The Six Five team discusses Huawei To Introduce A Chip In Chinese Market To Compete With NVIDIA?
If you are interested in watching the full episode you can check it out here.
Disclaimer: The Six Five Webcast is for information and entertainment purposes only. Over the course of this webcast, we may talk about companies that are publicly traded and we may even reference that fact and their equity share price, but please do not take anything that we say as a recommendation about what you should do with your investment dollars. We are not investment advisors and we ask that you do not treat us as such.
Transcript:
Patrick Moorhead: And that is, guess what, more AI rumor on the street, that Huawei is introducing another version of its Ascend AI Accelerator to compete with NVIDIA. Dan, what do you think here? What’s happening?
Daniel Newman: Hey, Pat, I think sometimes you say to me, “I don’t really like these speculation ones.” And then sometimes you get done with them, and you’re like, “I kind of like doing these speculation ones.”
Patrick Moorhead: Yeah, I do man. As an analyst, you need to be very… I mean there are people who do great rumor mongering, and they get great page views, but we typically don’t. But it is kind of fun to dive in. It’s what our audience wants. Are you not entertained?
Daniel Newman: And there is this theme though. The whole world really… I mean, the big question, the macro questions, can anybody compete with NVIDIA? And so anything that’s a sub theme to that, whether it’s AMD making the acquisition today or Huawei developing a next chip for China that’s going to compete. And again, the note here is that it’s competing with the H100. So the note is different because some people, when I tweeted it, they immediately went down the path of it competing with the China spec chip. And that’s not the point here. The point is that China has not been able to, through its own means, manufacture a competitive GPU to the H100, certainly not to the B series that’s coming down the line. And that’s been the question about China’s ability to maintain competitiveness.
So I think the first point, Pat, you and I would probably agree on violently is that of course China’s trying to do this. Of course they are. Why in the world would they not want to be able to take more control of their situation right now? The sanctions are adding complexity, it’s limiting their ability to innovate, and it’s forcing them to creatively or gray market a lot of hardware, Pat. What did we talk about those cigarette cartons full of GPUs coming across the border?
Patrick Moorhead: Exactly. I mean, to be honest, they’re less cards now. They’re more of these giant blocks, but you get the idea.
Daniel Newman: Yeah, right. Historic, that you’re painting the historic picture. But look, the sanctions are substantial. I don’t see the US moving away from them anytime soon. They need to do this to maintain technological leadership. Not to mention just capacity right now. There’s so much limited capacity. ASML can only make so many machines with laser beams on them, and there’s only-
Patrick Moorhead: We’re going to get there, big guy. Yeah, we’re going to get there.
Daniel Newman: I just wanted to talk about laser beams. But realistically, design, packaging, building, implementation and launching. So my take on the whole thing, Pat, is China is going to do everything in its power to get here. They are going to do whatever it takes to design in copy capabilities, and they’re going to try to manufacture competitive products. Can they do this as early as October, per the rumor? That seems like a very aggressive timeline, but what we don’t know is there’s a bunch of don’t know. How long has this been in flight? Any specific design updates, partnerships, packaging, relationships, how are they putting this thing together? And then what capacity to manufacture do they have? Is it a handful? We’ve seen them being able to take things down, I think as low as what? Seven through SMIC, but the volume has been very low, the volume of… So they’re not able to do these things in high volume, which has been part of the limitation.
So overall, Pat, I think they’ve been able to make a lot of progress in things like handsets and devices because of the differences in design, but I think with these large GPUs and systems, I just think it’s very interesting and provocative, but I just don’t see the horizon here. I think they’re going to be a year or two out, even if they can get it done. And I think this does make for great rumor mill fodder, but I’m not as sold that Huawei and its limitations are going to meaningfully be coming after the H100, maybe even the A100.
