AI Hardware Demand Pushes Nvidia and AMD Consumer GPUs Toward 2028

SANTA CLARA, Calif. — AMD’s sole reported exception to the delayed consumer graphics roadmap is the graphics processor codenamed “AT2.” The RDNA 5-based die remains on track for release next year, with unofficial technical details indicating 8 Shader Engines, each containing 5 Compute Units, for a total of 40 Compute Units.

The AT2 is intended for upper-midrange desktop graphics cards and is expected to carry 12GB of VRAM across a 192-bit memory bus interface. Its expected performance tier is occupied by Nvidia’s GeForce RTX 5080, one step above AMD’s existing flagship Radeon RX 9070 XT, which currently competes with Nvidia’s midrange GeForce RTX 5070 Ti.

A sustained surge in enterprise demand for artificial intelligence hardware is reshaping major chipmakers’ product roadmaps and stretching traditional consumer PC graphics hardware development and release schedules to historic lengths. Silicon allocation at leading semiconductor foundries, including Taiwan Semiconductor Manufacturing Co. (TSMC), has shifted heavily toward high-margin data center accelerators.

Enterprise platforms such as Nvidia Corp.’s Blackwell and Vera Rubin server architectures generate significantly higher profit margins per wafer than consumer graphics processing units (GPUs). Retail graphics card architectures from both Nvidia and Advanced Micro Devices Inc. (AMD) are consequently reportedly facing multi-year delays.

Earlier leaks from Kepler_L2 described AMD’s flagship Navi 5X processor with 12,288 stream processors, potential boost clock speeds of up to 3.4 GHz, and 32GB of high-speed GDDR7 memory. The higher-end RDNA 5 silicon therefore carries substantially larger reported core-count and memory specifications than the AT2 details.

Industry insider accounts posted by technology leaker Kepler_L2 to the AnandTech forums indicate that nearly the entire next-generation consumer GPU product stacks from Nvidia and AMD have been pushed to a 2028 release window. The revised timing disrupts the graphics hardware industry’s historic cadence, in which major architecture updates traditionally arrived every 20 to 24 months.

That reported delay covers Nvidia’s upcoming GeForce RTX 60 series, built on the “Rubin” microarchitecture, and AMD’s Radeon RX 10000 series, based on its next-generation RDNA 5 graphics architecture. It conflicts with an earlier report from YouTube technology channel Moore’s Law is Dead, which said Nvidia was preparing to launch its top-tier RTX 60 series, including a flagship GeForce RTX 6090, in the first half of 2027. Kepler_L2 dismissed those earlier H1 2027 claims as inaccurate.

Add-in board (AIB) manufacturing partners have also faced uncertainty over launch timing. In June, several AMD board partners told Dutch technology outlet Tweakers that they had received minimal official guidance from the chipmaker about RDNA 5 rollout dates. Some manufacturers expected initial shipments between mid- and late-2027, while others said retail availability could easily slip into early 2028.

Nvidia has publicly showcased its enterprise Vera Rubin “Superchip” architecture for AI data centers and cloud computing providers, but it has not publicly detailed its consumer GeForce roadmap for the Rubin family. AMD has likewise withheld official architectural specifications and formal product announcements for its RDNA 5 desktop gaming line.