Jensen Huang, CEO of Nvidia Corp., stated in a media interview on Tuesday that the company's latest generation of AI training chip, Blackwell GPU, will attract a wider customer base at a reasonable price. He is optimistic about the growth of artificial intelligence (AI).
This is Nvidia's latest flagship product, following the Hopper architecture data center GPU released in March 2022, and is expected to replace the market position of the H100, which has been heavily sought after. In terms of naming sequence, the latest product that appeared skipped the 100 sequence and was named directly as "B200".
Huang Renxun stated that the selling price of B200 may be between 30000 and 40000 US dollars. This pricing is lower than some analysts expected, and he told them that he hopes the pricing of the new chip can attract as many customers as possible.
However, he later added that Nvidia would incorporate its new chips into larger computing systems, and prices would vary based on the value they provide.
In addition, he expects the B200 to be shipped later this year. Huang Renxun said that NVIDIA is collaborating with TSMC to avoid chip packaging bottlenecks, which have slowed down the shipping speed of its previous generation flagship AI processors.
"Last time there was a sharp increase in demand, but this time we have gained a full understanding of the demand for Blackwell chips," he said.
Nvidia's CFO Colette Kress also stated that the new chip and the system it is based on will be launched later this year. She said that although the company is better prepared for the surge in demand than the current generation of products, there may still be supply constraints.
Optimistic about the global data center market
In this interview, Huang Renxun also mentioned that due to the wide variety of chips and software produced, Nvidia can generate a large portion of revenue from global investment in data center equipment. He said that global spending on data center equipment will reach $250 billion annually, and Nvidia will gain a higher share than other chip manufacturers.
He emphasized that the market's annual growth rate is as high as 20%.
In addition, Nvidia is also committed to developing software to allow artificial intelligence technology to be adopted in various industries. The company provides artificial intelligence models and other software, and charges based on the amount of computing power and the number of chips used to run the calculations required by customers.
Huang Renxun said at an investor event in California on Tuesday, "My expectation is that this will be a very large business.".
In summary, he pointed out that Nvidia is shifting from selling individual chips to selling entire systems, potentially winning a larger "cake" within the data center.
"Nvidia is not manufacturing chips, but building data centers," he said.