So tell me, what exactly is Discovery Loop doing? Obviously, it's pretty still in stealth, but is there a sneak peek you can give us at, at all?

Yeah.

So I would start by just highlighting the, the really unprecedented caliber of the, of the founding team.

Um, there's four founders who collectively really built Google's technol- technical infrastructure and AI foundation over the past fifteen to thirty years.

Um, Jeff Dean will be the CEO, who was the chief scientist, um, at, at Alphabet, founded Google Brain, created the TPU, um, was, has been, been behind, uh, Google's ascendance in AI from the very beginning.

Uh, Oriol Vinyals is, uh, is also a longtime leader at Google and DeepMind on the AI side.

Um, he was the co-lead of Gemini most recently, but also, um, wa- was a lead researcher on a lot of Google's most important breakthrough research over the years, you know, dating back to things like the original Mixture of Experts paper or the Chinchilla paper, et cetera.

As a VC, like, how do you kind of think about doing these sorts of deals that are so different from the normal structure that an early stage investor would even go after?

How do you see that playing out?

Yeah.

My view is that over the long term, demand for intelligence, including artificial intelligence, will basically be infinite.

And so we will find good use for basically any GPU and AI hardware that we develop and build out.

Now, that's in the long term.

In the short term, there certainly can be ups and downs, and it would not at all surprise me if there is a market pullback and we see many of these data center projects be abandoned or financing kind of gets pulled and so forth.

Those sort of near-term market gyrations are inevitable when there's this big of a technology boom and an investment cycle the specific prediction that I made was that This question of depreciation schedules for chips, which seems like kind of an esoteric Accounting topic will actually become really important this year because and just maybe to provide some quick background generally when companies build out data centers The computing infrastructure, including the GPUs and the CPUs, are depreciated over a five or six year time horizon, which means that each year you are only recognizing about 20% of the overall cost of the GPUs as an expense, and you sort of depreciate it over five years.

Maybe talk to me about what you would recommend our CEOs or kind of our investment team should be focused on as they think about the durability of those vertical software companies and the transformation into vertical AI companies?

Mm-hmm.

Uh, and that actually has already been accomplished by BCIs, and I, I wrote about it in the article.

Uh, this, this guy Eddie Chang, who's, like, one of the leading BCI researchers in the world, he runs, uh, the neurosurgery department at UCSF.

He's a longtime UCSF professor, and his lab has demonstrated in recent years that you actually can implant, uh, BCI chips into people's brains, and you can tell what they're trying to say, and then you can turn, you can turn the brain signals into actual words.

Um, that obviously has just been done on, like, small scale research, uh, clinical study context.

It's obviously not, like, available commercially.

Um, and how do we get from there to actually we need to give AI rights?

Yeah, this is another one that I think should sound ridiculous when you first read it because, like, any good prediction should sound a little bit ridiculous looking into the future.

subspecific stuff

yeah yeah so the way we think about the universe of ai startups is as a technology stack with multiple layers um and we we invest up and down that technology stack so we're full stack investors we invest actively at each layer of that stack Jensen Huang, the NVIDIA CEO, has recently started to talk about AI as a five-layer cake, which closely mirrors the way we think about it.

I feel like we were thinking about it as layers in a stack before the five-layer cake analogy, but now everyone thinks of it as a five-layer cake.

But anyway, I think it's basically right that The way we think about it, at the bottom most layer is the energy and the physical infrastructure that powers AI, whether that's energy generation or data centers, equipment and technology for data centers, the energy grid, et cetera.

and then a layer above that we think it was the is the silicon the chips um you know the hardware to power to power ai and obviously nvidia is the is the dominant player in at that layer but there's a lot of interesting opportunities and new challengers and alternative approaches on the silicon side that you know we think are really exciting A layer above that I would, I would think of as like the transition from the world of atoms to the world of bits or going from hardware to software.

And I think the kind of the foundational software layer is the, is the models, the foundation models.

And of course a ton of, you know, probably the most famous well-known AI companies are at this layer, like OpenAI and Anthropic, but I think there's a ton of activity at the model layer beyond language.

Maybe we could start with like, how does the incubation work? And how does like Radical kind of like think about the strategy since like the early stage front is like 650 million and they're like, what does a portfolio construction look like compared to maybe the later stage front?

Rob, Rob, what was your whole reaction to the, you know, the apps are cooked and, and SaaSpocalypse, uh, you know, narrative of the past months?

Yeah, that's interesting.

I think, like, market narratives have a way of, like, swinging so far in one direction or the other to the, to the extent that a lot of the nuance is lost.

So I do think, like, a couple things are true simultaneously.

Like, I, I do think a lot of traditional software companies, and even categories potentially, are at real existential risk on account of the frontier labs and their, and their product roadmaps.

And so I think, like, a lot of that re-rating was rational.

Um, and there...

[laughs]

They're not gonna go into surgery for it, but more of like a biohacker use case.

Yeah.

That, that's exactly right.

And those use cases have already been pretty well proven.

Like there are actually non-invasive BCI products on the market today that like can, you know, detect if you're stressed or if you're, if you need to sleep more or, you know, if you need to focus better, those sorts of things.

I think the, the really big question, like I, in my opinion, the single most interesting strategic question in the field of BCI right now is like the conventional wisdom has been that invasive BCI is necessary for the most sophisticated use cases like language that we were talking about-

Mm-hmm.

I do think neural data, brain data is a new modality that we're just right at the very beginning phases of starting to research into.