The question isn't whether DBs will be running in clusters - they all pretty much do, or can - but will those clusters be orchestrated by something first designed for dynamic scaling of stateless services, treating all nodes as varying-sized pools of otherwise mostly-homogeneous hyperconverged compute/storage? Or, how much are you willing to pay in $/performance/dev time to force them into such infrastructure?

Or will they be on something more purpose-built, either designed from the start to handle storage more flexibly at a higher level? Or conversely something more specific and lower-level to extract maximum performance for their particular storage patterns?