From "generations every two years" to "generations every year", NVIDIA significantly strengthens its new product release frequency

Author: Li Xiaoyin

Source: Hard AI

After the U.S. stock market closed on Wednesday, May 22, Huang Renxun revealed in the latest financial report conference call that NVIDIA will accelerate the pace of chip architecture updates, from once every two years to once a year:

"I can announce that after Blackwell, we will launch another chip. Our rhythm is one generation per year."

Huang Renxun also stated that NVIDIA will not only accelerate the development of AI chips, but also speed up the production of all other types of chips:

"We will advance comprehensively at a very fast pace."

"New CPUs, new GPUs, new network cards, new switches... a large number of chips are on the way."

It is understood that NVIDIA's previous chip architecture update cycle was approximately once every two years: Ampere was launched in 2020, Hopper in 2022, and Blackwell in 2024.

The decision to accelerate the transition of AI chips to an "annual update" is mainly due to strong market demand.

Huang Renxun predicted during the conference call that the situation of AI chip supply shortage will continue for some time.

"We expect that as we transition to H200 and Blackwell, demand will exceed supply for a period of time. Because everyone is eager to get their infrastructure online, they want to save money, make money, and achieve this goal as soon as possible."

At the same time, the demand for GPUs in the automotive market is still soaring. Huang Renxun stated that automakers will become "the largest vertical enterprise in data centers this year."

According to Tianfeng Securities analyst Guo Mingchi's previous forecast, NVIDIA's next-generation architecture "Rubin" is expected to be launched in the fourth quarter of 2025, bringing a brand new R100 AI chip, consistent with Huang Renxun's announced plan.