I feel like the easiest way to think of this is to have a grid of square gradients and when you use a greater than, you can change the size of the square. Think of it as changing the height that you cut through a pyramid, at the top the cross section would be small but at the bottom, it'll be large.
To make your grid of squares we can do a little bit of vector maths. I'll be using the UV coordinates as this will let you unwrap your cube so each face gets the whole effect.
I am scaling the UV coordinates 20x so that we get 20 squares and then breaking it into a grid with Fraction. I then centre each space by subtracting 0.5 and use absolute to make each space symmetrical. This is the easiest way to get a square grid. I then separateXYZ and use a maximum node in order to get our little pyramids for each square.
If we add a math node set to Less Than at the end, you'll see the threshold will change the size of all the squares. To make this wipe down like your animation, we need the threshold value to be a gradient.
Branch off from the Scale node and add a separateXYZ followed by a math node set to snap. Take the Y socket into the math node and set the increment to 1 (this is to match the fraction node which also repeats at 1).
We now have a gradient and that we can manipulate to control the animation. To change the position of the gradient, add a math node set to subtract, and then follow that with another math node, set to multiply. The multiply will act as the falloff control. Use a small number like 0.1 for now. If we plug this into the threshold of the Less Than node, you'll see the square gradient how you want. You can change the subtract value to move the transition up and down.
Finally, to animate the sequence, we need to bring the frame number into the shader. Add a value node and type #frame into the value field. This will create a driver that takes the frame number. We want to have a loop so we're going to add a math node set to modulo. We need to use the loop length twice so add another value node that we can use and plug this into the bottom of the modulo. Set the loop length value to something like 80. Follow the modulo with a map range and plug the modulo into the first value and the loop length value node into the From Max socket. Set the from Min socket to 1. Our To Min and To Max socket slightly depends on a few things such as what scale value we've used and what the falloff is set to and the loop length so the settings for a scale of 20 and falloff of 0.1 are:
To Min: 20
To Max: -5
Plug this map range into the bottom of the subtract node to control the position of the wipe.