What limits a physical screen
A screening campaign can only return compounds that are physically present in its collection, and its cost scales with how many it tests. Both limits are structural, not a matter of better equipment: the collection is finite and small relative to chemical space, and every additional compound costs another assay.
Where AI changes the arithmetic
AI does not make the space smaller. It changes the strategy from sampling an existing shelf to proposing new structures and ranking them, so experimental effort is concentrated on a short list of candidates that are more likely to be worth making.
Everything so far has been about the problem, not the method. How a molecule becomes machine-readable data is the next step, and it is what makes generation and comparison possible at all.