Huang Renxun's latest interview: Blackwell chip priced at $30,000-40,000, artificial intelligence climbing just beginning, unprecedented progress will be seen in the next 10 years

Author: Hard AI

Author: Zhao Ying

NVIDIA, regarded as the "AI vane", unveiled its latest next-generation AI chip architecture Blackwell yesterday. Huang Renxun made a heavy release, followed by answering media questions and addressing related issues on the Blackwell chip and AI development in some media interviews.

Here are the main points of the interview:

1. The price of the Blackwell chip is around $30,000 to $40,000, but NVIDIA will not sell individual GPUs, preferring to sell supporting network devices and software services.

2. The development of AI has just begun and will disrupt various industries in the future.

3. NVIDIA has created a new way of computing, with its technology integrated into all computer manufacturers' products, connecting the world together, which is why NVIDIA is ubiquitous globally.

4. AI on NVIDIA hardware is changing daily life, like the invention of electricity, and even comparing it to this underestimates the importance of NVIDIA.

5. Essentially, there is no such thing as discontinuity, and the underlying logic of multiple industries is similar.

6. In the next ten years, we will see unprecedented advances in computing technology, with NVIDIA researching a combination of traditional supercomputers and quantum computers.

Blackwell Chip Pricing: $30,000 to $40,000, with Supporting Services Sold

Huang Renxun stated in an interview with CNBC:

The price plan for the Blackwell GPU is $30,000 to $40,000, but this is just a rough price. The price differences for different systems are significant based on each customer's needs. NVIDIA does not sell chips, but rather data centers.

The global data center market reached around $250 billion last year and is growing at a rate of 20% to 25%, which is where NVIDIA's market opportunity lies.

In terms of cost, an analyst from Raymond James believes that the cost of manufacturing a B200 GPU accelerator is around $6,000.

Each Blackwell GPU actually integrates two Compute Dies, connected by NVLink-HBI (High Bandwidth Interface) technology at 10TB/s. In addition, around the two computing chips are 8 8-layer stacked HBM3e memories, with a total capacity of up to 192GB and a bandwidth of up to 8TB/s.

Analysis indicates that the cost of the dual-chip GB200 with 192GB HBM3E will be significantly higher than the single-chip GH100 processor with 80GB memory. The cost of each H100 is around $3,100, while the cost of each B200 should be around $6,000 Of course, this does not include NVIDIA's spending on research and design, as Huang Renxun stated:

The development of GB200 is a challenging task, with NVIDIA's spending on the new generation GPU architecture and design reaching as high as $10 billion.

Regarding B200, NVIDIA is more inclined to sell supporting network equipment and software services, rather than just the accelerator itself. For example, servers like the DGX B200 with 8 Blackwell GPUs priced at millions of dollars each, or even the DGX B200 SuperPOD with 576 B200 GPUs.

Some analysts point out that "the gross profit margin of GPU cards themselves coming down is a healthier state, the volume can exceed expectations, expand the base of GPUs, and then make profits through supporting network equipment and software services."

AI Development has Just Begun and Will Disrupt Various Industries in the Future

Huang Renxun pointed out in an interview with CNBC:

"The climb of artificial intelligence has just begun and will continue for several years, with investments in the field of artificial intelligence still in the early stages."

At the same time, Huang Renxun predicts that as AI technology impacts various industries including healthcare, there will be further growth in the coming years.

Huang Renxun emphasized the innovative ways in which artificial intelligence can be applied in fields such as science and healthcare, stating:

"Artificial intelligence can help 'understand the meaning of proteins, the meaning of life,' thereby accelerating the development of new therapies.

We can simulate life using computers, so we don't have to do a lot of screening work in the laboratory.

Therefore, no matter what drugs we ultimately decide to test, the likelihood of them passing the actual tests will be much higher."

Huang Renxun also mentioned that in order to apply their extreme chips to real-world tasks, NVIDIA needs to have a deep understanding of many disciplines.

"Without chips specifically synthesized for proteins, you have to understand proteins, biology, you have to understand what scientists want to do, and how we can better automate their work, all these different algorithms require quite a bit of research."

Huang Renxun stated:

"In the next ten years, we will see unprecedented advances in computing technology, such as climate technology, digital biology, general-purpose robots..."

Although many of NVIDIA's recent achievements are the result of investments made 5 to 15 years ago, they are still ongoing. At some point, the performance of quantum computers may surpass that of traditional supercomputers, such as those based on NVIDIA GPUs. However, Huang Renxun expects this to take another one or two decades, and he also believes that the most powerful computers in the world will be a hybrid of traditional and quantum, which is also an area of research the company is pursuing.

Will NVIDIA be Everywhere?

Huang Renxun also mentioned a series of NVIDIA's customers, stating that the company's technology has successfully greatly accelerated data processing speeds and reduced costs

We have created a new way of computing, our technology is integrated into the products of all computer manufacturers, connecting the world together, that's why NVIDIA is everywhere. We are present in every cloud, every data center.

This is a vivid metaphor, AI running on NVIDIA hardware is changing everyday life, just like the invention of electricity, and even comparing it underestimates the importance of NVIDIA.

According to Raymond James Financial, even though NVIDIA dominates the AI chip market with a market share of around 85%, this does not fully convey NVIDIA's contribution to AI.

Essentially, there is no such thing as discontinuity, the underlying logic of multiple industries is similar

As NVIDIA accelerated its development, AI was not only a distant future for it, the entire tech industry was skeptical of the potential of this technology, leading to disappointment and reduced investment known as the AI winter.

But Huang Renxun emphasized that the company's current success largely stems from the choices it made in the early days when it did not fully grasp where they would lead the company. He philosophically said:

As usual, the future unfolds in a continuous manner, essentially there is no real discontinuity.

For example, even when GPUs were only about graphics, NVIDIA used them for programming, so they could do more than just project predefined pixels onto a display.

This made them a platform in themselves, not just an accessory to the computer CPU.

The skills NVIDIA acquired along the way prepared it to face new challenges, Huang Renxun explained, illustrating how learning in one industry can be applied to another.

Particle physics in video games is similar to fluid dynamics in molecular simulations, the image processing we use for lighting in computer graphics scenes is no different from medical device image processing.

In many ways, these logics are very similar, so we can slowly and systematically broaden our horizons.