While the OP is wrong about being able to "do everything in parallel" via quantum superposition (we have plenty of no-go theorems when it comes to quantum computers being faster simply due to parallelization) some interesting algorithms that are relevant to data science in general have been developed. For example, this[1] one provides an exponential speedup (over known classical algorithms) for calculating persistent homology, a useful method for identifying scale-independent patterns.

[1]: https://arxiv.org/abs/1408.3106

with a proper algorithm which may well not exist it's in theory possible to e.g. perform deep learning on all training examples in a single pass or something like that.

Surely you jest? For AI (QC or not) to bring such a result, even with perfect AI, would require delegating governmental controls on a wide scale. Hard research requires hardware, people, money, etc. So does your AI first solve the problem of how governments make decisions? And how do you conclude that an AI would decide that researching fusion is the best choice?

Likely huge, I'm already aware of two QC companies in Canada who are staffing up to look at building a quantum cloud and offering quantum compute workloads "in the cloud" - so to speak.