Jenny Xiao, Leonis Capital Partner on AI Investing and Emerging Talent

Grace Gong (0:01)

We are live. Hi, Jenny. Welcome to Venture with Grace.

Jenny Xiao (0:04)

Yeah, I've been listening to your show for a while, so really excited to finally talk to you here.

Grace Gong (0:11)

I'm so excited for our conversation. Before we start, I want to give a quick shout out to our amazing sponsor. This episode is brought to you by Nebius, the ultimate cloud for AI innovators. Nebius provides AI infrastructure you can count on, combining reliability and speed with flexibility and engineering support unmatched by hyperscalers. AI leaders like Meta, Shopify, and Higgsfield already partner with Nebius to run their AI workloads. Plus, venture-backed startup can save up to 150K on compute costs when they apply for access. Visit nebius.com or nebius.com/startup to learn more.

Jenny, I want to start with your background. You're a researcher at OpenAI, you research on agents before agent is hot.

I want to start with, what were some core things that you have learned through your researcher era before getting to venture?

Jenny Xiao (0:58)

Yeah. I think the biggest takeaway that I had from my OpenAI era was really that a good AI product is supposed to unlock capabilities of the foundation model. AI products and AI applications fundamentally are there to amplify a foundation model. The model really is the star of the show. I think that's my biggest takeaway. The other biggest takeaway that's relevant for my choice into going to venture is that when I was at OpenAI, one of the things that I realized pretty quickly is that VCs don't really understand AI and this is like 2021, 2022 This is before everyone's heard of LLMs or foundation models. A lot of VCs at the time were still talking about blockchain, they were still talking about crypto.

In my view, they were totally blind about the AI revolution. I think even today, a lot of VCs are at least three to five years behind the technical frontier. A lot of the hottest topics that we're talking about in AI today are really old topics in the research world. Those are my two takeaways from my OpenAI era.

Grace Gong (2:07)

I love that.

I want to start with the foundational model, and you mentioned AI unlock the foundation models, which is totally true. So I guess applications will help you unlock the foundational model. I guess when you're thinking about investing, how do you think about the model war that's going on here versus application? Because I think nowadays in the AI land, there are obviously, though, everyone is saying foundational model doesn't really necessarily going to have much of like an investible opportunity because everything is so expensive. But yet, most of the money are still going into a few big laps. What's your thought on from investing perspective, what is interesting to you?

Jenny Xiao (2:56)

Yeah. So first of all, we primarily do precedence seed investing. We actually also did a check into AnthropA through a side vehicle, but not through our main fund.

I think there's a lot of value that will be captured by the foundation models. I think there will be continuous value capture in the closed source model front, even though open source models are becoming a lot better, really quickly. I still think my closed source models are going to capture a lot of value. There's a lot of value in being the number one foundation model. A lot of people will just use like an OpenAI or an ethnography model without thinking too much about cost. So there's definitely value capture there. From a startup investing perspective, and this is where I'm going to focus on because this is my day-to-day job, I think the best startups are usually the ones that are ahead of their time. The best AI applications are always, almost always ahead of their time. If you look at Cursor, they started building an AI coding back in 2021, 2022

This is like almost two years before coding took off as a space. If you look at Lovable, they started as GPD engineer almost a year, I think actually more than a year before Vibe coding became a thing. If you look at our own portfolio, we were early investors in Victor. They started thinking about AI employees even back in 2024 This is two years before AI employees took off as a category. This goes back to my earlier point, the best AI applications are the ones that anticipate the next model capability breakthrough, and they build ahead of their time. They're building something that's totally transformational instead of like, hey, let's do this incremental thing that can fix a certain limitation of a foundation model. They're really thinking about where does the next capability unlock happen and how do we productize that unlock? I think because of that, there's still going to be a lot of value capture happening in the application layer as well.