This shows some of the challenges in “industrially relevant” quantum advantages:

- It’s complicated to use quantum machines
- There are complexity speedups, but large prefactors
- It may improve precision, but some algorithms may not be accurate
- Not all advantages are relevant!
- Variational algorithms: hard to estimate convergence time

GPU analogy: GPUs had low adoption because speedups were only modest, and the algorithms had to be completely rewritten! Perhaps the same is true with near-term quantum computing.

Prediction: Expect modest speedups in CAS for 25<N<60, with emphasis on non-dynamic & degenerate correlations