That confused me as well. The Mode 13h graphics framebuffer is at address A000:0000. How does this program get that value?
The key is the LES AX,[BX] opcode. BX is initialized to 0 by DOS, and the "Set video mode" BIOS call preserves BX's value. So the LES opcode sets ES:AX by reading a dword from DS:0.
What's there? A COM file is a single-segment program, so DS equals CS, and the code segment starts with the Program Segment Prefix. (The actual code is loaded at offset 100h.) So it loads the first two words of the PSP. What are those?
The first word is an "INT 20h" instruction, 20CDh, for compatibility with CP/M.
The second word is the segment number of the end of the memory allocated for the program. But DOS always allocates all memory to COM programs, so this will be 0x9FFF (assuming you have a full 640K conventional memory installed).
So ES:AX is set to 9FFF:20CD. And with x86 segmented memory 9FFF:0010 equals A000:0000.
Writing "demo" scene is like writing "web" site: come on, this has been around since the 80s. We literally have a post on the front page about demoscene being a recognised cultural heritage now.
As alasdair_ mentioned, this is some impressive demoscene stuff. Which, if I may take the liberty, is all about maximal creative output (subjective, I know) with minimal data footprint. In case you missed in original post, a video of the demo:
My brain could not admit it was 32b and insisted on parsing it as 32kb anytime it was displayed. It's when I saw the demo, expected more, then go wtf!?...