That assumes scaling laws still hold up. A bigger model might end up only incrementally more intelligent.
"The future" being "whenever training and inference at increased scale becomes economical". Which is probably bounded by new generations of hardware, but might also be pushed forward by algorithmic advances.