I'm not sure what you mean, we have already scaled up AI exponentially. The amount of AI compute in the world has been doubling every 7 to 8 months [1], so it is already exponential despite not being able to do human or super human research. The % of all AI compute going towards academic style knowledge research is quite low as well. So it stands to reason AI Compute used to do research would in fact scale up exponentially if we did figure out how use weights in a datacenter to do frontier research.

That doesn't mean were going to go right to Dyson spheres so that every possible molecule is going towards scaling.

[1] https://epoch.ai/trends