DeepSeek Disruption: Impact on AI and Chip Market (Full Transcript)

Speaker 1: Joining us now to talk more about the DeepSeek disruption is Patrick Moorhead, more insights and strategy founder, CEO, and chief analyst. Pat, good to see you. So the big question to me here with all this DeepSeek stuff is, is this a minor shock that's going to lower the cost of AI infrastructure, boost demand as much or more, or is it the beginning of an ecosystem shift that's going to hit chip and infrastructure valuations for a while?

Speaker 2: So I subscribe to the former. I believe that this is going to radically shift costs down, which is exactly what we need to hit those downstream effects that I talk about in every one of these show. Consumers and enterprises have to buy into the benefits, and if they don't, everything crumbles. And what this does, it accelerates us to get there. And on the other hand, if I look at the future of, let's just say the end game is AGI, that's going to require a tremendous amount of training and the entire ecosystem that goes with it. So I think net-net, this is a positive for tech. And I think the reason that the markets are responding the way that they are is because of the uncertainty around it and what it means. And NVIDIA, I know they're in their quiet period, but it seems like people want to hear more from them than their statement that they put out today.

Speaker 1: Pat, where does it open up opportunity? I know you got some software names that you're going to share with us, but also I wonder about the hyperscalers themselves, because they're all designing their own AI chips that they've put out there with the idea of being, hey, you can run some things more efficiently with us. Is this going to cause customers to be more open to that efficiency argument and not just automatically say, give me as much NVIDIA as I can eat? I don't care what it costs. Yeah. So on the software and SaaS side, the

Speaker 2: Adobe's, Microsoft's, Salesforce's, SAP ServiceNow, I think are all going to do well there. When it comes to the AWS's and the Azure's and Google Cloud, folks like that, their homegrown chips are even more important because although they can do training, they primarily have been used to do inference. And that's exactly the breakthrough that we saw here. And if we look back five years ago, there was a lot of investment into training for machine learning, and then generative, and then it shifted to inference about, it went from 9010 to 1090. And here we are in generative AI, it's about 9010. We will likely see a shift like we saw previously with ML to where the action is inference. And this is where we've seen NVIDIA put a lot of effort into education where they're saying, hey, we're really good at training, but we're really good at inference too. And they do have a point because the low latency, the experience that you need to deliver on the inference side is a lot more complex than it was five years ago. And you do need more silicon to do that.