In relation to FSI, specifically Portfolio Optimization, implementing the sklearn.covariance.* packages would be helpful for shrinkage on covariance matrices. I have already requested LedoitWolf #7514
and EmpiricalCovariance is another popular technique in research, publications, code that would benefit from GPU acceleration.
scikiit-learn link:
https://scikit-learn.org/stable/modules/generated/sklearn.covariance.EmpiricalCovariance.html#sklearn.covariance.EmpiricalCovariance