No current-gen game utilizes 24-32GB VRAM, not even the most unoptimized one.
These cards are made for AI first and foremost, not gaming (that's a bonus).
I never said they should target 1440p.
Even RTX 4070 (1440p GPU) shouldn't necessarily target native 1440p in DLSS-enabled games...
Native 720p upscaled to 1440p (DLSS4 Performance) yields pretty decent quality and it reduces VRAM pressure quite a bit.
x60 GPUs should render at native 540p and upscale to 1080p via DLSS4.
With DLSS4 there's no need for Quality preset (unlike in the DLSS3 era).
PS5 has 12.5GB of available unified memory for both the CPU and GPU. There's no way the GPU utilizes 12.5GB of memory. More like 8-10GB at most (the CPU also needs to store data).
Also, keep in mind OG PS5 does not support AI upscaling (let alone techniques such as ray reconstruction), which means the framebuffer will consume more memory.
Regarding DirectStorage, it was advertised by Microsoft to reduce VRAM usage. Same with Sony's proprietary SSD API.
If consoles didn't have this special API and they still utilized an HDD, they would need 32GB of RAM for caching (there's a reason the PS4 had an enormous 8GB pool back in 2013).
Believe it or not, DirectStorage makes a huge difference in terms of reducing bottlenecks when properly optimized (remember when DX12 first appeared 10 years ago? it performed worse than DX11, but it's clearly a superior API):
Ideally you want the data (textures etc.) to go straight from the NVMe SSD to the GPU via PCIe, without using the CPU RAM as temporary storage/cache (nor taxing the CPU with decompression).
Imagine having a car and wanting to go from point A to point B: you want to follow the shortest route, not the longest one (more gasoline consumption)!