NVIDIA’s Rubin AI Architecture Update: Team Green Maintains Annual Release Schedule, Outpacing AMD’s Instinct Series

[Update – 8/14/25] – NVIDIA’s Rubin architecture and associated GPUs are progressing as planned, aligning with the company’s annual release schedule.

[Initial Report / Clarified] The anticipated launch of NVIDIA’s next-generation Rubin graphics cards may be postponed by several months due to a redesign aimed at enhancing competitiveness against AMD.

NVIDIA’s Rubin Redesign: Impact on Production and Competitive Landscape

The evolving landscape of AI hardware is projected to undergo significant transformations, particularly regarding market share dynamics between NVIDIA and AMD. Both corporations are in a fierce contest to develop cutting-edge technology. However, NVIDIA has recognized an unexpected surge in competitive pressure, prompting the decision to redesign its Rubin lineup. According to insights from Fubon Research, a notable banking institution in Taiwan, this redesign is expected to result in limited production volumes in 2026, extending the expected release window by approximately four to six months.

1) Rubin redesign leads to more limited volume2) TSMC CoWoS capacity to hit 130k in 20273) Blackwell Volume: 750k in Q1 25, 1.2mn in Q2, 1.5mn in Q3, 1.6mn in Q44) Broadcom is fastest growing CoWoS customer in 2026Source: Fubon$tsm $nvda $avgo pic.twitter.com/ckoZswOCkI

— Nomad Semi (@MooreMorrisSemi) August 13, 2025

While specific details of the redesign remain undisclosed, NVIDIA’s strategy appears to be a reaction to AMD’s anticipated release of the Instinct MI450 AI series. This new lineup is rumored to introduce configurations of IF64 and IF128, which would directly compete with the capabilities of the Vera Rubin series. Particularly noteworthy is AMD’s focus on high-bandwidth memory (HBM), potentially offering up to 432 GB per GPU. This could compel NVIDIA to innovate further on the Rubin configurations, although this remains speculative at this stage. Additionally, AMD’s adoption of Infinity Fabric over Ethernet enhances design modularity, facilitating more efficient implementation in data centers.

TSMC & SK hynix Forms a Strategic AI Alliance, Fueling HBM4 For Next-Gen NVIDIA & AMD GPUs 1

It’s anticipated that the redesign of the Rubin architecture will lead to reduced shipment volumes in 2026, reminiscent of the current constraints faced by the Blackwell Ultra series. This timeline aligns closely with the expected market entry of AMD’s MI450, suggesting that the launch of NVIDIA’s Rubin Ultra might also face postponement. Although the specific reasons behind the delay have not been clarified, it appears that NVIDIA’s aggressive product development cycle has contributed to these challenges, a topic we have explored in greater detail previously.

Source & Images