00:00Great to have you here, Rudina, with Emily and, of course, with Ed.
00:04I'm curious about, you know, how this conversation around AI continues to evolve.
00:09We are now talking so much about the memory names.
00:12How do you see it?
00:14Well, hello, Carolyn.
00:16Good to see everyone.
00:17I think fundamentally we are seeing the overall demand for AI-related infrastructure and all
00:23the way up to the application layer continue to be strong.
00:25And in my view, the shifts that we are seeing in the market sort of reflect the scarcity
00:31or overestimation of scarcity along the value chain.
00:35So on the point around NVIDIA and others, are we seeing the correction because we expected
00:41more scarcity while on the hyperscalers are the valuations coming back because it's becoming
00:46a lot easier to track the sources of revenue that these players will have.
00:51No, by the way, we think of the hyperscalers as that sort of fundamental layer, the software
00:57infrastructure layer, but each of them are much more vertically integrated with their
01:01own chips.
01:02So in many ways, we're seeing the market move toward vertically integrated players from the
01:07chip layer all the way to the foundation models rather than just the CapEx spend.
01:14I don't know that we're out of the, you know, CapEx demand.
01:17I think that's a multi-multi-cycle, multi-year level of demand.
01:23I do think that perhaps we're correcting a bit for the overestimation of scarcity.
01:28You mentioned revenue, and I'm curious, Rudina, if you can talk a little bit more about just
01:32how critical it is for investors like you to be able to pinpoint whether the spending
01:39on AI is actually translating into tangible revenue and whether we've seen that yet.
01:46Thank you, Emily.
01:47I think it's interesting because I'm a very early stage investor, so I often say I back
01:52to brilliant researchers out of a lab and we build out a monetization plan.
01:57If anything, I would bifurcate between the next-gen application layer AI companies where
02:04not only are we seeing revenue, but the scale from zero to multi-million, hundreds of millions,
02:10millions is a matter of months rather than a matter of years.
02:15So if anything, we are seeing acceleration.
02:17Now, beneath that, two fundamental questions.
02:20Why?
02:21Is it because, especially on the enterprise side and even consumers, we are all experimenting
02:26and trying a lot of things.
02:28Thus, is the revenue sustainable or is it a leaky bucket?
02:31Second question, though, is really pertaining to the adoption of AI more systematically.
02:38So having moved beyond the, I'm going to experiment with this in-house build tool or with a chat
02:45function of Claude or ChatGPT to something more serious.
02:50In that case, we're seeing AI, demand for AI products to become much more of an important
02:57component, not just to the tech stack of enterprises, but to their actually business models and go
03:02to markets.
03:02You know, Radina, what's interesting about what you and the team at Glasswing are looking
03:06at, it's so timely for today with this Morgan Stanley hyperscaler story.
03:11So you are looking at AI native SaaS, various layers of the software stack.
03:15Every time I speak to one of the CEOs of the hyperscalers, I always ask them, you seem to
03:21be pitching and working on the same kinds of technologies that your customers already
03:27do.
03:28You know, so how do you respond to that?
03:30The idea that the hyperscalers, you know, the cloud computing companies, let's call them
03:34what they actually are, are going to want to offer the same tools that you might back at
03:39that early stage.
03:41So, thank you, Ed.
03:43The underlying question is, where is the moat for the startups relative to the incumbents?
03:48And I think the moat depends.
03:50Some of it has to do with, you know, vertical plays.
03:54You know, I recently backed a company called Modern Industrials.
03:58They are developing an end-to-end demand planning for distributors of lumber and building materials.
04:05Very, very specific.
04:06A tens of billions type of opportunity a year in terms of the revenue they can generate.
04:10But if Microsoft or Anthropic or OpenAI are going to that level of specificity, we have
04:17something to worry about.
04:19By the same token, who will be their customer?
04:21So, they have to draw a line to your very astute question around, where am I going to
04:26stop competing with my customers?
04:29And, oh, by the way, is there value in being highly, highly specialized?
04:33My view is that there is.
04:35And the value doesn't necessarily lie in the models.
04:37It actually lies in the highly specific, albeit not as broad, data sets, which then, you know,
04:44deliver much, you know, superior outputs.
04:46I would say if I'm a pharmaceutical maker or, like, in the medical field, right?
04:50And, like, I would say that I want a data set that I'm going to play with that's very
04:54related to my field.
04:56I don't need to know about the rest of the stuff that some of the large language models
04:59are doing.
04:59Yeah.
05:00Or they…
05:00A hundred percent.
