Redpoint's Jacob Effron on AI's next application wave, robotics data breakthroughs, and valuation reality

guest is already in the reream waiting room. We have Jacob Efron from Redpoint.

There he is.

There he is.

Hey, how you doing guys? Great to be on. Great to finally have you on. Got it in before the end of the year. Welcome to the show.

It's an an honor to be here and a very very happy holidays to you guys.

Thank you. Thank you. Uh what's going on in your world? Maybe maybe uh first time on the show. Quick quick introduction uh for the folks watching.

Yeah. Uh I'm Jacob. Uh I'm a managing director over at Redpoint. We're a $6 billion fund and I co-run our our early growth fund. Uh which sounds like an oxymoron, but basically means uh series B and C. Uh so that's where I spend all my time uh and do a lot of our AI investing across a bunch of companies.

What's what's your favorite?

Uh you you know you can't answer that question.

I think you can answer it. I think I think it's uh I have a friend who does A's and B's uh and like he has like a very very clear a very clear clear preference around like one of I'm I'm not going to I'm not going to I'm not going to you know share his alpha but like he has a lot of reasons about liking one versus the other and I think I think they track but I'm curious how you think of that B to B TOC range if there's like a what

I will say the first the first investment I ever worked on at Redpoint point was uh was show sponsor Ramp uh which is sitting nicely up there in the uh in the top right. And so I feel like that that set the bar pretty high uh for uh for team velocity.

Was that the Was that the B or the C that you guys

That was the B. Yeah,

that was the B. Yeah. Yeah. No, I I uh I remember those days you guys were had a lot of conviction and it uh it certainly panned out. Um what uh yeah, what like how are you kind of reflecting on this year? Yeah, it's been uh it's it's been been quite the year, right? I think we've uh you know on the application side, I feel like there's been a tremendous amount of stuff that's really started to work. I feel like you know obviously coding being the uh the killer use case, but there's been a ton in customer support, healthcare, legal. Um and I think the big question going into next year is uh what's what's the next set of dominoes to fall on the application side? um you know are we in a time period where models have uh are are relatively plateauing and it's going to be this set of uh of applications for a bit or uh are we going to see some improvements and then get a whole next wave of of application areas uh unlocked. Um at the same time there's been a ton of interesting stuff on the model side uh you know outside of of core LLMs. Obviously you guys have talked at length about some of the cool stuff happening in the image and video world but even in robotics biology material sciences it's been fascinating to see. Uh and then next year I'm really fired up about infrastructure actually. I think with the with the models slowing down a bit uh I actually think it's a great opportunity. There's finally a stable set of things to to build infrastructure for. And so uh I think it'll be really really interesting 26.

Do you think that there's any chance that we'll get a new consumer AI company out of a company that didn't start as a consumer AI company? I'm just fascinated by this this

like OpenAI part two. Oh, I mean I mean that is one of my hottest takes is is that may maybe maybe the OpenAI nonprofit spawns another forprofit because perhaps there's something just innately uh you know like like just there's no better place to just do unstructured research and start a company than a massively funded nonprofit. It happened once in history, maybe it happens again. Um but but I mean to be clear, I was talking more about like there's there's these like pseudo midjourney. There's this interesting uh there's this interesting wave of companies where they have a model and they turn it loose and it seems like people are kind of playing with it themselves and and everyone reads on immediately as oh job displacement that's a proumer tool that's an enterprise product and maybe that's right maybe there is a big enterprise opportunity or big proumer opportunity or B2B opportunity but I'm I'm more interested in in these new patterns of consumer behavior right now and and and trying to find areas where um where where consumers just might land. But I I don't know if you've even looked at any of that or

Yeah. No, we've looked at all these and I think um there's obviously a ton of of interesting new consumer behavior around you know uh remixing these models, playing around with them. I think uh especially I mean we we we play around with that uh with that tool all the time. Um and so I do think that there uh you know I think that the key will be and you're starting to see this with uh OpenAI and Disney and even Sunno which uh had an arrangement with Universal recently. It's like bringing IP onto these platforms, I think, will just be fascinating, right? Because what's the It's fun to remix something that is, you know, maybe a friend or somewhat abstract. It's a lot more fun to remix something that involves uh, you know, Taylor Swift or uh, you know, Darth Vader or some form of IP.

