Nvidia’s Wednesday’s results are expected to show strong sales of the company’s current rack-scale systems. However, attention is currently focused on the next AI system, Vera Rubin, which is scheduled to be released in the second half of this year.
The Vera Rubin is made up of 1.3 million components and the company claims it delivers 10x better performance per watt than the previous generation Grace Blackwell. This is an important advance when energy consumption is one of the most important issues facing building artificial intelligence.
CNBC had an exclusive first look at Vera Rubin at Nvidia headquarters in Santa Clara, California.
Nvidia says its new AI system is a complex mix of components sourced from around the world. Its core chip includes 72 Rubin graphics processing units (GPUs) and 36 Vera central processing units (CPUs), which are primarily manufactured by the following companies: Taiwan Semiconductor Manufacturing Co., Ltd.. Other components, from liquid cooling elements to power systems and computing trays, are sourced from more than 80 suppliers from at least 20 countries, including China, Vietnam, Thailand, Mexico, Israel and the United States.
One of the major challenges facing the company is the rising cost of memory due to global shortages due to AI-driven demand. Dion Harris, Nvidia’s head of AI infrastructure, said in an interview that the company provides suppliers with “very detailed forecasts.”
“We’re coordinating to make sure everything we ship gets into the supply chain,” he said. “We are in good shape.”
This is a critical moment for Nvidia, which dominates the AI processor market but faces increasing competition from the United States. advanced micro device Not just custom silicone. broadcom and Google Tensor processing unit developed in-house. Nvidia plans to manufacture up to $500 billion worth of AI infrastructure in the U.S. through 2029, including manufacturing Blackwell GPUs at TSMC’s new Arizona factory.
Grace Blackwell went live in 2024, changing the landscape of how much computing is possible on a single system. Vera Rubin, scheduled to ship in late 2026, will take the company to a new level. Nvidia CEO Jensen Huang announced in January that the system was in full production.
NVIDIA founder and CEO Jensen Huang discusses the Vera Rubin AI platform during a Q&A with reporters at the annual Consumer Electronics Show (CES) on January 6, 2026 in Las Vegas, Nevada.
Patrick T. Fallon | AFP | Getty Images
“These are huge. They have all the computing, all the networking, all the cabling, all the cooling,” said Daniel Newman of research firm Futurum Group. “They’re built to put all these systems together in one rack for the absolute maximum efficiency and maximum performance. And that’s not how servers have been built historically.”
Last week, Meta announced plans to use Vera Rubin in its data centers by 2027. NVIDIA’s anticipated list of Vera Rubin’s other customers includes OpenAI, Anthropic, Amazon,Google, microsoft. The rack, which is manufactured at Foxconn’s new factories in the United States, Taiwan, and Mexico, weighs about 2 tons and contains approximately 1,300 microchips, compared to Grace Blackwell’s 864 microchips.
Vera Rubin is a simpler modular system intended for ease of installation and repair. Each superchip slides out of one of the rack’s 18 compute trays in seconds. In Blackwell systems, these components are soldered to the board.
Nvidia says the new system uses about twice as much power as its predecessor, but is much more efficient with a 10x performance return per watt.
Jordan Klein, an analyst at Mizuho Securities, said the “most important thing” is “how many tokens can you earn per unit of electricity consumed.”
“The more you tweak it or move it up the curve, the greater the return on the dollar spent,” Klein said.
Nvidia unveiled the Vera Rubin superchip, which features two Rubin GPUs and one Vera CPU and a total of 17,000 components, to CNBC on February 13, 2026 at Nvidia headquarters in Santa Clara, California.
Vera Rubin is also Nvidia’s first 100% liquid cooling system, which Harris says allows data centers to consume “much less water” than traditional evaporative cooling.
Nvidia hasn’t shared the price of the rack, but Futurum Group estimates the price will be about 25% higher than Grace Blackwell, bringing the system price to about $3.5 million to $4 million.
As major customers look to diversify their reliance on chipmakers, many are also filling their AI servers with their own silicon. In October, CNBC visited an Amazon Web Services data center filled with “ultra servers” made up of the company’s Trainium 2 chips. Meanwhile, Google’s data centers are loaded with racks of TPUs.
Nvidia will face significant competition later this year when rival AMD ships its first rack-scale system called Helios. The chipmaker just won a major contract from Meta for up to 6 gigawatts of capacity.
“We’re going to see a lot of adoption because customers want more capacity, but they also want a viable second source to keep Nvidia honest,” Klein said.
Regarding the competition, Harris said, “My hat is off to anyone who is willing to take on the challenge, but this is by no means an easy endeavor.”
Watch the video to learn more about Vera Rubin and how the system works.
