Defending a useful life for a capitalised AI asset
What IAS 38 requires
A finite-life intangible is amortised over its useful life, which is the period over which the asset is expected to be available for use IAS 38 §97. The estimate weighs expected usage, typical product life cycles, technical and commercial obsolescence, and the stability of the industry. For AI assets the obsolescence factor dominates: the relevant question is not how long the code runs but how long the model remains the one the business would choose to use.
The factors that shorten a token-built life
- The release cadence of superior foundation models the asset is built on or competes with.
- Announced deprecation timelines for the underlying model or endpoint.
- Falling inference prices that make a rebuild cheaper than continued use.
- Contractual or product commitments that fix a shorter horizon.
Judgement, honestly flagged
There is no settled market convention for the useful life of a capitalised model, and reasonable preparers differ. This reference does not manufacture a single number. The defensible approach is to estimate the life from the specific obsolescence evidence for your asset, document that evidence, and prefer the shorter life where the evidence is finely balanced, because a shorter life is harder for an auditor to argue is overstated IAS 38 §98.
Reviewing the estimate
Useful life is not set once. IAS 38 requires the amortisation period and method to be reviewed at least at each financial year-end, and a change is accounted for prospectively as a change in accounting estimate IAS 38 §104. For AI assets this review is substantive, not a formality, because the obsolescence evidence moves quickly. A model announced for deprecation mid-year is a clear prompt to shorten the remaining life.