NVIDIA to Accelerate AI chip updates, huge demand and industry impact
TapTechNews May 23rd news, as is well known, NVIDIA will introduce a new generation of GPU architecture on average every two years, such as Ampere released in 2020 and Hopper released in 2022, and Blackwell in 2024, which is true for both AI and gaming cards.
However, due to the explosion of the artificial intelligence industry, NVIDIA can rake in a profit of 14 billion US dollars (TapTechNews note: currently about 101.36 billion Chinese yuan) in just a single quarter with just its AI chips, and it is still accelerating further.
In order to adapt to the industry demand, NVIDIA CEO Jensen Huang now announces that the company will design a brand new AI chip every year from now on. 'I can announce that after Blackwell, there will be another chip coming. We will maintain an annual update rhythm.'
Famous analyst Ming-Chi Kuo also said a few days ago that NVIDIA's next-generation new architecture with the code name 'Rubin' will be released in 2025. Jensen Huang's words today seem to confirm this statement, which means that we can see the AI chip product adopting R100 as soon as next year.
Jensen Huang also emphasized that NVIDIA will also speed up the update speed of all other chips to adapt to this rhythm. 'We will advance all product lines at a very fast pace.' 'New CPUs, new GPUs, new network cards, new switches... A large number of new chip products are coming.'
In today's earnings call, when asked about how the latest Blackwell GPU can quickly gain volume when the previous generation Hopper GPU is still selling well, Jensen Huang explained that NVIDIA's new-generation AI GPU is downward compatible both electrically and mechanically, and can run the same software. He said that customers can 'easily transition from H100 to H200 and then to B100' in their existing data centers.
To explain the huge demand for NVIDIA's AI GPUs, Jensen Huang also shared some sales viewpoints in the call:
During our transition to H200 and Blackwell, it is expected to迎来 a period of shortage of supply. Everyone wants to get their infrastructure up and running as soon as possible, because these products can help them save costs and generate profits better, so they all hope to achieve this as early as possible.
He also humorously cited the 'Fear Of Missing Out' (FOMO) effect to support his view:
The next company to be the first to climb the next important peak will be able to release ground-breaking AI products, while the company that follows can only release products with a performance improvement of 0.3%. Do you want to be the company that leads the AI revolution or just the company that brings some minor improvements?
NVIDIA's chief financial officer also emphasized that the automotive industry will become its 'largest enterprise segment in the data center field this year', and revealed that Tesla has purchased 35,000 H100 GPUs for training its FSD system, while 'consumer Internet companies' like Meta will also continue to maintain a'strong growth momentum'.
NVIDIA said that some customers have already purchased or plan to purchase more than 100,000 NVIDIA's H100 GPUs - Meta plans to deploy more than 350,000 in its operating system by the end of the year.