AMD and Nvidia speaking up native AI for customers might be unhealthy information for these ChatGPT subscriptions however excellent news for us PC players

It is all a bit doom and gloom within the PC gaming market currently, what with reminiscence shortages and relentless discuss AI above all else. So why not put our idealist hats on for a bit and see that tumbler as half-full? I will assist us get began: There appear to be the slight rumblings of a give attention to native reasonably than cloud AI by the massive {hardware} firms, which may spell excellent news for the PC gaming trade and, dare I say it, the reminiscence market.

Primarily, these rumblings have come from AMD’s CES 2026 keynote, however there’s additionally a glimmer of hope granted by Nvidia, too. On the previous entrance, AMD CEO Dr. Lisa Su really spent a good period of time speaking about native AI.

Admittedly, a lot of this was within the type of bigging up the corporate’s new cellular chips, the Ryzen AI 400-series, which is basically the Strix Level 300-series with some clock velocity bumps. However that is to be anticipated, and hardly goes in opposition to my level: these firms nonetheless need individuals to be utilizing units—ideally ones with their silicon—and if AI is all the trend, native AI on units is a giant promoting level.

Since then, we have seen a large growth within the cloud AI trade, which now has a whole bunch of billions of {dollars} wrapped up in it. And Su sees, or maybe hopes, that it’ll proceed booming, exponentially: “To allow AI in every single place, we have to improve the world’s compute capability one other 100 instances over the following few years, to greater than 10 yottaflops over the following 5 years.” That, she tells us, is 10,000 instances extra compute than we had in 2022.

Nvidia CEO Jensen Huang on stage at CES 2026.

(Picture credit score: Nvidia)

CES 2026

(Picture credit score: Future)

Meet up with CES 2026: We’re on the bottom in sunny Las Vegas overlaying all the most recent bulletins from among the largest names in tech, together with Nvidia, AMD, Intel, Asus, Razer, MSI and extra.

Su additionally highlighted the software program facet of the equation, by bringing on-stage Ramin Hasani, cofounder and CEO of Liquid AI. Liquid’s entire objective is to make AI that’s “processor-optimised” and scales down nicely to on-device, native AI. Its present mannequin, LFM2.5, makes use of simply 1.2 billion parameters, and in response to Hasani, does higher than DeepSeek and Gemini Professional 2.5 on-device.

“The objective”, Hasani says, “is to considerably cut back the computational price of intelligence from first rules, with out sacrificing high quality. Which means liquid fashions ship frontier mannequin high quality proper on a tool.”

Native AI has had much less particular airtime over the previous couple of days from Nvidia—except we’re counting DLSS 4.5, that’s—however there’s nonetheless been one thing to kindle our hopes. Specifically, one particular slide for the GeForce On Neighborhood Replace which states: “PC fashions closing the hole with cloud.” The graph similar to this textual content would not record any precise metrics, nevertheless it’s exhibiting native fashions seeming to converge on cloud fashions.

Three charts from Nvidia, one of which shows the seeming future convergence of PC and cloud AI models.

(Picture credit score: Nvidia)

The elephant within the room right here is that that is presumably all referring to cloud-based AI subscriptions and AI inference, not AI coaching, which can nonetheless depart an enormous bulk of labor for all these big servers. It is sort of a cake-and-eat-it state of affairs for Nvidia, with all of the cloud firms utilizing its GPUs within the datacentre and its shopper graphics chips being completely fitted to accelerating native AI workloads.

Nonetheless, that there is at the least some give attention to native AI might be optimistic for us puny customers. Particularly as a result of working AI fashions domestically requires reminiscence, both on the system facet, à la Strix Halo, or with a ton of VRAM, corresponding to on the RTX 5090.

So, if these firms need us working native AI they’d higher put the stress on these reminiscence suppliers to set some apart for us particular person customers. Yeah, I am not holding my breath both. I will take my idealist hat off, now.

Secretlab Titan Evo gaming chair in Royal colouring, on a white background

Finest PC gaming equipment 2026

Supply hyperlink