05:01Or, Radina, they, you know, something you said a moment ago is really interesting.
05:05You are backing not companies necessarily.
05:08We'll work out the business plan later.
05:09You're backing the researchers out of the lab.
05:12And, like, that applies at the late stage as well, right?
05:15The founder-led thesis.
05:17But those people also have specialisms, right?
05:20Does it matter to you if they have a core team of not just computer scientists, but in
05:26biology, biochemistry, et cetera?
05:30Yeah.
05:31So let me parse what I intended with what I shared.
05:34In the case of the modern industrial, I'm very much going in with a very clear understanding
05:40of how it's monetized, what's the ROI for the customer.
05:42So in many ways, that's extremely tangible and very clear around the TAM pricing competitive
05:48dynamics.
05:49My earlier comment related more to what I would call frontier tech.
05:54So we have already achieved what we have achieved with the large language models.
05:58What comes next?
06:00You know, we backed a company called Recursive AI.
06:02It's about the models correcting themselves to get around the next set of issues like hallucination
06:08and other problems that the large language models have.
06:11There's a whole paradigm of neo-labs and new researchers that are bypassing in many ways
06:18the capabilities of the current incumbents.
06:22Think about Liquid AI, multi-multi-multi-billion-dollar company.
06:25They're actually doing all the processing at the edge.
06:29It's actually a whole different sort of way to bypass the need that we currently have on
06:35compute.
06:35So in those cases, you solve one of the compute problems or the energy problem or the sort
06:42of the performance of the models, and then you worry about monetization.
06:47So it's not pie in the sky.
06:48I'll beg the smart people, and someday they will deliver something.
06:51It has a purpose, but you might not know the pricing.
06:54Kind of like how many of us would have guessed that pure consumption pricing was going to be
06:59the model.
06:59Does that make sense?
07:00One contemporary difference, though, is that that group of three or four people can raise
07:05a billion dollars out of the gate or several hundred million dollar seed round at crazy
07:10valuation based on the value of their own intellectual capital.
07:14How does Glasswing even participate in early stage valuations like that?
07:20Very good question.
07:21So our core product is actually the more typical pre-seed stage rounds.
07:27We have a pool of capital and a product called Access Checks, where we write, you know, small
07:32one to two million dollar access checks in these gigantic rounds to have a seat at the
07:37table, to have a seat to the ecosystem, to support the founders and participate for, you
07:41know, with those that actually make it big.
07:44So while our core checks are focused on building the companies from the ground up in the more
07:49typical fashion that you think of early stage venture, we absolutely get access to the
07:53big guys.
07:54And it's important.
07:54There are benefits for both.
07:56Hey, speaking of the big guys, I just got time for maybe one or two more last questions.
08:00I'm thinking about SpaceX, which does go into the NASDAQ 100 at the end of trading today.
08:04XAI, like what's your thought about that company, which seems to be to some extent coming from
08:11behind, but it's got certainly deep, deep pockets.
08:15It's emerging more and more.
08:17I think, you know, in many ways, I think of Elon's companies as all being intertwined with
08:22one holding and the holding being Elon himself.
08:25I think it will be interesting to do to see what he does with XAI vis-a-vis SpaceX and
08:30maybe even what happens with Tesla eventually.
08:33They are catching up.
08:34I think performance has improved.
08:37And to be perfectly honest, in the circles that sort of are leveraging the most cutting
08:42edge models, XAI is starting to become a real contender.
08:47Definitely coming from behind, though.
08:48Are people talking more about XAI?
08:51Yeah, increasingly because of Cursor, right?
08:54Like the Cursor acquisition moving so quickly, people pay close attention because it's such
08:58a widely used tool.
08:59All right.
09:00And yet, even with Cursor now, you know, the next generation saying we've bypassed Cursor.
09:06So things are moving so very fast.
09:08What becomes, I go back to what becomes sustainable versus what is the flavor du jour?
09:14We shall see.
09:16You know, and it's fascinating, too.
09:18Rudina, thank you so much.
09:19Rudina Ciceri, founder and managing partner at Glass Wing Ventures.
09:22We always appreciate your view.
09:23I have to say, I had more and more conversations I have with folks where companies are using
09:28a lot of different LLMs.
09:31They're saying they're starting to get more discretionary in terms of who's got access to
09:34what because there is a cost.
09:35Well, real quick, the reason people loved Cursor is you could swap in Anthropic, OpenAI,
09:39Gemini model and change it.
09:41Now Elon owns it.
09:42They're going to be allowed to do that?
09:44It's a big question.
Comments