Up until recently they have simply broken the image up into rectangles and only kept enough visual data in each block to convey to image without so much high frequency noise. They have also used frame which only remember differences from the last frame with full frames called Iframes are intermittent.
Now with the advent of deep learning NVidia has developed a codec that takes one full frame then just changes the data of the head position and facial expressions which can work better on weaker bandwidths for video conferencing.
So where do we go in the future.
well what you will have will be a video codec that utilizes low detail changes, physics and other data in order for the AI to generate a broadcast quality video that could be up to 1/5 the the data size of the equivalent AV1.
So what does that mean.
Well
you could have about 120 good mpeg4 SD streams on the UHF and VHF Bands. 2Mbit.
you could have about 170 good AV1 720p streams on the UHF and VHF Bands. 1.6Mbit.
you could have about 850 good next gen 720p streams on the UHF and VHF Bands 300Kbit.
Properly distributed this could lead to something like 200 radio stations, 80 2K channels and 500 720p channels over terrestrial television bands.
Video upscaling is destined to improve the situation also.
This level of data efficiency would mean that 10 years from now using some next gen codec you could fit a youtube like site up in a 200PB stack. When you consider there will be SSDs offering 1PB for a couple of thousand dollars you may see that with a couple of million you could have a system like Youtube especially when you consider it won't be the size of Youtube at first until enough stuff is submitted on it.
As circuit stacking becomes more viable and fabrication becomes more precise. 80PB SSDs might not be out the question for a couple of thou by 20 years. meaning that you could make a youtube like site for under 100,000 dollars and for the same price you could probably do ok with such a service in the cloud.
Today it would cost you more like 200 million to 1 billion to offer a youtube like service.
This means youtube has to evolve to keep up especially by 10 - 20 years from now.