I don't see how this is even remotely true. Unless there's some super breakthrough into a fundamentally different architecture, there's not really a path to a 50% reduction in price, much less a 99% reduction.
And yet 90% drops for the same level of quality every 18 months have happened like clockwork...
And the technology already exists on the algorithmic front TODAY to lock in another 10x gain -> when, typically, algorithmic gains only account for ~30% of that drop and the other ~70% comes from better data (often synthetic) and knowledge distilation from frontier models.
Just look at DeepSeek's pricing...
In fairness, I think _current_ capabilities will be cheaper. So the models of today will be run drastically cheaper in 4 years.