NVIDIA Rolls GeForce Gaming Products into “Edge Computing” Category for Financials, Now Attributes to Less than 8% of Overall Revenue

The way world model LLMs are being developed, pretty soon games could be nothing but a collection of prompts describing every scene or level. And most of the creative people in game studios will need to become scene description writers.
 
nVidia quit well before RAM shot up. In fact, I’d argue nVidia quit PC gaming when they rolled out the RTX branding

nvidia cancelled the 50 series super versions and postponed the next gen because of the RAM crisis. intel cancelled gaming gpus for the same reason and AMD is still out of the limb on whether PS6/Helix will even come out in 2027, the SteamBox is rumored to be postponed/cancelled already and RDNA5 in '27? don't count on it.
 
The way world model LLMs are being developed, pretty soon games could be nothing but a collection of prompts describing every scene or level. And most of the creative people in game studios will need to become scene description writers.

Ever since RT and DLSS came out, I was sure AI would replace RT eventually. First, we got hybrid raster/raytracing, more recently hybrid raster/RT/AI, soon we'll have hybrid path tracing/AI and maybe in a generation or two, full AI. You just can't do brute force rendering anymore.
 
60% shrink you say...

Google and ChatGPT beg to differ, but don't take my word on it, do a quick search.
Great, then why didn't you post what you asked them and the answers?

60% was just an example to demonstrate that you can sharply increase market share while still loosing revenue if the market shrinks. I didn't mean the gaming GPU market actually shrank by 60%, although it wouldn't surprise me at least in units shipped, since gross revenue is propped up by exuberant prices.

When nvidia last reported gaming GPU revenue separately it was already down 13% from the previous quarter. With no new hardware and the prices being what they are it is a safe bet that the trend continued to this latest quarter.

It's highly unlikely for gaming gpu sales to be up, when the overall DIY market is at its lowest point. Here is a snippet from a real article, not conjured up by AI (at least I hope, who knows these days):

Desktop CPU sales have fallen to record lows, signalling a sharp slowdown in the DIY PC market as rising component costs push consumers to delay upgrades.
New data from German retailer Mindfactory for week 13 of 2026 shows CPU sales hitting their lowest level on record, according to figures shared by TechEpiphany.
The decline reflects broader trends across PC components, with prices for CPUs, GPUs, memory and storage all climbing in recent months.
 
The way world model LLMs are being developed, pretty soon games could be nothing but a collection of prompts describing every scene or level. And most of the creative people in game studios will need to become scene description writers.
The way how AI works currently this is impossible. Unless you want the same room to look completely differently each time you go back to it. AI video generators are barely able to keep coherence within one inference. Barely meaning with tons of tricks and hooks you might be able to limit the decay to where it is not immediately noticable.
 
The way how AI works currently this is impossible.
It's in its infancy but improving rapidly.


 
It's in its infancy but improving rapidly.


Retaining consistency for minutes - well that's tiktok covered :D

But seriously even if it can be improved to the point where every inference gives 100% consistent results forever, the compute cost is too much. It's much more practical and feasible to generate a static world using AI that then can be interacted with in the traditional way.
 
It's highly unlikely for gaming gpu sales to be up, when the overall DIY market is at its lowest point. Here is a snippet from a real article, not conjured up by AI (at least I hope, who knows these days):

The article is for Desktop CPU sales, not gaming GPUs, apples to oranges. It claims record lows but doesn't mention how much (I'm willing to bet not anywhere near 60%), also it's also from a german retailer, not worldwide sales.

On a side note, MindFactory tends to sell lots of AMD cpus and gpus, easily eclipsing intel and nvidia respectively. Maybe germans know something we don't.
 
Maybe germans know something we don't.

I guess many of them just grew up seeing AMD laptops/desktops around them and they also appreciate the jobs AMD created when they had a fab in Dresden. I remember reading that they supported AMD staunchly during their worst Bulldozer era too.
 
60% shrink you say...

Google and ChatGPT beg to differ, but don't take my word on it, do a quick search.
The AI/Datacenter market is crazy right now.

It races until it hits the next constraint: Chips, Power, Water, whatever hits next.

I read somewhere that while nVidia has sold a metric buttload of AI chips, only a handful of those are actually installed and working - the rest are sitting in warehouses, because they hit some constraint to allow them to be run.

Everyone is just kinda hording assets. Working in the Power industry, I'm seeing it right now with turbines and engines. I'm betting it's the same deal with RAM and whatever else is being jacked over right now.

There are something like 85 major data centers in planning around the US right now. These are massive - like... use more power than the entire rest of the state they are being located in massive. Take a peek at Stratos, Stargate, Hyperion: https://orennia.com/insights/the-20-largest-data-centers-being-developed-in-north-america Just add those top 20 projects up, and it just about doubles the entire electrical production of the entire rest of the United States.

My opinion - I think there are really only about 5 or 10 major projects of this scale, and they are being shopped around to find the most lucrative/exploitative location for installation - so you see a lot of planning and permits and applications, but there aren't really that many projects that could possibly go online in the next 2-5 years. The US couldn't support all that are in the planning stage right now, it's going to keep hitting the next constraint if we keep trying to develop at this scale.
 
Last edited:
Become a Patron!
Back
Top