Nvidia and AMD's next gaming GPUs could be delayed until 2028, because AI pays better

Connecting the dots: Nvidia's RTX 60 series Rubin GPUs and most of AMD's RDNA 5-based Radeon 10000 series gaming cards have reportedly been pushed back to 2028. The delay would come down to persistent AI demand, which continues to make data center hardware far more profitable than consumer products.

Responding to a forum question about the release timing for Nvidia and AMD's next-generation graphics cards, the generally reliable leaker Kepler_L2 said on the AnandTech forums that nearly the entire stack from both companies is now expected to arrive in 2028, as AI accelerator demand keeps eating into available capacity. The lone exception is AMD's "AT2" GPU, still reportedly on track to launch next year.

Kepler_L2's post appears to directly contradict a recent report from YouTube channel Moore's Law is Dead, which claimed Nvidia was gearing up to launch its next-gen RTX 60 series – including the flagship RTX 6090 – in the first half of 2027. Kepler_L2 dismissed the claim outright.

Speaking to Dutch outlet Tweakers in June, several of AMD's board partners said they had little clarity on the release window for the next-generation cards. Some expected shipments to begin by mid-to-late 2027; others said availability could slip into early 2028.

Neither Nvidia nor AMD has said much publicly about its next-generation gaming GPUs, but rumors point to the so-called AT2 as an RDNA 5-based die with 8 Shader Engines, each carrying 5 Compute Units, for 40 Compute Units in total. It's expected to power an upper-midrange card, possibly with 12GB of VRAM on a 192-bit memory bus.

Also check out: Cost Per Frame: Nvidia Stops Offering Good Value GPUs to Gamers – see what's the "best value" GPU in your region

AT2 could end up going up against Nvidia's current GeForce RTX 5080, a step up from AMD's existing flagship, the Radeon RX 9070 XT, which trades blows with the mid-range RTX 5070 Ti. All of this, though, comes from unverified third-party sources and should be treated as such.

Nvidia has already shown off its Vera Rubin "Superchip" for AI data centers, but has stayed quiet on the consumer side. AMD hasn't said much about RDNA 5 either, though an earlier Kepler_L2 leak claimed the flagship Navi 5X GPU could hit clock speeds of up to 3.4GHz, with 12,288 cores and 32GB of GDDR7 memory.