There are many workloads for which all available database engines are poor. If you have one of those workloads and the esoteric technical expertise, it is entirely plausible to improve performance, scale, etc metrics by 10-100x versus whatever is currently available in the market. 10-100x is qualitative if that is central to your business.

Of course, almost no one should attempt this. The number of people with the technical expertise to pull it off successfully is much, much smaller than the number of companies with workloads that would benefit from this.

It doesn't have to be an exotic workload. Sometimes the market is just full of weak implementations e.g. graph databases.