How do you look at that category evolving? because I was debating uh somebody Monday night who has managed some of the biggest global superstars and we were talking about music creation as a category and consumer AI and this kind of proumer category and I was taking the side that like I like you have to assume that every lab or not every lab but a lot of the big players get into you can imagine Gemini letting you make uh music OpenAI every I think a lot of people will get into this because it's just like a it's a great hook and it's something consumers clearly want. I was much more on the side of like I think music is like nuanced enough and like there's a lot of depth to it and I think it deserves a standalone app and I think that there's a lot of people that will just pay for the Sunno experience for a long time even if they have some of these other subscriptions. Meanwhile, like clearly like you don't need like the average person is not going to necessarily have like an image gen app and then also like their normal LLM. So I'm curious you guys you guys are obviously quite uh biased here given your position in Sunno but like how you see that structure evolving.

Uh we're not investors in Sunno just a fan but uh really sorry

I'm from Boston and so anytime there's like an epic Boston based company I get very excited and so being from there uh is uh is is awesome. But I think that uh to your question you know I I I agree with you like it it probably doesn't live within uh a larger consumer app. I think if anything what you're going to see is there's a whole set of experiences to be built around how artists interact with their fans through these platforms. And I think that when someone captures a lot of the volume of folks that are interested in this, that's where the artist will go. It's not like an artist is going to go to 10 different, you know, to Gemini to Claude to Chatbt where music's embedded and like, you know, have back and forth with fans that are, you know, remixing their songs and interacting with it. I think that the, you know, uh, the IP holders here will go to the consumer, uh, landing page that aggregates as many people as possible and so I actually think it makes a ton of sense as a as a space for a standalone category because it's not really just about the quality of the model, right?

Yeah,

totally. Uh, sorry.

I wanted to chat was asking about pricing and I was curious too how you guys are how you guys handle some of these companies that are being formed have are getting off the ground basically with like series A pricing out the gates. How do you guys handle that at Redpoint when you have an early stage team and then you're doing early growth and you're like where do these even fall in? No,

you incorporate the company and you say I don't want to talk to the Redpoint seed investors. I want to do a growth round on day one. I want to talk to Jacob. That would happen from day one. They call I don't have an idea, but I need a growth round right [laughter] now. Put a hundred million in my account.

Yeah. I mean, this is where, you know, I think it's just one of the really fun parts of working at a a small firm, right? There's, you know, there's eight of us that are that just work super closely together uh on all this stuff. And, you know, a company can be a seed one day and five days later be a series B. Uh and so we, you know, I think it involves us just working super closely together. But it's, you know, it's it's an interesting time, right? It's both uh a crazy valuation environment and also uh these markets that folks are going after is just absolutely massive. And so I think you have to hold both of those uh both of those truths simultaneously.

Okay. Uh on the valuations broadly, um just in the last half of this year, we've seen uh Oracle sell off post big excitement around the backlog and the RPO. We've seen Coree sell off. We've seen uh Rich Sutton and uh Andre Karpathy and Ilasgiver on the Doresh Patel podcast uh talking about how maybe the models are slowing down. Uh is any of that

though I love that Andre's take was taken as bearish. He was like I think this stuff will automate all enterprise work but it will take a [clears throat] decade and everyone's like oh it's over. Like if it takes a decade I don't

Well, there's the question is that is that is that how is how is the venture community? How is how how are all of those different data points? Some of them bullish, some of them bearish, potentially many of them up for interpretation, how are they being interpreted in series A, B, C prices?

