NVIDIA's next-generation AI platform, Vera Rubin, has officially entered full production.
Jensen Huang, CEO of NVIDIA, announced on June 1 at the GTC Taipei 2026 event that "Vera Rubin is now in full production" and confirmed that the platform will incorporate memory from Samsung Electronics, SK Hynix, and Micron.
Vera Rubin is the successor to NVIDIA's Blackwell platform, which currently dominates the AI server market. Named after astronomer Vera Rubin, NVIDIA is developing a new AI superchip architecture that combines the GPU "Rubin" and CPU "Vera." It features next-generation HBM4 (high-bandwidth memory) and significantly enhances both AI training and inference performance.
According to NVIDIA, the Rubin platform offers much higher AI processing capabilities compared to previous generations and is designed for the era of AI agents and inference-based AI. Huang stated, "The AI agent market will grow thousands of times larger than it is now," forecasting explosive growth in the AI industry.
Notably, the presence of South Korean semiconductor companies has drawn attention during this announcement. Samsung Electronics and SK Hynix are being highlighted as key players in the next-generation HBM4 supply chain. Industry experts believe that as Vera Rubin enters mass production, demand for HBM will surge, benefiting South Korea's memory sector.
Indeed, Samsung has recently begun shipping HBM4E samples, and the market is closely watching NVIDIA's potential supply chain expansion. SK Hynix is also regarded as a critical supplier for the next-generation HBM4.
Vera Rubin is more than just an AI semiconductor. NVIDIA is pursuing a "vertically integrated AI ecosystem" that encompasses AI chips, networks, data centers, and AI agent software. Huang emphasized, "NVIDIA will integrate all stacks of agent AI," leading to increased growth expectations across data centers, power infrastructure, AI cloud, and the HBM memory industry.
Vera Rubin is the successor to NVIDIA's existing Blackwell platform. It is not just a GPU but a next-generation AI infrastructure platform that integrates CPU, GPU, network, storage, and AI software into a single massive AI supercomputer.
The platform includes core technologies such as NVIDIA Vera CPU, Rubin GPU, BlueField-4 DPU, Spectrum-6 Ethernet network, and NVLink 6 switch.
Notably, NVIDIA claims that Vera Rubin can enhance AI agent throughput by up to ten times compared to the previous generation, Grace Blackwell.
The AI industry anticipates that AI agents capable of planning and executing tasks will lead the next market wave, positioning Vera Rubin as dedicated infrastructure for such AI agents.
One of the most notable changes in this platform is its modular design.
Traditional AI servers often require complex installation and maintenance, leading to system shutdowns even for minor component issues.
In contrast, Vera Rubin is designed with replaceable major components, significantly reducing maintenance time. NVIDIA explained that its new design minimizes cables, cutting the installation process from about two hours to just five minutes.
Considering that a substantial portion of AI data center operating costs comes from maintenance and power expenses, this change is seen as a crucial improvement that enhances profitability rather than just convenience.
NVIDIA emphasizes that Vera Rubin is not merely a performance-enhancing product.
The AI industry is entering a phase where inference costs are becoming more critical than model training costs, as substantial computations are required for AI to deliver actual services.
Vera Rubin focuses on significantly reducing inference costs by utilizing ultra-fast data connection technologies and optical communication-based networks.
Huang stated, "AI agents perform thousands of steps of inference, tool usage, and information retrieval with a single request," explaining that "Vera Rubin is a platform designed for this new era of AI."
South Korean companies are also growing their expectations for benefits.
Vera Rubin is set to incorporate next-generation HBM (high-bandwidth memory) HBM4. Industry analysts believe that Samsung Electronics and SK Hynix will play significant roles in the supply chain.
As demand for HBM in AI servers continues to rise, the expansion of Vera Rubin's mass production could provide new growth opportunities for memory companies.
Particularly, as the AI agent market begins to grow in earnest, competition for data center construction is expected to intensify. Consequently, memory, server, network equipment, and cloud companies are all anticipated to benefit.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.