Nvidia’s new AI system Vera Rubin is 10 times more efficient than its predecessor — here’s a first look

Nvidia is set to make waves in the tech industry with the introduction of its latest AI system, Vera Rubin, which is anticipated to launch later this year. This advanced system boasts a remarkable tenfold increase in performance per watt compared to its predecessor, Grace Blackwell, addressing one of the most pressing challenges in AI technology: energy efficiency. At its headquarters in Santa Clara, California, Nvidia provided an exclusive preview of Vera Rubin, which comprises an impressive 1.3 million components. The system integrates 72 Rubin graphics processing units (GPUs) and 36 Vera central processing units (CPUs), primarily sourced from Taiwan Semiconductor Manufacturing Co. The assembly draws from over 80 suppliers across 20 countries, including the U.S., China, Vietnam, and Mexico. Nvidia faces hurdles, particularly with rising memory costs due to a global shortage driven by increased demand for AI products. Dion Harris, the head of Nvidia's AI infrastructure, highlighted the company’s proactive approach in managing supply chain forecasts to ensure timely delivery of components. As Nvidia continues to lead the AI processor market, competition is heating up from companies like Advanced Micro Devices and custom silicon designs from Broadcom and Google. The company has ambitious plans to manufacture up to $500 billion worth of AI infrastructure in the U.S. by 2029, including the production of Blackwell GPUs at TSMC's Arizona facilities. Vera Rubin is expected to roll out in the latter half of 2026, taking Nvidia’s capabilities to new heights. The system's design focuses on maximizing efficiency and performance, marking a significant evolution in server architecture. Daniel Newman from Futurum Group noted that the system integrates all necessary components into a single rack, emphasizing efficiency in its design. Meta has already announced plans to utilize Vera Rubin in its data centers by 2027, with other major clients like OpenAI, Anthropic, Amazon, Google, and Microsoft also expressing interest. The new racks are notably heavy, weighing nearly 2 tons and containing approximately 1,300 microchips—substantially more than the 864 found in Blackwell. One of the standout features of Vera Rubin is its modular design, simplifying installation and maintenance. Unlike Blackwell, where components were soldered to the board, Vera Rubin allows superchips to be easily replaced from its 18 compute trays. While the new system may draw about twice the power of Blackwell, its efficiency promises a tenfold performance improvement per watt. Analysts like Jordan Klein from Mizuho Securities point out that the efficiency of power consumption will be crucial for clients. Nvidia is also making strides in cooling technology, being the first to implement 100% liquid cooling in its systems, which reduces water usage significantly compared to traditional methods. While Nvidia has not disclosed the exact pricing of the racks, estimates suggest a 25% increase from the Blackwell system, placing the cost between $3.5 million and $4 million. As clients seek to diversify their supply chains, many are integrating their own silicon into AI servers. The competition is poised to escalate, particularly with AMD launching its first rack-scale system, Helios, later this year, backed by a substantial commitment from Meta. Despite the competition, Harris remains confident, acknowledging the challenges but expressing pride in Nvidia's innovations. A detailed video showcasing Vera Rubin and its functionalities has been made available for those eager to learn more about this groundbreaking AI system.