Nvidia's new AI computing platform makes leap with new TSMC-made chips

Taipei, Jan. 6 (CNA) Jensen Huang (黃仁勳), CEO of AI chip giant Nvidia Corp., introduced on Monday the company's latest supercomputer platform, featuring six new chips made by Taiwan Semiconductor Manufacturing Co. (TSMC), and said it is now "in full production."

"If Vera Rubin is going to be in time for this year, it must be in production by now. And so today I can tell you that Vera Rubin is in full production," he said during his keynote speech at the 2026 International Consumer Electronics Show (CES) in Las Vegas.

The rollout of six concurrent chips for "Vera Rubin" -- the company's next-generation AI computing platform -- marks a strategic departure from Nvidia's previous "one or two chip" cadence, Huang said.

He said more than one or two new chips needed to be developed in a world where AI models are growing 10-fold and the tokens generated are increasing five-fold annually.

"It is impossible to keep up with those kinds of rates ... unless we deploy aggressive extreme co-design, innovating across all the chips, across the entire stack, all at the same time," Huang said.

The six-chip roadmap was first teased by Huang during his visit to Taiwan in August 2025, when he confirmed the Silicon Valley giant had "taped out" the new designs at TSMC.

At the time, he described the Rubin as "revolutionary" because all six chips were new.

The suite -- comprising the Vera CPU, Rubin GPU, NVLink 6 Switch, ConnectX-9 SuperNIC, BlueField-4 DPU, and Spectrum-X Ethernet Switch -- succeeds the Blackwell architecture, with the chips primarily manufactured using TSMC's advanced 3-nanometer process.

The platform's flagship system, the Nvidia Vera Rubin NVL72, is a liquid-cooled supercomputer weighing nearly 2 tons.

It is named after the American astronomer Vera Rubin, whose observations of galactic rotation provided the first robust evidence of dark matter -- the invisible mass that holds galaxies together -- and revolutionized people's understanding of the universe.

The new supercomputer slashes inference costs to one-seventh of the Blackwell platform and reduces the GPU count required for training Mixture-of-Experts (MoE) models by 75 percent, according to Nvidia.

Leading AI labs, cloud service providers, and system builders, including Amazon Web Services (AWS), Meta, Google, and Microsoft, are expected to be among the first to adopt the new platform, the company said in a press release on Monday.

Meanwhile, another Taiwanese company, Hon Hai Precision Industry Co. (Foxconn), is said to be the main manufacturer of AI servers using Nvidia's Rubin platform.