and rapidly evolving nature of AI, the cost model (cost less accumulated amortisation and impairment losses) is generally more prevalent due to the difficulty in reliably measuring fair value using a revaluation model in active markets for unique AI assets. Amortisation should be systematic over the asset's useful life, which for AI can be challenging to estimate due to rapid technological obsolescence. An impairment review is also critical, particularly if there are indicators that the AI asset's carrying amount may not be recoverable. Specialist valuation expertise, often from firms in Sydney with specific experience in intangible asset valuation, may be required to robustly support these measurements and useful life estimations.
The Role of FCPA in Navigating AI Asset Recognition Compliance
The complexities of applying AASB 138 to cutting-edge AI developments underscore the critical role of experienced accounting professionals, particularly those with FCPA credentials. A Fellow of CPA Australia (FCPA) brings a deep understanding of Australian Accounting Standards, ethical obligations under the APESB Code of Ethics for Professional Accountants, and practical experience in complex financial reporting scenarios. For AI asset recognition, an FCPA can provide invaluable guidance by:
- Interpreting AASB 138: Applying the nuanced definitions of research and development to specific AI projects, ensuring correct classification of expenditure.
- Establishing Robust Cost Tracking: Designing and implementing systems to reliably capture all directly attributable costs for capitalisation, a fundamental requirement of AASB 138.
- Assessing Recognition Criteria: Providing objective judgment on whether the six criteria for capitalisation (technical feasibility, intention, ability, benefits, resources, reliable measurement) have been met, often requiring collaboration with technical AI teams.
- Determining Useful Life and Amortisation: Advising on appropriate useful lives for AI assets, considering technological obsolescence and commercial lifespan, and establishing suitable amortisation policies.
- Impairment Testing: Guiding management through the process of assessing AI assets for impairment indicators and performing impairment tests as required by AASB 136 'Impairment of Assets'.
- Disclosure Requirements: Ensuring compliance with the extensive disclosure requirements of AASB 138, which provide transparency to financial statement users regarding the nature and value of intangible assets.
Our principal-led practice ensures that every file receives the rigorous scrutiny and expert judgment required for such intricate accounting matters, adhering strictly to the CPA Code of Ethics [APESB: APES 110 Code of Ethics for Professional Accountants].
Practical Steps for Capitalising Your AI Development Expenditure
Successfully capitalising AI development costs requires a structured and disciplined approach. Here are practical steps Australian businesses should consider:
- Define Project Scope and Phases Clearly: From inception, delineate AI projects into distinct research and development phases. Document the objectives, expected outcomes, and technical milestones for each phase. This clarity is crucial for applying AASB 138.
- Implement Granular Cost Tracking: Establish a dedicated project code or cost centre for each AI development project. Track all direct labour, materials, software licenses, cloud computing usage, and external consulting fees specifically attributable to the development phase. General overheads are typically excluded.
- Document Technical Feasibility and Commercial Intent: Maintain detailed records of technical proof-of-concept, successful pilot tests, and internal assessments confirming the AI's functionality. Concurrently, document business cases, market analyses, and management approvals that demonstrate the intention to use or sell the AI and its expected economic benefits.
- Regularly Assess Recognition Criteria: At each reporting period, formally review whether all six criteria under AASB 138, paragraph 57, continue to be met. Any failure to meet a criterion necessitates expensing further costs or potentially impairing previously capitalised amounts.
- Engage with Accounting Professionals Early: Collaborate with your FCPA or experienced accounting firm from the outset. Their expertise will be invaluable in structuring cost capture, making critical judgments, and ensuring compliance with all reporting standards. This proactive engagement mitigates risks and ensures robust financial reporting.
By following these steps, businesses can build a strong evidentiary basis for capitalising their proprietary AI investments, aligning their financial statements with the true value of their technological advancements [ATO: TR 97/24 Income tax: research and development tax concession].
Future Implications: AI, AASB 138, and Australian Business Growth
The ability to correctly account for proprietary AI development under AASB 138 has significant implications for Australian businesses. By capitalising eligible costs, companies can present a more accurate and robust balance sheet, reflecting the true investment in their intellectual capital. This not only enhances financial transparency but also can positively influence key financial metrics such as return on assets and equity, potentially improving access to capital and investor perception. As AI becomes increasingly integral to competitive advantage, the financial reporting of these assets will become a key differentiator.
Furthermore, the discipline required to meet AASB 138's stringent criteria encourages better project management, clearer strategic planning, and more rigorous cost control in AI development initiatives. This internal rigour can lead to more efficient R&D spending and a clearer understanding of the commercial viability of AI projects. For Australia's innovation ecosystem, clarity in AI asset recognition fosters greater investment in cutting-edge technologies, as businesses can more accurately reflect the value created by their R&D efforts. As the regulatory landscape evolves, particularly with the rapid pace of AI advancement, ongoing vigilance and expert accounting advice will be paramount to ensure continued compliance and to fully leverage the financial reporting opportunities presented by AI innovation. This approach aligns with the broader goal of fostering sustainable growth and innovation within Australian enterprises.
Frequently Asked Questions About Capitalising AI Costs
Q.Can all AI development costs be capitalised?
