I am deeply sceptical of the 7x claim. It's easy to find a great improvement when you chose a bad baseline. For example if you just plonk a traditional fan-on-a-stick in the same place, it will probably be terrible, but there are non-AI upright designs that might be in the same ballpark. I don't doubt they have a way to make substantial site-specific design adjustments, but I do doubt that much improvement over the most suitable but not customised design you can already buy today.

Also the AI angle is pretty silly hype because multidimensional optimisation techniques have been part of the industrial design toolbox for decades. Wind turbine companies are very keenly aware of optimisation algorithms for squeezing everything out of a blade geometry. CFD is an entire field for a reason.

Random 2006 article in numerical optimisation of vertical wind turbines from a careless search: https://iawe.org/Proceedings/CWE2006/MC3-01.pdf