Nvidia’s once-tight bond with players is cracking over AI

For its first 30 years, Nvidia wasn’t a family title except you have been a gamer. Now, a few of its unique fan base really feel left behind as synthetic intelligence has made the chipmaker the world’s most respected firm. 

“The gaming section is now not the driving drive of the corporate. There was one level when it clearly was,” mentioned Stacy Rasgon of Bernstein Analysis.

Nvidia popularized the graphics processing models, or GPUs, that allow quick body charges and rendering that make one of the best online game play doable. 

When Nvidia launched its first GPU in 1999, the GeForce 256, it laid off the vast majority of staff and approached chapter to make it occur. Players snapped up the brand new sort of processor, bringing Nvidia again from the brink.

Now, with demand for AI hovering, almost all of Nvidia’s income comes from its merchandise that serve that business, as a substitute of gaming. And as AI chipmaking shrinks the accessible reminiscence provide, Nvidia has been compelled to make robust choices about priorities.

In a memory-constrained actuality, it is not surprising that Nvidia would prioritize its much more worthwhile information middle GPUs corresponding to Hopper and Blackwell.

Nvidia’s working margins in its compute and networking section averaged 69% over the previous three years, in comparison with a 40% margin for the consumer-forward graphics section.

“I perceive that they’ll chase that. And that breaks my coronary heart,” mentioned Greg Miller, co-founder and host of in style online game podcast Kinda Humorous Video games Every day in an interview with CNBC.

“Dance with the one who introduced you. Players have introduced you this far,” Miller added.

If analyst predictions are right, 2026 would be the first yr in three many years that Nvidia does not launch a brand new technology of its consumer-facing GeForce line of graphics processing models.

Players are “vastly vital” to Nvidia, in line with an e-mail the corporate despatched to CNBC, including that it is “all the time innovating, testing and releasing” new gaming-focused applied sciences.

The present RTX 50 collection of GeForce GPU was unveiled at CES in January 2025. 

However with 2026 CES and GTC within the rearview mirror, some fear this would be the first yr with no new technology, though Nvidia does generally reveal new {hardware} as late as September.

Whereas it represents a giant technique pivot, some players say it is not a foul transfer for his or her budgets. 

“It is form of arduous to maintain up. You possibly can’t improve each single yr, so having a little bit of a break and ready for a technology to actually matter I feel is definitely in service of the players on the market,” mentioned Tim Gettys, Miller’s co-founder of Kinda Humorous Video games.

Nvidia’s present period of AI dominance began twenty years in the past with the 2006 launch of its CUDA software program toolkit. Abruptly, builders may use GPUs for general-purpose computing as a substitute of simply graphics.

Then, in 2012, Nvidia’s deep studying capabilities have been made clear throughout what many take into account the large bang second for contemporary AI. Nvidia’s GPUs and CUDA have been used to construct a neural engine known as AlexNet that blew away the competitors throughout a distinguished picture recognition contest.

Though Nvidia did not cease making gaming GPUs, it signaled a brand new give attention to GPUs for AI in 2020 when it bought high-performance computing chipmaker Mellanox Applied sciences for $7 billion.

The corporate has been releasing new generations of high-end GPUs ever since, together with full rack-scale methods for AI workloads corresponding to the brand new Vera Rubin platform, which CNBC acquired an unique first look at in February.

Nvidia does not reveal costs for its AI chips, however analysts say one Blackwell GPU prices as much as $40,000, whereas the Futurum Group estimates a full Vera Rubin system will value as much as $4 million.

In distinction, Nvidia sells its RTX 50-series gaming GPUs for between $299 to $1,999.

In the course of the cryptocurrency peaks of 2018 and 2021, Nvidia’s GPUs offered in on-line marketplaces for as much as thrice itemizing worth as a result of they have been as soon as key to mining Bitcoin and Ethereum. 

Though costs fell when mining modified course in 2022, Nvidia’s present RTX 5090 GPU continues to be offered on-line for as much as double the retail worth. 

Loads of demand for final yr’s technology could make Nvidia much less motivated to place out a brand new model this yr.

However the reminiscence scarcity is a extra possible offender for Nvidia’s gaming downside.

Business reviews recommend Nvidia has made plans to cut back manufacturing of its newest gaming GPUs by as much as 40% because it faces a serious scarcity of the general-purpose reminiscence that is crucial for making a GPU.

Dynamic Random Entry Reminiscence, or DRAM, permits quick, short-term information storage so the GPU can run parallel duties.

Private computer systems, the place Nvidia’s gaming GPUs find yourself, have borne the brunt of DRAM shortages. When reminiscence costs go up, manufacturing a GPU prices extra, and that value trickles right down to shoppers.

Gartner predicts PC costs will rise by 17% this yr, inflicting PC shipments to say no 10.4%.

“With how costly all of this has gotten, it is regarding to see costs go up on the gaming facet with no indicators of ever coming again down, after which Nvidia clearly chasing a totally completely different class of shopper,” Gettys mentioned. 

If the entry-level shopper PC market disappears by 2028 as Gartner predicts, the marketplace for Nvidia’s entry-level gaming GPUs is prone to contract, too. 

As a substitute, Nvidia is probably going saving restricted reminiscence stock for its greater value, greater margin AI chips. 