No, not all AI development costs can be capitalised. Only costs incurred during the 'development phase' that meet all six specific recognition criteria under AASB 138, paragraph 57, are eligible. Costs from the 'research phase', which involves gaining new knowledge, must be expensed as incurred. This distinction is crucial and requires careful judgment based on the project's specific activities and milestones. For example, exploring various neural network architectures is research, while training a proprietary LLM for a specific commercial application, once technical feasibility is proven, may qualify as development [AASB 138, paras 54-57].
Q.What kind of costs are typically included when capitalising AI models?
Capitalisable costs for AI models typically include directly attributable expenditures necessary to create, produce, and prepare the asset for its intended use. This often comprises salaries and wages of AI engineers and data scientists directly involved in development, costs of cloud computing resources for model training, specific software licenses, data acquisition costs for proprietary training datasets, and legal fees for intellectual property protection related to the AI. General administrative overheads or costs from the research phase are generally excluded [AASB 138, para 66].
Q.How is the useful life of an AI intangible asset determined?
Determining the useful life of an AI intangible asset is challenging due to rapid technological advancements and market changes. It requires careful consideration of factors such as expected technological obsolescence, the stability of the underlying AI architecture, anticipated changes in customer demand, and the entity's plans for using the asset. The useful life should be the period over which the asset is expected to contribute to the entity's economic benefits. If a reliable estimate of useful life cannot be made, the asset is considered to have an indefinite useful life, though this is rare for AI. Regular review of the useful life is essential [AASB 138, paras 90-97].
Q.Are open-source LLM fine-tuning costs capitalisable?
Costs associated with fine-tuning open-source LLMs can be capitalisable if the resulting fine-tuned model meets the AASB 138 recognition criteria. The key is whether the entity creates a proprietary asset through its unique data, training, and customisation efforts, leading to identifiable future economic benefits. If the fine-tuning process creates a distinct, controllable, and valuable asset (e.g., a model uniquely tailored for a specific business function that provides a competitive advantage), and all six criteria are met, then the directly attributable development costs may be capitalised. Simply using an open-source model without significant customisation typically would not create a capitalisable asset [AASB 138, para 21].
Q.What happens if an AI project fails after costs have been capitalised?
If an AI project fails after development costs have been capitalised, the entity must assess whether there are indicators of impairment. AASB 136 'Impairment of Assets' requires an entity to assess at each reporting date whether there is any indication that an asset may be impaired. If such an indication exists (e.g., the project is abandoned, technical issues prevent commercialisation, or expected economic benefits diminish), the entity must estimate the asset’s recoverable amount. If the carrying amount of the AI asset exceeds its recoverable amount, an impairment loss must be recognised immediately in profit or loss, reducing the asset's value on the balance sheet [AASB 136, paras 9-12].
Principal's Insight: Navigating the AI Accounting Frontier
In principal-led practice at Local Knowledge, we observe a growing demand from founder-led businesses and SMEs to understand how their significant investments in custom AI can be reflected accurately in their financial statements. The challenge often lies not just in the technical application of AASB 138, but in bridging the communication gap between AI developers and financial reporting teams. Our approach involves working closely with both sides, translating technical AI milestones into accounting realities. We emphasise meticulous documentation and a clear audit trail for all development costs. The future of business is intrinsically linked with AI, and ensuring these valuable assets are correctly recognised is paramount for strategic planning, investor relations, and ultimately, sustainable growth.
Unlock the Value of Your AI Investments
Navigating the complexities of AASB 138 for proprietary AI models requires specialist expertise. Our FCPA-led practice, Local Knowledge, is equipped to guide your business through the intricacies of capitalising custom AI logic and LLM costs, ensuring compliance and accurate financial reporting. Don't let your significant AI investments remain invisible on your balance sheet. Speak with our principal, Graham Chee, to discuss your specific AI development projects and how we can help you unlock their true financial value.
About the Author
Graham Chee, FCPA, CPA, GRCP, GRCA
Principal and Founder, Local Knowledge
Graham Chee is the principal and founder of Local Knowledge, an FCPA-led Australian practice that brings institutional-grade compliance, investment-structure and intellectual-property experience directly to owner-managed businesses. Graham is a Fellow of CPA Australia (FCPA since November 2005, continuous CPA member since 1986) and holds the OCEG Governance, Risk & Compliance Professional (GRCP) and Governance, Risk & Compliance Auditor (GRCA) designations. His prior career includes senior roles at Goldman Sachs, BNP Investment Management and Merrill Lynch. Graham was previously portfolio manager of the Asian Masters Fund (IPO December 2007 – 31 December 2009), which returned +29% in AUD terms versus the MSCI Asia Pacific (ex Japan) benchmark. He signs off on 100% of client files personally.
Areas of Expertise:
Industry-specific insights
This article is especially relevant to these industries. See how we tailor our services for each.
Contact Us Today
This article provides general information only and does not constitute financial or accounting advice. Speak to us for advice specific to your situation. Every file is signed off by our principal under the CPA Code of Ethics.
Graham Chee FCPA, CPA, GRCP, GRCA · Principal, Local Knowledge · Mascot NSW · CPA-signed files