Nvidia has presented its next-generation artificial intelligence platform, Vera Rubin, a rack-scale system built from 1.3 million components and designed to deliver ten times more performance per watt than its predecessor, reports Baltimore Chronicle via CNBC. The system is scheduled to begin shipping in the second half of 2026 and is expected to play a central role in the company’s expanding AI infrastructure portfolio.
CNBC was granted exclusive access to Vera Rubin at Nvidia’s headquarters in Santa Clara, California. The company states that the platform represents a significant engineering shift at a time when power consumption has become one of the most pressing constraints in global AI infrastructure deployment.
At the core of Vera Rubin are 72 Rubin graphics processing units (GPUs) and 36 Vera central processing units (CPUs), with primary chip manufacturing handled by Taiwan Semiconductor Manufacturing Co. The system integrates components from more than 80 suppliers across at least 20 countries, including China, Vietnam, Thailand, Mexico, Israel and the United States. Beyond processors, the platform incorporates liquid cooling systems, advanced power delivery infrastructure and compute trays assembled into a single rack-scale architecture.
Nvidia says Vera Rubin will consume approximately twice as much electricity as its predecessor, Grace Blackwell. However, due to architectural and efficiency gains, the company claims the new platform delivers ten times greater performance per watt. The increase in efficiency is considered critical as hyperscale data centers expand to meet AI-driven demand while facing grid and sustainability constraints.
The company acknowledged ongoing supply chain pressures, particularly rising memory costs driven by global shortages linked to AI workloads. Dion Harris, Nvidia’s head of AI infrastructure, said the company has provided suppliers with detailed demand projections to ensure production targets are met. He stated that Nvidia remains aligned with its supply chain and positioned to fulfill shipments.
Grace Blackwell entered production in 2024 and significantly expanded compute density within a single rack. Vera Rubin builds on that architecture and is already in full production, according to Nvidia CEO Jensen Huang, who confirmed the milestone in January. Huang also discussed the platform publicly during a question-and-answer session at the Consumer Electronics Show in Las Vegas on January 6, 2026.
Industry analysts describe the system as a departure from traditional server construction. Daniel Newman of Futurum Group said the racks integrate compute, networking, cabling and cooling into a unified structure engineered for maximum efficiency and performance, contrasting with historically modular server approaches.
Each Vera Rubin rack weighs nearly two tons and contains approximately 1,300 microchips, compared with 864 in Grace Blackwell. The system is designed with modularity in mind: each superchip module can slide out from one of 18 compute trays within seconds, simplifying maintenance and replacement. In the Blackwell system, comparable components were soldered directly to boards.
Nvidia demonstrated the Vera Rubin superchip at its Santa Clara headquarters on February 13, 2026. The superchip integrates two Rubin GPUs and one Vera CPU and contains roughly 17,000 components. Vera Rubin is also Nvidia’s first fully liquid-cooled rack-scale system. According to Harris, this design enables data centers to consume significantly less water compared with traditional evaporative cooling systems.
While Nvidia does not publicly disclose rack pricing, Futurum Group estimates that Vera Rubin systems will cost approximately 25% more than Grace Blackwell, placing the projected price between $3.5 million and $4 million per rack.
Major technology companies are preparing to deploy the system. Meta announced plans to integrate Vera Rubin into its data centers by 2027. Nvidia has also identified OpenAI, Anthropic, Amazon, Google and Microsoft among expected customers. The racks are manufactured in the United States and other locations, including Taiwan and a new Foxconn facility in Mexico.
Despite Nvidia’s dominance in the AI processor market, competitive pressure is intensifying. Advanced Micro Devices is preparing to ship its first rack-scale AI system, Helios, later this year. AMD recently secured a substantial commitment from Meta for up to six gigawatts of capacity. In parallel, major cloud providers are expanding internal silicon initiatives. Amazon Web Services has deployed data centers equipped with ultra-servers powered by its Trainium 2 chips, while Google continues to scale infrastructure built around its proprietary tensor processing units.
Jordan Klein, an analyst at Mizuho Securities, said efficiency metrics remain central to purchasing decisions, particularly the number of tokens generated per unit of power consumed. He noted that incremental improvements in that ratio can materially increase returns on AI infrastructure investments.
Addressing competitive efforts, Harris said that developing systems of this scale and complexity is not a simple undertaking, emphasizing the technical barriers involved in rack-scale AI platform integration.
Earlier we wrote that US Stocks Slide After January PPI Surges and Block Unveils AI Workforce Overhaul
