Calculated demand load is less than 385A right now. However, we are talking about feeder even if at some point load is greater than 385A next round up rule would make 750kcmil size would be ok no? Otherwise what point of having next round up rule it does not state calculated load has to be less than 385A?
The 750 kcmil AL wire is OK as long as the calculation that "rounds up" to 400A, is 385A or less, and it is a circuit where 240.4(B) applies. If the scope of the loads served by this disconnect increases beyond "adding up" to 385A, the 750 kcmil AL is no longer compliant. If it "adds up" to 395A, you'd need 800 kcmil AL, but verify that you can connect it first. If it "adds up" to 396A, you'd need the full 400A worth of wire, i.e. 2x 250 kcmil AL in parallel. If it is my choice in this example, I would parallel anyway, because it is less total kcmil of metal, and because smaller wires are more flexible.
I put "adding up" in quotations, because it isn't necessarily a direct addition of each device's full load amperes. Some loads do have their operating currents simply add, but loads in general do not. There are load diversity factors based on the reality that not every device is turned on simultaneously. There are also continuous load factors that may apply, depending on the details of how long the loads are expected to operate continuously. A safety factor applied so thermal-based OCPD's don't nuisance trip.
I do not understand the physical basis for why the next-size-up rule even exists in the first place. It would seem that the OCPD should be the weak link, such that it trips before the wire exceeds its ampacity. Especially when the calculated load isn't a fixed sum of current-limited devices, but rather a weighted sum that could be exceeded in a rare situation of high demand user behavior. But it is in the NEC, and we get to take credit for it when convenient for us.