Accounting Standards Update on AI Software Costs
The one-paragraph answer
FASB ASU 2025-06 is the Financial Accounting Standards Board's Accounting Standards Update on internal-use software costs, updated to address AI implementations. It clarifies which AI-related costs can be capitalized, which must be expensed, and how to disclose them. Companies investing in AI systems face significant P&L implications depending on classification. FASB ASU 2025-06 shapes how AI investments appear in earnings and how CFOs plan around them.
AI implementations are expensive. The question of whether costs are capitalized (spread over time, protecting current earnings) or expensed (hitting the P&L immediately) is a real quarter-by-quarter concern. FASB ASU 2025-06 provides guidance but leaves substantial judgment. CFOs and controllers need to apply the guidance carefully to specific AI investments, which requires understanding both the accounting rules and the technical nature of the AI project.
Costs incurred during the application development stage, once technological feasibility is established, generally can be capitalized. For AI, this includes certain model development, integration, and deployment costs.
Preliminary project costs (research, alternative evaluation) and post-implementation costs (training, maintenance, most retraining) must be expensed.
Model retraining costs, prompt engineering, and ongoing model maintenance raise classification questions that FASB ASU 2025-06 addresses. The line between "development" and "maintenance" is thinner for AI than for traditional software.
Companies must disclose the nature of capitalized software costs, the amortization period, and impairment considerations.
Public company financials must comply. Private companies following GAAP must comply. Investor communications and analyst expectations increasingly reference AI capital treatment. Impairment analysis for AI-related capitalized costs is a growing audit area.
The academic literature on FASB ASU 2025-06 is ahead of most corporate practice, and it is unusually blunt. Two findings are worth putting in front of any executive who thinks this is a compliance formality.
“AI adoption significantly enhances corporate governance effectiveness and improves risk management”
That is the gap between having AI and governing it. The second finding is the one that tends to change the room.
“The introduction of AI algorithms in public services modifies the chain of responsibility.”
Neither of these is a fringe position. Both come from peer-reviewed work, and both describe the condition most organisations are actually in when the question about FASB ASU 2025-06 arrives from the board, the buyer, or the regulator.
This is the sequence that works, and it is not the sequence most organisations choose. They start with the framework and work backwards toward reality. Start with reality.
Done in this order, FASB ASU 2025-06 becomes tractable. Done out of order, it becomes a document nobody uses and a control nobody exercises.
Depends on classification and timing. Data purchased for a specific development project may qualify; general data acquisition may not.
Ongoing operating costs are typically expensed. Development-phase usage may qualify for capitalization in specific contexts.
The CFO addendum in Volume III of The Operating Discipline for AI Library™ addresses AI accounting decisions and links them to strategic AI planning.
The preliminary project stage is expensed. Once technological feasibility is established and the project enters application development, direct costs can generally be capitalised. After the system is placed in service, most costs are expensed again. The difficulty with AI is that the three stages blur. A model is rarely "finished". It is retrained, re-tuned, and re-evaluated continuously, and each of those activities has to be classified as either continued development or ongoing maintenance.
Four questions carry most of the accounting weight under FASB ASU 2025-06. Is a retraining run maintenance of an existing asset or development of a new one? Is prompt engineering configuration or development? Are training data acquisition costs a direct cost of the asset or a general expense? And are foundation model API charges an operating cost or, during a defined development phase, capitalisable? Reasonable auditors differ. What they will not accept is a policy that changes to suit the quarter.
Capitalised AI costs sit on the balance sheet as an asset, and assets get tested for impairment. AI moves fast. A model capitalised eighteen months ago may have been superseded by a commodity API that does the job better for a fraction of the cost. Auditors are beginning to ask whether the carrying value is still supportable. Companies that capitalised aggressively to protect earnings should expect that question, and should have an answer that does not depend on optimism.
The authoritative texts and agency pages behind this summary. We keep this page current, but where a compliance decision turns on exact wording, read the source. Anything concerning FASB ASU 2025-06 that carries legal consequence should be confirmed against the enrolled text or the issuing body, not against a secondary summary, including this one.
The AI Business Enablement Audit™ measures your organization against every framework in this library, including FASB ASU 2025-06, and delivers a defensible governance dossier. Start or finish your audit below.
Start or finish your AI Audit →