Nvidia's "Rubin" architecture coming soon: A new era of AI computing power
一、Official releases and core information
1.Launch event
Nvidia CEO Jen-Hsun Huang will officially unveil the next generation AI chip architecture "Rubin" at the GTC Developer Conference on March 18, 2025 (March 19, Beijing time). The architecture was originally scheduled to be released in 2026, but mass production has been brought forward to the second half of 2025 in response to challenges from competitors such as AMD and Amazon.
2.Technical specifications
Process & packaging : The Rubin-architecture GPU will use TSMC's 3nm process with next-generation HBM4 high-bandwidth memory supporting 8 to 12 HBM4 stacks .
supporting CPU : The next-generation CPU architecture supporting Rubin is named "Vera", and the two will form a super chip platform to achieve 3600GB/s ultra-high bandwidth through the sixth generation NVLink interconnection technology.
二、performance improvement and market layout
1.Performance leapfrog upgrade
reasoning ability : Rubin's AI reasoning (model generation output) performance is expected to be 30 times better than that of the previous generation Blackwell architecture, continuing the Blackwell upgrade of the previous generation .
Thermal optimization : Despite delays in mass production of the Blackwell architecture due to thermal issues, Nvidia accelerated Rubin research and development to solidify its 80-90% monopoly in the data center AI chip market .
2，Supply Chain and mass Production planning
TSMC cooperation : TSMC plans to increase CoWoS advanced packaging capacity to 80,000 units per month in the fourth quarter of 2025 to meet Rubin chip demand .
Core suppliers : Hon Hai, Wistron and other manufacturers have joined R100 AI server research and development, Rubin's first product R100 is expected to be put into production by the end of 2025 .
三. Industry influence and competition pattern
1.market demand driven
cloud giant demand : Companies like Microsoft and OpenAI have begun discussing Rubin's application potential, and Morgan Stanley and others believe it could be a key catalyst for Nvidia's stock to break above $200 .
market size : The global AI chip market is expected to reach $150 billion by 2025, and Rubin's early release could further expand Nvidia's lead .
2.competitive pressure
Rubin's early debut is seen as a key move to keep Nvidia's technology generation down in the face of competition from AMD's MI300 series and Amazon's own chips by accelerating its iteration strategy (update once a year) .
四、Future outlook
1.The Technology roadmap
Nvidia has planned product lines through 2027: Blackwell Ultra in 2025, Rubin in 2026, Rubin Ultra in 2027, continuing to update the architecture at a generation per year pace .
2.Ecological synergy
The Rubin platform will support CX9 SuperNIC (1600Gb/s bandwidth) and X1600 switches to build a full-stack, chip-to-network AI computing solution.
