The Six Five team discusses GSA US Executive Forum
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: The Global Semiconductor Alliance Executive Forum that both you and I attended. We both attended the dinner. I did a presentation for their board of directors, which by the way, had everybody from the Mag Seven there, which was pretty cool. And then I did an innovation panel on stage. So the conference, it really reflects semiconductors, right? Discussions on AI, manufacturing challenges, how do you play China? And then, challenges in some of the key elements of AI, compute storage and connectivity. And the conversations are very similar to we talk about even with the board, I can’t give exact details, but essentially, talking through the downstream benefits of AI.
One thing I did notice though is, Dan, as you and I researched not just semiconductors but the markets they go in and the use cases that drive it in the cloud and on the PC, I thought our conversations and my conversations with the board of directors, it was enlightening to them. They don’t spend a whole lot of time, it appears, on the downstream application. So we had a very lively conversation on that. A lot of conversations about GPU and accelerators, and what does that mean? And sustainability and power. Not sustained, power, like power generation, what’s going to happen? And also maybe, disintermediation of smartphones in five to 10 years, given generative AI devices and these goggles.
Just to wrap this up, some of the discussions with Astera Labs, SambaNova and Celestial AI on my panel. No surprises, but three to five years of innovation, model size and complexity, still going up. Cluster size, moving from 100,000 nodes to two million, and across data centers. The shift from mega-training to inference, going from 90-10 to 10-90. And specialization, right? Like ASICs and XPUs versus GPUs. And then following up the conversations and obviously SambaNova is biased. Both Celestial AI and Aster Labs, they support both, right? But both these mesh and networking companies are both working to lower power, increase performance, and lower cost, and that’s the game here in semiconductors. It is interesting, both Astera Labs and Celestial AI very much have clients out there that both you and I would know, they’re integrated into their solutions. I did get a tour of Astera Labs, saw public stuff. I saw basically an H series rack with their chips all over it, as well as an MI300 platform, and I’ll just leave it at there.
Daniel Newman: Yeah, I didn’t have quite as much time as you did there, Pat, but one thing is I wish we could actually tell the people some of the great conversations we had. Sorry, you’re going to have to get into the room and it’s a tough room to get into. And so by the way, thanks bestie, for making that happen. I’m humble pie, I’ll admit it. The overall sentiment though, Pat, you know chips are a boom and bust industry. Overall sentiment though, is extraordinarily bullish and just you know that there’s been times where it hasn’t been, but nobody seems to feel that this is coming to an end anytime soon. Now, that could be misconstrued as sometimes the exuberance of a top, right? That sometimes when nobody sees it coming is exactly when it’s coming. But I think most agree that it’s not because they don’t see it coming from some place of exuberance. They don’t see it coming because they believe it’s still really early, that we’re really just getting started. We’re just getting up to speed on capacity, of capabilities of new packaging and designs and the amount of innovation that’s coming up.
But there are a lot of downstream challenges, that was the one thing I will lead, and some of the conversations I had is, talking to a lot of the people on the design and delivery side, there are some kind of waiting and waiting for that kind of utilization rate to go up, because everyone believes the build out is very, very substantial. But at some point on one-year cycles, how do the cloud providers make money? How do the ISVs make money? But hey, where there’s great software, Pat, behind it, there should be great hardware. So I also see a big opportunity, one of the things I’m predicting now is with the simplicity that’s been put into being able to design and build your own chips is, are more ISVs going to do it? Are we going to get Uber chips for Uber software? Are we going to get, of course we’ve seen it with Meta and Alibaba and Google and they’ve sort of had the ability to do it in a dual capacity, but do companies like Salesforce with Agentforce, are they going to build a specialized chip down the line that’s going to make them more efficient, going to give them more control? And of course, the ability to make sure their solution is absolutely optimized for their use case. It certainly would be more price efficient if they’re using tons and tons of GPUs. So something to think about down the line, Pat, but very exciting times.