“If there may be push-outs or delays on the gaming roadmap, it is most likely largely that they most likely cannot make the playing cards in any case as a result of it is arduous to get the reminiscence,” Rasgon mentioned. “Each little bit of reminiscence that is on the market, I feel is absolutely getting prioritized to AI compute.”

Greater-performance GPUs like Blackwell and Rubin are lined with dense stacks of a selected sort of DRAM often called Excessive Bandwidth Reminiscence, or HBM. Rasgon mentioned it takes about 4 instances as many silicon wafers to make a gigabyte of HBM because it does to make the identical quantity of extra conventional forms of DRAM.

“That dynamic is ravenous the general business of the kind of reminiscence that’s historically used for extra shopper sort purposes. It is simply not accessible,” Rasgon mentioned.

Nvidia informed CNBC that it is persevering with to ship all GeForce GPUs because it sees sturdy demand, and is working intently with suppliers to maximise reminiscence availability.

“In the event that they’re making thrice the cash and the stockholders are thrice happier, then yeah, I do suppose that they are going to abandon gaming regardless of it being what acquired them there,” Gettys mentioned.

‘Seems like a slap within the face’

CEO Jensen Huang did make a giant gaming announcement at first of his keynote handle at Nvidia’s annual GTC convention in March, however the gaming group was lower than enthused. 

Huang introduced the following technology of its rendering software program known as Deep Studying Tremendous Sampling or DLSS, coming within the fall. It is well-known for enhancing body charges by rendering video games at decrease resolutions and utilizing AI to scale up the picture, serving to video games run extra easily on much less highly effective {hardware}.

The controversy with the brand new DLSS 5 is that players fear it makes use of generative AI to vary the look of the sport. Huang unveiled DLSS 5 with a sizzle reel of photorealistically enhanced variations of characters in in style video games corresponding to Resident Evil Requiem, Starfield, and Hogwarts Legacy.

“I play video video games as a result of they’re an artwork kind. And so I prefer to see the thumbprint of the creator in what I am doing,” mentioned Miller of Kinda Humorous Video games. “That raised numerous hair on numerous necks within the online game business as we take care of so many layoffs, so many studio closures.”

Because it grapples with a post-pandemic slowdown, the gaming business has seen studio closures, canceled video games, and 1000’s of job cuts throughout giants like Epic Video games, Microsoft’s Xbox, and Sony’s PlayStation.

Gettys was a fan of earlier variations of Nvidia’s DLSS for making gaming extra accessible on a decrease finances.

“The know-how is mind-blowing for what it might probably do to make video games run on lower-end PCs,” he mentioned. “However then so as to add this generative AI stuff, it appears like a slap within the face.”

Gettys’ large worry is that this can be a step towards absolutely AI-generated video games, which he thinks is “100% the objective.”

Elon Musk has already addressed the potential for it. In an October put up on X, Musk mentioned his xAI recreation studio will launch “an ideal AI-generated recreation” earlier than the tip of 2026.

“You are actually altering the artwork created by the builders. After which at a sure level you are changing the builders after which their studio will get closed down,” Gettys mentioned.

Nvidia mentioned in an announcement to CNBC, “Video games are a artistic artform that give builders the chance to inform partaking tales and immerse gamers in unbelievable worlds. Our RTX applied sciences are instruments that allow recreation builders to realize their artistic imaginative and prescient – these embody rendering strategies corresponding to ray tracing and path tracing, and people enhanced by AI, like DLSS Tremendous Decision, DLSS Body Technology, and DLSS 5, all working collectively to offer one of the best efficiency and picture high quality.”

Throughout his GTC keynote, Huang mentioned AI goes to “revolutionize how pc graphics is finished.” 

In a question-and-answer session the following day, Huang responded to assertions from the gaming group that DLSS 5 makes video games seem homogeneous.

“They’re utterly flawed,” Huang mentioned.

He emphasised that recreation builders will nonetheless be in management, in a position to “fine-tune the generative AI” to match their fashion.

For over a decade, Nvidia has additionally provided gaming within the cloud via a service known as GeForce NOW. The mannequin has advanced to incorporate completely different subscription tiers — together with a free choice — that lets customers stream video games they personal on providers like Steam, operating on Nvidia GPUs in information facilities, reasonably than on private units. 

“You see XBox and also you see PlayStation, you see different rivals attempting to get the cloud into players’ palms in a manner that really is smart. And Nvidia GeForce NOW has actually cracked that code,” Miller mentioned.

Gettys informed CNBC that Nvidia’s streaming platform is one of the best “by a landslide.”

“It permits thousands and thousands extra folks entry to gaming on the highest degree, even when they do not have the newest playing cards and all of that. And it is actually unbelievable know-how,” he mentioned.

Superior Micro Units is Nvidia’s high competitor in gaming, with its Radeon line of GPUs.

However the reminiscence crunch stays a problem for each.

“If Nvidia cannot get the reminiscence, AMD ain’t going to get the reminiscence,” Rasgon mentioned. “Sentiment sensible, each manufacturers have their followers and they are often die arduous.”

“There is a clear favourite,” Gettys mentioned. “In case you’re enjoying on PC, you are going to need an Nvidia card.”

Watch: How AMD turned a chip large and at last caught Intel