I'm not directly sure about CS2's backend, but VRAM can address a lot of optimization issues in games with many cosmetic skins. CS2 at least has a bit on the clutch due to 5v5 (primary cosmetic, secondary cosmetic, knife, glove cosmetic, agent model). At 10 players in a match and 5 skins per round, that's 50 unique skins in play during any match.

In CS2's case, it's a bit more of a minor of a concern Valve can quite easily calculate a "worst case scenario" for, let's say 300mb/skin. While the number is significantly less than this, if we estimate 300mb per skin and have 50 unique skins in play during a match.. that adds up to 15GB real quickly. And, of course, VRAM concerns more than just skins (map textures themselves are VRAM hogs). But at least Valve has this in a reasonable range with CS2.. Given that Valve and especially CS2 are known to be optimization kings, a VRAM shortage on an 8GB card would perform fine.. but you may see some "texture pop" especially when you pick up an item on a ground with a skin that hasn't loaded (texture streamed) its HQ model in yet.

Each of these skins are often based off of several up to 4k texture maps. In many cases, you'll see at least 5-6 images for things like normal maps, roughness maps, diffuse maps, metallic maps etc., The way it works is that the GPU stores these actively in VRAM and, with texture streaming, if there is a shortage it "streams" the ones it prioritizes most.

Now let's talk less optimized titles with cosmetics. This is especially true in a battle royale or MMO type games that may not have gotten as much optimization love, especially in the early stages of development. In those cases, many 2025 ultra quality cosmetics become a great performance concern. In some case, cosmetics are after-thoughts or pushed by a less tech savvy company who bought it after development. Or developers will develop a crazy beautiful prototype knowing that novice investors might not understand the impact on future monetization models (IE: we used all of our budget and while there are solutions to it, it probably be a step-down for the currently the next-gen looking presentation that currently looks a step ahead of our competitors for a reason. IE: too good to be true and the same tradeoffs have to be made that other developers have made to make it work).

I like to compare it to a budget. A novice developer may see 8GB of VRAM as their early development target and inch that boundary as it will give them the best results if everything remains the same.. Only to then have a second team jump in and add a bunch of pretty cosmetics and not realizing that the original developers already were edging the 8GB limit to conform to hardware standards. Now all of a sudden, instead of a 12 hero limits across 60 players where the same model is shared across about ~5 players.. with even just 5 unique skins for each hero you're already hitting 5x more skins being potentially loaded in at once (all 60 players can use a unique skin). And that's not even including weapon skins etc.,

Again, solutions like texture streaming always make it playable.. but now you're implying that the VRAM budget that's already borderline 8GB is being cut into and something else has to be sacrificed. Let's add other novice mistakes on top of that: if the max VRAM and player models are prioritized over environment, which some teams may prioritize as "you don't want the things people are specifically buying to look worse" and further very likely for an early developer whose project lead initially misinformed them that "no, these are the only skins we'll ever have in the game".. you start experiencing another project that could effectively become "unfixable" compared to the original, on-release performance. So yes, neat back-end optimization can help address an issue that begins to occur because of it, but it probably will never look or perform as barebones as t original pre-cosmetic prototypes.

Again, you'll probably will never see this with a Valve title as they are the kings of that model and understood the limitations fully.. but if you start playing some Early Access Battle Royale titles or other free titles with cosmetics that VRAM is a massive game changer especially as we see new releases conforming to what 3D modelers often refer to as 2k/4k textures (In games with them we'd call this 'ultra quality', but ultra quality is less of a specification vs. a word synonymous with "the maximum for this specific title" and ultra can and often is even just 1k textures).

So imagine a perfectly optimized early-prototype of a title with 12 default skins across 12 classes/hero. Now imagine 60 players jumping out of a spawn ship, in 60 different cosmetic skins, with their uniquely skinned gear on their back etc., That's especially where VRAM shortages start really kicking you in the butt.. and if a game developer is pushing their beautiful prototype that looks generations ahead of existing, successful titles.. it's often not a good thing but instead a bad thing as they've probably underestimated how massive of a trade-off cosmetics will eventually cause - especially if they're a hero title pushing player-model qualities beyond what any other company has ever succeeded at.