Venture with Grace
00:34:43 · Jay Zhao
[00:34:43] S1: kind of decisions. Yeah, yeah. So I think the older VC firm, I forgot what the other is, a sales of SMR, they have this term called Prepare Mind. I think it's a very old term, it sounds very old type of term, but I think the essence of it does stay, right? So what that means is that, for example, everybody has been talking about fiscal We have not placed bets in physical AI space yet for some concern about the whole subspace story, some part of that stack is very hyped with capital excitement, and the fact that the cap investors, most investors are not technical, they are attracted to a very impressive demo. And I think physical AI is one of the spaces that leave a lot of imagination for investors. But what we did as an example of that, right, internally we have a private research that we don't put up on the website and we don't share publicly, but it's like a 90 page of deck where we actually break down the whole value creation chain of physical AI. I think some of the opinions that we share internally, we share with some LPs and stuff, it's quite, I would say, should be quite censored. Because we have opinions about some public companies, we think it's very not deserved what they are right now. But to answer your question, so stuff like that, we have the form of a big understanding about the space. And in Fund III, we do want to play some bets in the physical AI space, but not in the traditional companies like the interactive world model layer. That part we actually feel pretty bearish about right now. But there is some other pocket opportunity. We think it's actually good timing to make some place bets on bets in 2017. And I think in that sense, kind of like Vipshop, kind of like MaintainX, in the early round, there's always something that the market does not appreciate or understand. In Fund one, one thing that we form as a thesis of the fund is that when everybody was talking about the early days of LM, we actually looked for more edge play, especially debt play. So MaintainX is a good example of that. They were actually rolling out into manufacturers like property maintenance and stuff. We think that can actually fit into LMs much easier and much more unique high value way because nobody was able pay attention to that. And now, like MaintainX, after acquisition by Autodesk, now it becomes Autodesk Industrial AI solution, right?
00:37:22 · Jay Zhao, Grace Gong
So that takes five years to kind of fully roll out. But it was not obvious early on. Same with Spectro Cloud. Spectro Cloud, at the time, the way we're looking at this, obviously, Temari and the team really fit with the shine score super well, like four star. They had access to Cisco. Wait, what happened to [00:37:42] S0: the five? What's the last mile that
00:37:45 · Jay Zhao
[00:37:45] S1: they did? They had a good access to Cisco, but it's not like a billion, like the typical headline, it was a type of exit, but a great exit regardless for the team. But the thing is that they were working on Kubernetes platform at the time. But the idea is that how can you help enterprises to deploy Kubernetes and then eventually do the edge deployment. So now fast forward to now, they've become the category leader in edge AI deployment, and their customers are US Air Force and GE Health and all the high compliance sensitive industries. But that was the investment that we made five years ago, and we try to be a little bit trying to find an area where we view a little bit different than the majority of the capital market, or at least one or two years before they come in. In Victor's case, it was competitive in some sense. It was not competitive in other sense. I will tell you why. It was competitive in the sense that the founders were researchers at Meta, so they were the pattern recognized founders. They got some good backings from European VCs at the time. Leonis was the only US VC that backed them at the C stage. And the reason is kind of interesting because the team was actually based in Europe, we were following their work when they writing papers and stuff. And when they happen to be in town for actually just five days, we'll reach out to them, we'll grab coffee, we send out report and say, Hey, we have to share the same world view with you, we want to build an end to end AI agent. This is what we think internally and let's talk. So for the next five days we just spent a lot of time talking about the tech stack, the product road mapping, and you know, it became one of the first investments we did Alpha two. They actually talked about other VCs in Valley at the time as well, but I think either nobody really pulled the trigger or they don't really buy into that worldview vision, but it ended up Leonis is only US VC firm on back the team at the time. But now they're kind of growing and they probably raised a new round as well. Now they've become pretty hot and obvious, so we try to apply that same methodology into Fund three and beyond.
00:40:07 · Grace Gong, Jay Zhao
[00:40:07] S0: I wonder, so how do you react to like the newer things? For example, I feel like I wouldn't call like AI Adrian something like super new, but like obviously I feel like it's just like within like this year, because of the like I would call it like blow up but like the blow up of like cloud or like so basically a lot like these like AI agent that can do a lot of crazy tasks like you know Instinct and kind of like many others and So how do you budget out the money that's for? Maybe it's something that's not in your prepared mind thesis, but that just happened to be a really hot thing everyone's chasing. [00:40:50] S1: Yeah, it's interesting. [00:40:52] S0: I think for us Obviously you're anti hypo
00:40:56 · Jay Zhao
[00:40:56] S1: a That's our swag we give other people. So it's interesting, we try to there's two mechanisms that we think. We think each fund lasts for two or three years for a reason, right? Because there is certain high confidence, eighty-ninety percent chance, you should be willing to make bold prediction of what's gonna happen in two or three years. It does not need to be nitty gritty detail on the product form and product shape, but you can start from first principle thinking and then infer some implications for companies. So for our Fund three, which we'll deploy from 2027 to 2030, we talk about like the agent will become more popular in both enterprise and personal setting. So it has implications for the Asian infrastructure side because that's right now is not a worldview out there are some companies doing this but the market is not fully mature yet. So this is one of the space that we will continue to monitor and obviously work with best founder in the space. But you need to have that world view to find companies and think about token economics and think about how does it moat being formed in the agent's era and agent space, which is something that we put out in the research post, called agent line graph. Along the way, right, so at Leonis as a forum, each partner every month that we have what we call the internal monthly learning session. So each one of us has to be a professor one day. We have to share a new area and new things that we're curious about and get feedback from other partners. So Physical AI is one example. It actually comes from an initial debate about hot deals that's being shared with us. That company, I would just say the first round of valuation is 2,000,000,000.
00:42:58 · Grace Gong, Jay Zhao
[00:42:58] S0: Only $2,000,000,000 [00:42:59] S1: I'm just kidding. [00:43:00] S0: I feel like Nodd is like the physical AI space is raising $2,000,000,000 at CNBC. Yeah, it's quite wild. [00:43:09] S1: So we got invited to the deal and then, you know, who is who is on the cap table it's one class of hot deal, right? So that actually started a pretty intense internal debate because question is that it's actually not so much about the valuation per se, it's like do we actually believe the tech direction is correct? If it's correct, then maybe even at $2,000,000,000 it can compound into a $20,000,000,000 deal, right? So there's a lot of back and forth debate within the partnership, and that translated into what we call the monthly lesson. So basically one partner will share and prepare and lead those discussions and that map out our deeper understanding about the physical AI space. And that will we'll start some experimental check into Neo4j in 2007. But that is kind of how we integrate Neo theses along with the three main theses that we have. And I think it makes sense for three years because after, you know, when it comes to 2029, 2028, I'm sure new things will come up, new types of things will be unlocked, and that just provides new building blocks for the builders, and that's why we will have new funds that back when they evolve thesis. I don't know if that answers your question [00:44:27] S0: or not, but... [00:44:28] S1: I would