Yeah. I mean, I think what's what's very clear is there's both going to be, you know, massive uh massive new companies created in this wave. And I think even with just the capabilities of models, even if even if you believe that models aren't going to get that much better without some algorithmic breakthrough and you believe that we're overbuilding on the data center side, it doesn't change the fact that this current generation of models uh works really well in a bunch of different application areas. And so I do think that like on on the app side there's going to be massive businesses built. They'll require a bunch of infrastructure behind them. Uh I think the you know you were talking earlier with Sarah about some of the Neolabs. I think that's like the most interesting question of you know what what is the form of model progress going to be in the next five t years and is that something that occurs within the labs or or outside of them but I don't think that uh anyone's questioning whether really large companies will be curated today

yeah um what about the overall venture comm uh the venture e like the the amount of dollars flowing into venture funds right now I'm sure you have a view and you have a couple data points there's this chart that looks incredibly bearish It looks like it's completely over for venture capital and yet that doesn't match with anything I was feeling because there's huge rounds happening. I mean, I guess it doesn't count as venture capital, but open AAI is raising 10 billion from Amazon. Like there's big deals getting done, but for some reason there's a chart that shows venture dollars going down. How did you process Did you see that chart? Did you process that the same way I did? Uh walk me through your interpretation of that. What's going on?

Yeah, I think you're spot on on the vibes. it didn't it didn't totally match uh what it feels like the game on the field is. I do think there's probably, you know, it's definitely harder for uh emerging managers or first-time funds and so it doesn't surprise me that the overall count of funds was was down a bit. But like, you know, a lot of that data can be skewed by some pretty outlier large funds that were raised in kind of 20 21 as well as the fact that those funds in 21 were deployed in like, you know, 6 12 months or something, right? And so I think actually maybe we're back to slightly more normal fund deployment bases.

Yeah. [laughter] And then do you think there's also uh a shift over the last few years uh to more SPV usage amongst fund managers so that maybe the maybe the dollar value that's flowing through is high or is growing right now this year certainly is

look at look at rate we just had Doug on from Radiant the two funds that led yeah are not

to my knowledge

multi-billion dollar funds and yet he raised $300 million that had to come from

assuming that came almost entirely from SPVS.

Sure. So yeah, are are you seeing an uptick broadly? How does your firm have a view on SPVS?

Yeah, I mean I think there definitely uh you know seems to be both SPV activity as well as uh you know Nvidia like you know large uh the hyperscalers a lot more investment coming in from from all sorts of other places. I I think for us you know we we keep our fund size focused. We're, you know, we want to be, uh, you know, direct investors and and long-term partners for the companies we work with. But I do think, uh, you know, I do think that's probably part of the story here, too.

Yeah. Do you think there's any are funds losing like uh losing rounds to hyperscalers? Like, oh, yeah, I put in a term sheet and they went with Nvidia instead of me. I think usually these rounds are are kind of a combination of uh you know traditional uh venture firms and and some of the other folks but obviously people have different incentives in those arrangements and so uh you know it can change uh change the pricing sometimes.

Yeah. Yeah, that makes a lot of sense. Uh

what are you uh what are you expecting out of uh robotics broadly next year or maybe the next two years?

Did you guys see the news from from physical intelligence today? It's actually pretty nuts. What? Yeah. Um

I didn't I didn't see it. Sorry, break it down for the audience.

Yeah. So, for those that didn't see, uh, and I am not a robotics AI researcher to be clear, but I'll do my best to play one. Um,

you guys are investors in physical intelligence, so expert.

Let's ring the gun.

Exactly. Exactly.

