NEA: Enterprises Want to ‘Own’ Their AI Intelligence

I think the simple truth that there are many voices in your position is like, if the frontier lands pace, the frontier life resumes. Is that fair? Yes, I think that's fair. First of all, thank you for having me and greatly excited to be here. Um, you know, one, one actually, way I like to think about this is, you know, you have the frontier, you have where the technology is today, and then you actually have the kind of reality of adoption. Um, and I actually think Waymo is a good early example of this. Probably the cars were ready to be on the road far ahead of when they actually got on the road. Right. And what did it take to actually get them on the road? It took trust. It took regulation, and it took, you know, obviously shipping them into production. Um, and so I almost think, yes. You also took regulation. Yeah. City by city, state by state or federal. Totally. And trust by consumers that when I get into a car, uh, you know, it's it's going to be safe. Um, and now I think you get into a Waymo and you're like, wow, the magical moment of riding in an autonomous vehicle. And so I think we will see that with I, we've had some of that right to close that gap between where we are in adoption, from where we are on the technology front? Um, obviously, like every single model that gets released as more and more capable, we've seen just incredible breakthroughs on the intelligence front. Um, but if you ask me, like, I think we're still we have so far to go on the adoption front, both on the enterprise side and then actually, you know, as a as a personal consumer as well, right, so far to go, I mean, we had this conversation earlier with, with the Insight partners about Ford deployed engineers, which is not a new conversation, but there's the admission there that you need forward deployed engineers because like loads of companies just haven't worked out what it is they're supposed to be doing with IoT. So you might see that as time, though. The time is there. The Tam is huge. I actually do think like this just means even if we're like early single digit adoption, let's say to today on an overall broad curve, there is so much room to grow. Um, you know, and I think like a couple points make me really believe that intelligence is not enough. And I think it's incredible. We need to keep, you know, kind of keep going up that intelligence curve. But we really need to figure out how to get up the deployment curve of the adoption curve. And so I think, you know, you had OpenAI and anthropic publicly launch, uh, deployment companies, right to say until raw intelligence is not enough, we need to translate that into real outcomes, translate that into the enterprise. Uh, we just saw anthropic, um, you know, actually talk about their frontier Academy. So they're investing $100 million to train 10,000 engineers, kind of between now and the end of next year, right? To actually teach people how to take this intelligent and how intelligent and how to actually get it into real world examples where it's, you know, producing outcomes, where it's dealing with end to end workflows, where it's producing, you know, real, tangible ROI. I think today everyone thinks obviously, yes, like AI is powerful and it's going to help me do X, Y, and Z. But now we're at the point where rubber meets the road and say like, okay, well, what is that ROI based on? You know, what is the value I'm driving based on what I'm spending? There is, uh, a trend, but there is a, um, approach of owning your own intelligence. First of all, like, I think that's something that you would prescribe to you. Right? Uh, yeah. Subscribe to, like, explain that part to me. And then within that is, you know, the thing I see quite a lot is, um, depending on where you are in the stack, which layer of AI you're in, it's like, what are you paying for? Um, rum run with both ideas. Yeah. So. Okay. So own your own intelligence. What does that really mean? I think it means a couple things. One, from an enterprise perspective, there is this tailwind of enterprises wanting to own a piece of their intelligence stock. You know, whether that actually means training their own models, um, having more control over their data, more control over their intelligence. So I think that's a trend will continue to see more of a very, very important trend and actually a really important part of the ecosystem. Um, you know, I just saw kind of overnight, um, or I guess yesterday that, uh, reflection is supposed to be launching soon. It's Western version of an open weights model, which is super important. I think that will help this trend of, uh, own your own intelligence. Right. And it really just comes down to, I think three things. If you think about it. You know, one certainly is this idea of enterprises want to keep their moat and they want to keep compounding their moat. That means data private, that means compounding workflows, etc.. So kind of keep it within their, you know, their own ecosystem, if you will. The second thing or the second, two things which are related is just performance and cost. Um, so if you think about owning your own intelligence relative to the frontier, there's a wide spectrum of kind of what tasks need to be taken care of by the frontier. Right. Not every single query or workflow or agent, um, you know, agent behavior needs to be handled by a frontier model and kind of what? You know, maybe, let's say the 80% that gets done or the everyday things that get done could actually be tackled by an enterprise's own model. Right. And so you actually have a whole ecosystem building around this. Right. And where you're actually seeing higher performance than some of the base models at a lower cost as well. It's the idea that you avoid paying for tokens twice over. Um, everything that we've discussed. You know, I really, uh, I think we, we under appreciate any a size and scale, you know, where it's deploying capital. How does everything you've you've discussed with me impacts where you guys are investing? Sure. So we I would say we are investing across the AI ecosystem and have been for, you know, since the earliest days of I, um, you know, where I spend a majority of my time is actually on applied AI. And that's kind of what I think about as the deployment layer. Um, right. So we talked, I think, earlier today about for deployed engineers, you know, you see a lot of applied AI. Companies saying, how can I take this raw intelligence? How can I package it? You know, in many cases the answer would be let me let me get some forward deployed engineers into my customers, right. Figure out how companies are using the models, how they need to use the models, package it in a way that, you know, kind of makes sense. Um, so I spent a lot of my time on applied. I, we also spent a lot of time on, I would say, you know, unique models, unique, uh, tooling, everything that comes around, not including cyber security, of course. Um, you know, data memory, semiconductors really like the entire AI ecosystem. Um, and then I would say the other area we're super excited about is this physical AI, right? So we've had the, uh, kind of aha moment, if you will, uh, from a text modality perspective with lens. And now we see, you know, world models and unlocking physical intelligence or physical eye, um, also physical intelligence, it's a company. But and more broadly and I think that that's another really exciting wave that will know I'm following you completely. I mean, we just had, um, Lisa Su and Feifei Li here in this year. And now after that deal closed and, um, clearly AMD was thinking about it as well.