Patrick Moorhead: Yeah, so let me just get some stuff out there. The rumor was around a 910C. Huawei’s high silicon already offers an Ascend 910B that was going after the A100 for AI training and inference. So that’s actually done a done deal. And companies like Baidu in China are already buying this. This is the potential follow on. By the way, you know there’s going to be a follow on. Is it called C? Likely. So this rumor is, again, we try to vet our rumors. This has a very high probability of being true. Now, what’s interesting when it comes to wafer capabilities, you have Groq. The silicon that Groq has right now is done on global foundries.
Daniel Newman: 14.
Patrick Moorhead: 14 nanometer.
Daniel Newman: By the way, Groq with a Q, not the Twitter thing with the K. Right?
Patrick Moorhead: Yeah, that’s exactly right. And well, wait a second. How can they do this? Well, first of all, you have to recognize that design is, for the Huawei Ascend is an ASIC, which is like Groq, like Maya, like Inferentia, like the TPU. So likelihood is that it doesn’t have to be on bleeding edge. And then by the way, it’s a lot less efficient on a process standpoint, but an ASIC is more efficient than a GPU. And the way that you do this, by the way, if let’s say you’re hitting 8200 performance, well, how do you get to an H series type of performance? You have to string a lot of them together through clustering. And it’s done. I mean, how do you think that OpenAI, ChatGPT 1.0 was trained? It wasn’t done in an H, it was likely done on A series or something before that. And what you do is just less efficient. It takes more servers, it takes more clusters to make this happen.
Now, NVIDIA is not just going to let all that business disappear. Another rumor that we cover on our show that I would say is very a high likelihood is this idea of a B20. H20 didn’t actually see the light of day, but an A20 did in China. And that was the Chinese cut down version of NVIDIA’s Silicon. I believe that NVIDIA is likely going through the process. There is an actual process. Check out my website if you want to know what that is. But essentially, it’s a red, yellow, green status. And this would be in the yellow where NVIDIA has to ask permission of the government to determine whether they can ship it in there or not.
Daniel Newman: By the way, Pat, you mentioned something interesting about these XPUs basically being put up head to head with these more flexible GPUs. And I think that is an important thing we can’t dive into today because we don’t have time, but understanding that a lot of these accelerators are being built with more logic cores on them, like we’re seeing with Gaudi right now, where you’re hearing Intel making claims about Gaudi being able to compete meaningfully with the Hopper architecture. I think this is similar to what, I’m sure, Huawei’s doing. I haven’t spent as much time in the Huawei portfolio. Hasn’t been as relevant over the past few months, but to your point, they’ve been able to align something to the A series. Have you heard anything about scale on that? I’m just curious. Have they ever been able to hit any volume? Because that’s ane that I’ve heard the trap over there, is they just haven’t been able to get anything produced in volume, which seems like that would be a big limiter for this.
Patrick Moorhead: I haven’t. They’re limited in the amount of 7 nm smartphone chips they can crank out, but I do not know about these basics.
Daniel Newman: We should put them in the lab, run them side by side. Let’s make it happen.
Patrick Moorhead: Totally. You hear that Huawei? Let’s get your infrastructure into our labs. Pronto.
Author Information
Daniel is the CEO of The Futurum Group. Living his life at the intersection of people and technology, Daniel works with the world’s largest technology brands exploring Digital Transformation and how it is influencing the enterprise.
From the leading edge of AI to global technology policy, Daniel makes the connections between business, people and tech that are required for companies to benefit most from their technology investments. Daniel is a top 5 globally ranked industry analyst and his ideas are regularly cited or shared in television appearances by CNBC, Bloomberg, Wall Street Journal and hundreds of other sites around the world.
A 7x Best-Selling Author including his most recent book “Human/Machine.” Daniel is also a Forbes and MarketWatch (Dow Jones) contributor.
An MBA and Former Graduate Adjunct Faculty, Daniel is an Austin Texas transplant after 40 years in Chicago. His speaking takes him around the world each year as he shares his vision of the role technology will play in our future.
