"Unprecedented Chip? Unprecedented Stock Price?" Nvidia Teases Debut of 'Monster AI Chip' with Double the Performance

As Nvidia prepares to hold its annual developer conference, GTC 2026, in California on March 16 (local time), the global semiconductor industry is turning its attention to the event. In particular, as CEO Jensen Huang has hinted at unveiling a "chip the world has never seen" during the event, there is growing anticipation over whether the detailed specifications of "Feynman," Nvidia's next-generation artificial intelligence (AI) semiconductor, will be revealed for the first time.

According to the semiconductor industry on March 4, Nvidia is expected to unveil its next-generation AI chip, Feynman, at this year's GTC. CEO Huang has drawn attention every year at GTC by introducing new semiconductors. At last year's GTC, he stated, "We have laid the foundation to scale up hundreds of thousands of graphics processing units (GPUs)," and also announced, "We will launch a new architecture and a new product line every two years."

Jensen Huang, CEO of Nvidia, introduced the next-generation graphics processing unit (GPU) Rubin and the data processing unit (DPU) Bluefield4 at Nvidia CES 2026 held at the Fontainebleau Hotel in Las Vegas, Nevada, USA on January 5 (local time). Photo by Yonhap News.

원본보기 아이콘

Feynman is a next-generation GPU for data centers that Nvidia is developing with the aim of launching in 2028. According to the AI chip roadmap unveiled by Nvidia at last year's GTC, Feynman will combine next-generation high-bandwidth memory (HBM), the eighth-generation NVSwitch, and a 204Tbps (terabits per second) class network, which is expected to further accelerate the expansion speed of AI factories. An industry insider explained, "According to the roadmap announced so far through 2028, the next version to be unveiled is likely to be Feynman."

원본보기 아이콘

Currently, a significant portion of Nvidia's revenue comes from its data center division. As demand for AI development rapidly increases, Nvidia is likewise focusing on enhancing the performance of its semiconductors for AI. Feynman is projected to deliver at least twice the computing performance of "Vera Rubin," which is scheduled for release in the second half of this year.

While the specific specifications of Feynman have not yet been disclosed, it is highly likely that TSMC's 1.6nm (1nm equals one billionth of a meter) advanced process (A16) foundry will be applied. According to reports from Taiwan, Nvidia is expected to secure priority access to TSMC's advanced process semiconductor production lines, and the A16 foundry technology will be used for the Feynman chip. If Nvidia adopts a 1nm-class foundry process, it will be the first time this has occurred. The start of mass production is anticipated for the second half of this year.

There are also signs that the landscape of Nvidia's partner ecosystem may shift with the introduction of Feynman. To diversify its supply chain, Nvidia may have a portion of Feynman chips produced by Intel. While TSMC would continue to handle mass production of the core GPU components, discussions are reportedly underway to outsource semiconductor manufacturing and packaging to Intel's 18A or 14A processes.

© The Asia Business Daily(www.asiae.co.kr). All rights reserved.