Probably just a video distribution/converter and a bunch of daisy-chained LED drivers?
I think there will be zero to none difference compared to how LED TV screens are done.
I know of similar installs that used a system to convert video data to pixel-mapped streaming ACN (transported over Ethernet), which was then converted to DMX to feed to groups of fixtures. That allows all of the distribution infrastructure to be built with off-the-shelf hardware, and you can even use off-the-shelf LED fixtures if you can find something that suits, plus sACN has a nice sync feature that allows receivers to temporally align the output of several DMX universes very simply. The downside is that sACN relies on IPv4 multicast, and many consumer/office switches simply can't handle more than a couple of multicast groups.
One of the challenges with installs like this is that not only is it a fair bit of data to throw around, it's spread out over a fairly large area, so it can be difficult to find the best way of fanning out the data to balance the cost of hardware against the cost of installation. mikeselectricstuff has a few videos on similar topics, although AFAIK he's never worked on an installation this physically large. (Maybe those snowflakes, but I don't think those required anywhere near the amount of data flying around as the building display in the linked article.)
I'd think that DMX could not do this job very well, unless you have a significant amount of universes. You know, that puking all 512 channels takes some significant time, over 22ms if I remember correctly. That would create significant visible lag a worsen refresh rate, if there will be too many channels used per single DMX universe.
You can't just fit too many RGB LEDs to a single DMS universe. Tens of DMX universes would be too low a number I'd guess. (Of course, depends on the scale of a specific installation, but just taking the building imaged above as a reference)
I think there may be some other proprietary solutions to video projection equipment, as they do exist for the typical LED screens.
That's why you use something like sACN to bundle up as many universes as you need over ethernet, and break it out to DMX closer to the final fixture. There are media servers that can directly emit pixel-mapped sACN, and boxes that convert up to eight universes of sACN to DMX, all available off-the-shelf. sACN supports up to 32K universes, so that's around 5.6Mpixels if you have the bandwidth (a full universe sACN packet is something like 7-800 bytes on the wire, so it's about 6Gbps).
You could certainly do something custom, but I'm not sure that you could do something sufficiently better to justify the additional development cost for a one-off project.
I do not really know, I am not from the lighting business, but DMX for this sounds quite strange. I would rather expect some vendors making stuff specifically for these strip building displays. These LED shows are not that uncommon in larger cities. However they (gov and buros) try to suppress them here as much as possible (by not allowing them to be installed), due to them being a clear distraction for drivers, and annoying for people living nearby (which is kind of expected, uh).
(They have even banned the use of static paper billboard ads near intercity roads! Which, by the way, I have absolutely no problem with).
What about some other low level light fixture communication standards? I've heard about Martin Light (or hows that spelled), have even got a USB converter for it from somewhere. But that's some closed-source proprietary thing I do not have any clue about.
(DMX on the other hand is pretty well known in the industry and even between hobbyists)
I am in the lighting business, and DMX is really it as far as low-level lighting protocols in common usage that can do a reasonably high channel count/refresh rate. There are other protocols in the architectural/building automation world, but AFAIK they tend to be fewer channels and/or lower refresh rate. Certainly in the entertainment lighting world DMX is the only low-level protocol in common use, but there are more options in network protocols. For open standards you have sACN (ANSI E1.17/E1.31) and Art-Net (privately developed, but royalty-free and openly documented). Just about every lighting control manufacturer also has their own network protocol, mainly used for sharing show file data between control consoles/NPUs, but some of them can also be used to transport DMX data to and from DMX I/O nodes.
Again, it is possible they're using something else, but I'm not aware of any other off-the-shelf solutions. Even if you were doing something from scratch for whatever reason, the same topology makes a lot of sense--IP network for large-scale distribution, and somewhere closer to the fixtures convert to something (probably RS485-based or similar) that's cheap and easy to implement on the thousands of small MCUs you need to run all of those LEDs.
If any part of the system is fully custom, I would guess that the fixtures are, if only because they need to be architecturally appropriate, but there are some off-the-shelf solutions that could be used as well. It really depends on the kind of effect you're after, the architectural considerations, environment, etc.
That would be one heck of an interesting project to orchestrate.
