Martin Casado: AI market is expanding, not consolidating — open and closed source will both win big

discuss open source AI, his views there, what's cursor doing. Uh, welcome to the stream, Martin. So good to have you here. How you doing? Good. Thank you. You got to welcome some live the live audience. The live audience is really just they couldn't control it. They're out of control. Yeah. Yeah. Anyway, thanks so much.

Yeah. Yeah. Thank you so much for uh for joining us. Um I'd love to know your take. We Let's just jump right into it since we're running a little bit behind. Um cursor uh obviously it's becoming this front-end almost aggregation play. It's the front end to AI coding. It's uh potentially model agnostic.

Uh what does that mean? What have you learned from the cursor journey around how value occurs across application layer versus foundation model layer? Yeah. So um I mean I think that there's two things you know that that led up to this cursor moment. And one of them is one of the first use cases for AI was code, right?

we had like Microsoft Copil and a lot of people used it. Um but like and so like you know there there was you know a bunch of users and their behavior was trained by Microsoft but the models weren't quite good yet and um in the last you know year we had this RL movement. Mhm.

And as a result, the models got way way better. And it actually just happened to have like this kind of magic timing where it, you know, it did something very similar to what Cold Pilot looked like, but you know, use these greater models.

And now that they've kind of caught the wave, they're able to, you know, not only serve these amazing models, but like for every magic experience you see the the the users have, they've starting to like have their own smaller models.

And so it's just you know very unique point in time very unique uh position and then of course a very unique technology way behind it. Going back to the foundation model layer uh can you talk about how your thinking around opensource in the foundation model wars has evolved.

Uh I remember you and John Ludig were talking about it last year. Since then, we've had Llama 4, which kind of fell backwards, but then DeepSeek was a huge moment.

And uh we just had a a guest on the show yesterday who was talking about the importance of offering the world uh open- source foundation models that they can build on because if we don't do it, our near-per adversaries might. How is your thinking evolved if at all?

And where are you seeing the most opportunity in the foundation model wars? Yeah. So it's not clear to me that the model space is significantly different than you know the software space when it comes to open source. So in the software space we've got you know 20 years of history of this.

Um you know the closed solutions tend to aggregate the value first and the quickest and the reason is is you kind of need to make money to change people's behavior right and you know this is like you know you need to have Unix before you have Linux and you need to have Oracle before you have my SQL etc.

And the same thing seems to me to be playing out in in in this AI space which is you know of course open AI was first and they've aggregated a ton of value and you know anthropic for code and um and then you know even if you look at the non- language models like you know image you know midjourney was first and that's closed and 11 labs and so it seems to me like the first movers will close source and they'll continue to close source and they'll aggregate a lot of value but then for every first mover there's always the opportunity to enable the masses I mean like everybody else.

Uh and there's always a lot of value there, right? And this is kind of where these open source models come in. And what's amazing and everybody should realize is just because Deepseek is out there hasn't changed the calculus or the growth of these other companies, right? Is and fully are creative.

And so I think the only sin anybody can make in AI from an investor standpoint is zero someum thinking because the reality is is every time something new gets introduced, it's not at the expense of something else. it kinds of finds new territory to gain value in and and open source is no different.

So it's a very important part of an ecosystem. It's been part of the software ecosystem for a very long time. It'll continue to be I do think that you know us um should have open source models and and I think it's very important for like continue to go presence in technology in that conversation.

Um but it feels very healthy to me um both the dynamic with existing closed companies and then you know new companies that can be built on on the open source. Yeah. Are you a fan of aggregation theory or this idea of like the front door to artificial intelligence?

Cursor is kind of that for coding and we're seeing that in other applications that are popping up where uh the underlying model might not even matter. And my where I'm going with this is that there's one world where uh another country wants their own foundation model.

So they know that the weights are trained in a way that they want aligned with their views on speech for example.

Uh but there's another world where the individual countries should be thinking about how do we create the dominant consumer app that people actually use because even if we have the model if nobody's using it because they're just opening up a certain URL and they go to another country's model. What do you think of there?

I mean I think I think the story of AI has been the markets are much larger and they're growing much faster than anybody had expected and that's just resulted in actual fragmentation. So, everybody had these theories early on, if you'll remember. Oh, there'll only be one model. Oh, the models are not defensible.

