China's AI chips: What would be needed to catch up to Nvidia

02/06/2024 19:37
China's AI chips: What would be needed to catch up to Nvidia

In a note from Bernstein laying out their predictions for China and AI, app development, use cases, and race to build AI-based platforms will continue to surge. Due to Biden administration restrictions around how and what sort of AI chips companies are allowed to export to China, it may put a damper on some of China's ability to compete in the AI race. Bernstein Managing Director and Senior Analyst Stacy Rasgon, joins Yahoo Finance to discuss the AI race in China and their ability to compete with American companies. When asked about the gap between China and America chips, and what they have to do to catch up, Ragson explains that there is a big gap and "What does that mean? That means you have to shovel more chips at the problem. For example the Huawei Ascend supposedly is roughly equivalent to a Nvidia A100, and that part from Nvidia is two or three years old...If they want to get the same amount of performance as Nvidia's or others are delivering with their current parts, they'll have to put a lot more silicon in place in order to do that, so that costs money in terms of chip cost, it cost a lot of money in terms of power and infrastructure and everything else, and so is it going to be cost competitive? Like if they had to sell that stuff elsewhere? Absolutely not, it's not competitive, but they have no choice, so they'll do it." For more expert insight and the latest market action, click here to watch this full episode of Yahoo Finance Live. Editor's note: This article was written by Nicholas Jacobino

In a note from Bernstein laying out their predictions for China and AI, app development, use cases, and race to build AI-based platforms will continue to surge. Due to Biden administration restrictions around how and what sort of AI chips companies are allowed to export to China, it may put a damper on some of China's ability to compete in the AI race.

Bernstein Managing Director and Senior Analyst Stacy Rasgon, joins Yahoo Finance to discuss the AI race in China and their ability to compete with American companies.

When asked about the gap between China and America chips, and what they have to do to catch up, Ragson explains that there is a big gap and "What does that mean? That means you have to shovel more chips at the problem. For example the Huawei Ascend supposedly is roughly equivalent to a Nvidia A100, and that part from Nvidia is two or three years old...If they want to get the same amount of performance as Nvidia's or others are delivering with their current parts, they'll have to put a lot more silicon in place in order to do that, so that costs money in terms of chip cost, it cost a lot of money in terms of power and infrastructure and everything else, and so is it going to be cost competitive? Like if they had to sell that stuff elsewhere? Absolutely not, it's not competitive, but they have no choice, so they'll do it."

For more expert insight and the latest market action, click here to watch this full episode of Yahoo Finance Live.

Editor's note: This article was written by Nicholas Jacobino

Video Transcript

JULIE HYMAN: I want to turn to China for just a second, because you and your team were out with a note looking at some predictions for China AI. And they're sort of specifics, so I want to zoom out a little bit and just talk about China versus the US and sort of where we are in that development. Can you talk to us about that and sort of where we are and who stands to win? Or is it not zero sum here?

STACY RASGON: Well, so there are export controls. Now that were put on US companies selling AI parts into China. Mostly, they affect Nvidia. They don't really affect others, because most of the others, the revenues were very, very small. Nvidia had a sizable business in China. And under the export controls, as they currently send, Nvidia basically can't sell any of these parts in China at all.

They are working on new parts that will be legal to sell. However, given the performance thresholds that are around these sanctions, they have to drop the performance very considerably. And in fact, there are local Chinese alternatives from companies like Huawei that at least on paper will have better performance than what NVIDIA is allowed to sell into China.

What Nvidia will have is software compatibility. And like a lot of the existing AI infrastructure and models that have been built in China were already built on Nvidia hardware, and so they're leveraging Nvidia-- the Nvidia software ecosystem. And so the Nvidia parts will allow seamless integration with that. And so they're probably more suitable I would think for more of the tier two developers. The larger ones in China may go with other alternatives if they've got the resources available to port everything over.

I would say long term, though, I kind of view this-- my view is once they put sanctions on a year ago, I think long term, the China TAM for the US guys is going to be impaired because they've lowered these thresholds. But even if they never did, they were never going to be raised. And so the gap between what is legal for US companies to sell into China in terms of performance and the rest of the world is going to get bigger and bigger and bigger over time anyways. And so I think you do need to think about that TAM ideally as potentially as impaired.

Now I think the overall TAM long term is enormous. I think it's big with or without China. So I think it's manageable. But you need to do that. I will put aside, by the way the whole question of, was this whole thing the right thing to do? I mean, I can understand the national security implications of wanting to constrain China's ability to build this up. But at the same time-- and it's not just AI, it's AI and it's equipment and all kinds of other things-- we are forcing China to become creative, right, in all senses of that word.

JULIE HYMAN: Well, that's what I was going to ask you, Stacy. I mean, how far are Chinese chip developers from matching Nvidia? You said the Huawei chips allegedly can do what the lower-- you know, the ones-- the lower performance ones can do. But I mean, you know, it's still seems like there's still a pretty big gap between homegrown chips and what Nvidia is producing elsewhere.

STACY RASGON: It is. And so what does that mean? That means that you'll have to shovel a lot more chips at the problem. So for example, the Huawei ascend supposedly is roughly equivalent to an Nvidia A100, that part for Nvidia is two or three years old. On a performance basis, it's roughly equivalent. Again, it doesn't have the software ecosystem and everything. But if you just want to take that, I mean, fine.

So if they want to get the same amount of performance that Nvidia's or others are delivering with their core parts, they'll have to put a lot more silicon in place in order to do that. So that cost money in terms of chip cost. It costs a lot of money in terms of power and infrastructure and everything else. And so is it going to be cost competitive like if they had to sell that stuff elsewhere in the world? Absolutely not. It's not competitive.

But they have no choice. So they'll do it. We saw the same thing like on smartphones. Huawei is making smartphone chips again using supposedly SMIC 7 nanometer processes, which are not going to be competitive with what say Qualcomm or other main suppliers can provide. And they're going to burn more power. They're going to cost a lot more. But again, but if you have no choice, you will do it. And then over time, you will try to figure out ways to optimize within the constraints that you have. Again, they are going to have to be creative.

I don't know what this means 10 years from now, 15 years from now. But it will be more challenging. And China in general is a big market for US and other suppliers. It's going to be tougher. It's going to be tougher. I don't know what it looks like though as we kind of roll this out.

JOSH LIPTON: All right. We'll keep watching. Dr. Stacy Rasgon, Thanks for joining the show, my friend. Always love having you.

STACY RASGON: My pleasure.

Read more --->