Pictures and frame buffers. If you're running fullscreen at 4k, you're probably gonna take two roughly 3840x2160x3 byte frame buffers for the window. Your designers want a background image which moves as you scroll in some parallax style; that's over 3840x2160x3 bytes more for the pixel buffer backing the image layer. And let's say roughly 50% of your screen is text with subpixel (aka full color) anti aliasing; that means another 3840x2160x3x0.5 bytes for the pre rendered text.

3840x2160x3x3.5. That's 87MB, in pixel data only. And it's a very minimal example; for the parallax image, you're gonna want the image to be significantly taller than the window; you're gonna want a ton of smaller (tho still high DPI) images for icons; a few different font atlases for different font faces you've loaded at once; maybe pre rendered pixel buffers for all sorts of UI components; etc.

And lord help you if your designers want any part of this to be animated.

(I'm playing a bit fast and loose with what lives on the GPU and what lives on the CPU here. On many systems, they share a memory pool anyway. But on systems with discrete GPUs, most of this is gonna be video memory. Though applications may wanna store CPU-side copies as well for various reasons.)