CEO Jensen Huang stated that Nvidia does more than just design chips, it's a "data-center-scale" corporation.
At the recent GPU Technology Conference (GTC) event, Nvidia announced its latest Blackwell architecture AI GPU series. Huang pointed out that the overall data center market scale Nvidia faces is expected to have an annual growth rate of up to 25%, with the potential to surpass the US$250 billion mark.
In an exclusive interview with CNBC, Huang stated that 30 years ago, everyone believed that CPUs could handle many tasks well and were only bad at some tasks like parallel computing. Thus, GPUs were used to offload these tasks from CPUs.
However, it was later discovered that the tasks CPUs were bad at took up 95% of the entire computing process. It seemed that general-purpose products were not very efficient in almost all sectors.
He noted that the Blackwell computing system is so powerful that it could replace thousands of general-purpose computing systems. The costs of general-purpose computing systems not only come from itself; the cables used to connect the systems are also quite expensive. With the Blackwell system, this one unit can replace thousands of general-purpose systems, resulting in significant savings in energy and other costs.
It is precisely because of such high computational efficiency that the development of AI technology has become a possibility. If computing systems can write their own programs with AI, what do humans need to do?
Huang responded that humans need to guide the software and develop algorithms to help computing systems design programs on their own. This type of algorithm is called deep learning.
Besides chips, Nvidia also develops systems, network products, and software, covering almost the entire composition of data centers. So, do any other semiconductor companies need to exist?
Facing this sharp question, Huang stated that the semiconductor world is "gigantic", and Nvidia is merely investing in "this one niche called accelerated computing and AI." However, this is an important niche market because it is the foundation of future computing.
Huang also revealed that the Blackwell architecture GPU has a staggering 208 billion transistors. The R&D budget Nvidia invested in it exceeds US$10 billion. Its sales price will be between US$30,000 to US$40,000.
According to a report from Reuters, Huang later clarified that these chips are bundled in much larger computing systems, and the price will vary depending on the value these products can provide to customers.
When asked by investors at the GTC event, he estimated that the data center market size facing Nvidia is growing by as much as 25% annually, potentially surpassing US$250 billion.
An analyst from Morningstar stated that Nvidia's products may continue to hold the top spot in the overall AI hardware market. The report also said that Nvidia's ability to expand its business to other software and hardware, network-related products, and platforms is commendable.
Quoting several analysts, Reuters pointed out that as competitors introduce new products and several important customers begin to invest in in-house chip R&D, Nvidia's market share in 2024 may drop by a few percentage points, but its market-leading position will not be affected.
