A superseding model or price collapse as a trigger
Supersession is an external indicator
A superior model that a competitor can adopt, or that makes the asset's output look dated, changes the market the asset earns in. That is an adverse change in the technological environment, which IAS 36 lists among the external indicators to consider at each reporting date IAS 36 §12. The asset need not stop working; it is enough that its expected economic benefits have fallen.
Price collapse changes the economics
A sharp fall in inference prices cuts both ways. It lowers the cost of running the asset, but it also lowers the cost of rebuilding equivalent capability from scratch, which can erode the value that justified capitalising the build. Where a rebuild becomes cheap relative to the carrying amount, that is evidence the recoverable amount may sit below carrying value and a test is warranted.
Reassessing value in use
Value in use is the present value of the future cash flows expected from the asset IAS 36 §30. After a supersession event those cash-flow forecasts are revisited: lower expected usage, a shorter remaining horizon, or margins compressed by cheaper alternatives. The revised value in use, compared with fair value less costs of disposal, gives the recoverable amount against which any loss is measured IAS 36 §18.
External evidence and reversal
The trigger should be evidenced with external facts: the superseding model's release, benchmark or capability comparisons, and observed price movements. IAS 36 permits a later reversal of an impairment loss on an intangible, other than goodwill, if the indicators that caused it have reversed and recoverable amount has recovered, capped at the carrying amount that would have existed absent the original loss. Reversals of this kind are rare for AI assets, because obsolescence seldom un-happens.