Interview with NVIDIA CEO Jensen Huang on AI Chip Pricing and Values, and the China Market
On March 20, local time, NVIDIA CEO Jensen Huang gave an interview to global media on the morning of March 19. He highlighted that the global data center market is where NVIDIA’s market opportunities lie. Regarding the pricing of AI chips, he mentioned that the intention was to provide a general sense of pricing rather than specific quotes.
Here are some key points from the interview, as compiled by IT Home:
Regarding the pricing of NVIDIA’s latest generation AI chip Blackwell, which was rumored to be priced between $30,000 to $40,000, Jensen Huang responded by stating that the aim was to give customers a general idea of the pricing without intending to provide exact figures. He clarified that NVIDIA does not sell chips but data centers, and prices vary significantly based on individual customer needs and different system requirements.
Jensen Huang pointed out that the global data center market reached approximately $250 billion last year and is growing at a rate of 20% to 25%, representing a significant market opportunity for NVIDIA. With a diverse range of chip and software products, NVIDIA can generate substantial revenue from global investments in data center equipment.
Furthermore, he emphasized the importance of the Chinese market, stating that efforts are being made to maximize NVIDIA’s business in China. New chips called L20 and H20 have been introduced specifically for the Chinese market to meet the required standards. He also mentioned that many components in NVIDIA’s chips are sourced from China, highlighting the challenge of breaking the globalized supply chain.
In a recent update, NVIDIA officially unveiled the AI accelerator card Blackwell GB200 at the GTC 2024 developer conference, planning to ship later this year. Jensen Huang revealed that Blackwell offers AI performance of up to 20 petaflops, while the H100 only reaches 4 petaflops. The additional processing power will enable AI companies to train larger and more complex models effectively.
