On March 21st, local time, Jensen Huang, founder and CEO of NVIDIA, disclosed new artificial intelligence and chip technologies at GTC, an annual technology summit for software developers. The most impressive announcement was that the company will rent powerful and expensive supercomputers that were used to develop artificial intelligence technologies such as ChatGPT to almost all enterprises, according to reports from YICAI.
NVIDIA announced that it will open the DGX Super AI Computing System integrated with eight flagship A100 or H100 chips to enterprises by way of leasing access, with a monthly rent of $37,000, in order to accelerate AI development as led by big language models.
On March 22nd, Beijing time, Jensen Huang said: “We will cooperate with cloud service providers in Europe and America to provide NVIDIA’s DGX system AI supercomputer capability. In China, we have specially customized Ampere and Hopper chips. These will provide Baidu, Inc. capabilities through Chinese cloud providers, such as Alibaba Group Holding Limited, Tencent and implementation. I fully believe that they have the ability to provide top-level system services, and Chinese start-ups will definitely have the opportunity to develop their own big language models.”
The Ampere and Hopper chips available in China referred to by Jensen Huang are the A800 and H800, which are also the chips used by most Chinese developers when developing large language models.
At present, in addition to US giants such as Microsoft and Google, which have invested heavily in large-scale AI language models, Chinese Internet giants and technology start-ups have also invested in R&D like GPT models. NVIDIA dominates the field of artificial intelligence chips, and a large number of chip demand has pushed NVIDIA’s share price up by more than 77% this year. Nvidia’s market value soared to $650 billion, about five times that of Intel.
Jensen Huang also mentioned the three stages of AI’s vigorous development and how NVIDIA made technical preparations for AI development in this process.
He said that 12 years ago, NVIDIA realized that deep learning would change the way software was deployed, so the company reinvented modern computers and updated them every generation. “This is the first stage of AI development, establishing infrastructure,” he said. “The second stage is AI learning perception, such as machine vision, automation use cases, etc. The third stage is the AIGC computer-generated content stage that we are experiencing. AI is a co-creator and is participating in all of the work.”
Jensen Huang believes that AI can help human beings create the first draft, establish a preliminary design, help brainstorm ideas, stimulate human creativity and improve overall production efficiency. He said that NVIDIA has made several preparations. First of all, establishing infrastructure and deploying software. The second is to put this computing power into the cloud and let the infrastructure be shared faster by cooperating with cloud service providers. Third, under the background of the explosive growth of AIGC’s demand, adequate supply chain preparation should be made to meet the global demand for AI.
“The iPhone moment of artificial intelligence has already begun,” Jensen Huang said in a keynote speech at GTC. He further explained that he believes that the impact of artificial intelligence on society since its development may be the same as that of Apple iPhone opening up the smartphone market.
Hans Mosesmann, chip semiconductor analyst at Rosenblatt Securities, said NVIDIA’s latest product was “years ahead of its competitors”. “NVIDIA’s leadership position in AI software is not only a milestone, but also accelerating development.”
At present, NVIDIA has helped Microsoft and other partners build a huge system for their ChatGPT services. At the latest GTC conference, NVIDIA cooperated with Oracle and other companies to provide access to DGX supercomputers. Large supercomputer systems can be equipped with up to 32,000 NVIDIA chips.
NVIDIA has also launched a service called AI Foundations to help enterprises train their customized artificial intelligence models, and several stock image database manufacturers have planned to use this service.
In addition, NVIDIA recently announced cooperation with quantum computing researchers to accelerate software development projects, and cooperation with chip industry giant TSMC to accelerate chip development.
In the chip manufacturing process, NVIDIA announced new software technology for lithography machines, which will use NVIDIA chips to accelerate the steps between software-based chip design and the physical manufacturing of lithography masks for printing the design on the chip. Traditional computing chips take two weeks to complete, but Nvidia’s chips and software are used at the same time, which can handle these tasks overnight and reduce the power used from 35 MW to 5 MW.
NVIDIA said it was working with ASML, Synopsys and TSMC to bring the technology to market. Huang Renxun expects TSMC to start preparing to produce the technology in June.