During the NVIDIA 2024 GTC, a developer event in the field of artificial intelligence (AI), NVIDIA founder and CEO Jensen Huang confidently stated that NVIDIA will seize more in the data center market that can "dig" hundreds of billions of dollars a year. Lots of "cake".

At an investor event held during the GTC conference on Tuesday, the 19th local time in California, Huang Renxun told attendees that because Nvidia produces a wide variety of chips and software, the company is in a favorable position and will account for one of the global spending on data center equipment. Most of them are products from NVIDIA.

Huang Renxun estimates that the annual global investment in data center equipment will total US$250 billion, and Nvidia's products will account for a larger share of this than other chip manufacturers. NVIDIA is developing software to enable the adoption and use of AI technology in various industries. Nvidia provides AI models and other software and then charges customers based on their computing power and the number of chips they run.

Huang Renxun said, "Based on my expectations, this (software) will be a very big business."

During Huang Renxun's speech, Nvidia's stock price, which had fallen by about 3.9% in early trading on Tuesday, further narrowed its losses and turned higher at midday. After that, the intraday increase expanded to more than 1%, and finally closed up by nearly 1.1%. It closed for the second consecutive day after falling for three consecutive days. rose, and the closing increase exceeded this Monday, the opening day of this year's GTC conference.

Before making the above speech, Huang Renxun was asked in a recent media program why Nvidia is a company with a market value of US$2 trillion. He answered that the most important tool for mankind is the computer, and today’s computers can understand all kinds of information. This is a $100 trillion industry, and the impact that NVIDIA's work has had on this industry is absolutely extraordinary.

On Monday, Nvidia just released the Blackwell architecture GPU, which is known as the world's most powerful AI chip. The architecture's first GPU B200 improves cost and energy consumption by 25 times compared with the previous generation H100, and increases reasoning capabilities by up to 30 times. It has a second-generation Transformer engine that can provide up to 20 petaflops of FP4 computing power, and the floating point operation speed is five times that of the H100 4 petaflops.

Combining two B200 GPUs with a single Grace CPU, the superchip GB200 delivers 30x performance for LLM inference workloads and significantly improved efficiency. In the GPT-3 large model benchmark test with 175 billion parameters, the GB200 chip performance is 7 times that of H100, and the training speed is 4 times that of H100. The industry exclaimed that the new Moore's Law was born. Online comments said that Blackwell marked the historic achievement of a 1,000-fold increase in the computing power of Nvidia's AI chips in just eight years.

In his keynote speech at the GTC conference on Monday, Huang compared Blackwell to Hopper, Nvidia's previous-generation chip architecture designed for data centers. He said that if you want to train a 1.8 trillion parameter GPT model, it will take about three to five months:

If you do it with Hopper (architecture), you may need 8000 GPUs and consume 15 megawatts (power). With 8,000 GPUs and 15 megawatts, it would take 90 days, about three months.

If you do it with Blackwell (architecture), you only need 2000 GPUs. 2000 GPUs, same 90 days. But here's the amazing part, requiring only four megawatts of power.

Huang Renxun revealed that Amazon Cloud AWS, Google, Microsoft, Oracle, etc. are all preparing to support Blackwell. At the same time, NVIDIA will continue to strengthen the ecosystem based on AI. For example, NVIDIA NOmniverse Cloud will be able to connect to Apple's hybrid head-mounted display Vision Pro, enhance models and general robot ecosystems, etc.

Huang also introduced the basic humanoid robot model Project GR00T, the new humanoid robot computer Jetson Thor, and interacted with a pair of small NVIDIA robots from Disney Research.