The Core Debate is this:
Bulls (Cantor, IREN management) argue:
GPUs cascade from training → inference → batch processing
Hyperscalers use 6-year depreciation, so IREN should too
Total economic life is 7-9 years when you factor in secondary uses
Bears (Chanos) argue:
NVIDIA releases new GPUs every year now (not every 2 years)
GPUs become economically obsolete in 2-3 years for frontier AI workloads
Physical life ≠ economic life
Using 9-year schedules is accounting gimmickry to inflate earnings