Oh, the apps are just GPT rappers. Oh, like the apps are going to take all the value. And you know what? They're all wrong. Like, everybody's been wrong. Like, the only have been right are like, you know what? These things are massive.

I mean, let's just take OpenAI, which is one of the most remarkable companies and maybe forever in tech, right? But, you know, they were the first to image. Remember that with Dolly? They lost. Yeah. Yeah. They were the first to code with Copilot. They were the weights.

they lost that they were the first of video and like you know Sora was amazing and remarkable but like they're not the leaders in video um and yet they're still the dominant AI company on language right and so what are the takeaways you can take from that like in my view it's you know every one of these is a massive massive market that's too big for one company and so like you know open a great job focusing on chat GPT and you could have the same discussion between the model and the app layer right and so I know I mean you're talking about aggregation theory so there's two views of the world one view of the world is like everything consolidates into one company and there's another view which is there's like a model per thing that's a behavior I'm definitely in the second one and I just you know even though we're relatively early we do have four years of experience with this and the story has just been disagregation it really has yeah that makes a lot of sense how are you uh how do you think about revenue quality in the enterprise in an environment when you know the fortune 500 sort of has a mandate to just you know buy AI some some things they're going to that's a great yeah so okay so so there's there's a lot to this question and I'll try to be as succinct as possible so the first most important thing is um AI today is mostly a proumer and consumer movement so even though the people are working in the enterprise it's like it's an individual use and individual behavior and it's not like a big budget buying thing and so the I would say the vast majority of the companies we work with it's this more kind of bottoms up proumer thing and so that's clearly real individual demand This happens in all super cycles.

Like it was the same thing for the internet, right? It was the same thing for PCs. It was the same thing for like the iPhone, like the true smartphone like post Blackberry. It was like individuals. And so, you know, from that standpoint, like I think we're very very comfortable.

This a real secular movement because it is individuals that are driving the behavior. Um, you know, there there there's there's a second point to all of this.

um which is um in the early internet days you had all sorts of funny accounting cuz you know and it was quite frankly cuz people just didn't know how to do it successful but at the time people weren't even paying right so they would just kind of say everything and and of course in these early super cycles you have all of that like they don't know the difference between like GMV and take rate and you know run rate is not a gap thing and so they kind of had deal there's clearly all these funny numbers out there um and you know that's just an artifact of massive massive growth and success and like you know we'll rationalize it as an industry but for sure that's going on but it's less about like enterprise budget moving and it's much more about I've got this consumer phenomenon I have no idea how to really do the accounting you know and so you know it just requires from our standpoint you know just to kind of go through things and kind of really normalize it that's great uh I mean we could talk for another hour about this we'd love to have you back on the show this is fantastic uh but thank you so much for even taking the time that you did more time this was great Love it.

Thanks so much. It was a real pleasure. Cheers. Have fun. Cheers. Bye. Um, yeah, we didn't even get to Feay Lee and World Labs, the deal that he led there. Um, he spatial intelligence Martin, you know, I've seen his post, never heard him talk. Potentially a Hall of Fame Hall of Fame yapper.

Can we get a soundboard for that? Hall of Fame yapper for Okay, we got David George leaves the Growth Fund coming in next. I'm going to want the Ashton Hall sound effect. I'm going to want lots of sound effects for this one. We're getting too serious right now. We got to take it down a notch.

Everyone thinks, "Oh, they managed 40 billion. " You got to be serious. You got to be on Talking Points. We have a fully executed lease on our new studio. Oh, let's hear it for us. Let's hear it for TVPN in the chat. Oh, thank you. And you know why that's exciting, John? Because we're going to be walk.

We're going to be able to walk around sitting for 4 hours. Sitting for 4 hours is brutal. Brian Johnson is shaking his fist. Yes. Thank goodness. sitting is is, you know, don't they compare it to smoking cigarettes in terms of, you know, the terrible body. It's terrible for you.

And uh we're going to be walking around hitting the biggest gong you've ever seen that we've yet to reveal on the show. Uh we'll be whacking that. We're going to be you're going to be height mogging me, but I had an idea. We can put a uh like a measuring tape up on the wall so that I