I love that. So what they basically what they basically showed uh is uh there's been this question for a long time of can you use you know egocentric human video data to improve robotics models cuz god we have a lot of of human video uh what we don't have a lot of is uh teleoperated data and other data that's more difficult to gather for robotics and so what they were able to show is that basically they they built these models already uh on you know they had this model pi.5 that they built using a bunch of teleoper operated data and once the model got to a certain ability they then threw a bunch of egocentric human video data at it and it actually made the model way better and this was what's called an emergent capability. This capability didn't happen when the model was way worse. And what's exciting about it is it kind of parallels actually a lot of what's happening in in reinforcement learning right now. In, you know, after Alph Go, there was this huge push to to do reinforcement learning for everything. And it didn't really work because the base models weren't powerful enough. But then fast forward, we did this massive pre-training. We got these much more powerful base models. And then reinforcement learning started working really well. And so it seems like as we're starting to scale some of these robotics models, uh we're starting to see some pretty cool effects, uh of the ability to apply, uh video data, which we have a lot more of.

Yeah. Yeah. Yeah. That Yeah, that makes a lot of sense. It's awesome. Um

yeah, those guys every any uh any predictions around just like I feel like for people you see stuff coming out of labs, it's really exciting, but for people to really start to like actually believe that robotics are going to transform our entire world, right? You have to like get Zipline delivered to your house and have that like

moment, right? You have to be you have to be in a Whimo. Uh

yeah, everyone seems to be waiting for like the touring test of robotics is like do my dishes and and it's like it's unclear that that's going to be the first really big business in robotics. Like it could be something else that comes out, but we already have the robot in the house. It's a vacuum cleaner. Um, you know, a variety of companies have duked it out there, but like we we have some robotics that deliver some value, but we want something that's like the full humanoid will probably accept the dishwasher robot in the interim with like one arm.

I'll I'll definitely accept the dishwasher robot. Like, I'm here for that. The laundry robot like dishwasher. Uh, sounds sounds

dishwasher robot. I'll sell you a dishwasher. [laughter] It's an AI dishwasher. 20 grand. You want it?

Yeah.

Yeah.

For for under my Christmas tree, I I'll take it. Uh but no, I I do think that uh you know, basically as these models get better, we'll start to see some pretty interesting uh probably to start in like you know, some of the enterprise use cases where you control the environments a bit more uh and it's less uh on the home side. But I think there's going to be a ton of really interesting stuff people build on top of the models. To be clear, there still needs to be improvement uh in in the underlying models until we get there. But I think it's so clear everyone from the government to the largest uh companies in our economy are are are quite focused on robotics right now and and rightly so. Have you looked up have you looked at other AI hardware stuff? It there's this weird uh phenomenon happening where there's a lot of AI hardware companies that go viral, get tons of press, and then it doesn't seem like there's massive sales in the short term at least. Uh and then we wound up finding some company that was just doing basically AI device for note-taking. It just records and gives you the notes. It's a very simple like you know you very I can explain you you already know exactly what I'm talking about in terms of the product value prop and they're making like a hund00 million in sales maybe 250 or something already. Have you looked at that? Have you thought about how uh like the future like AI hardware plays out or is it scary because uh the big device companies like Apple might want to just go after that and just eat that at some point.

Well, I think all all big markets uh the big companies will want to go after, right? And so nothing nothing uh I think that's the price of it being really interesting. Uh we're we're investors in a company called Sesame that is going after you know uh both kind of better voice models but also like you know the the AI glasses world and I think that uh it's interesting to see so many people coales on that as like the form factor for uh for a bunch of you know of these products going forward. And so I think that like that is hugely exciting and I would be really surprised in the next few years if there wasn't uh much more prolific usage of hardware that has a bunch of these AI capabilities built into it.

Yeah. No, no, it's exciting stuff. Um Jordan, anything else?

The amount of glasses that are going to be available in a few years is is wild. Probably going to Whereas Sesame, Snap,

I mean,

Apple,

that's less than you walk into a sunglass hut, there's 25 brands. Why not 25 brands of electronic products?

Yeah. No, no. I I I just think it's exciting. It's going to be very cool.

It's going to be pretty sweet.

Yeah.

Yeah.

Yeah. We we got the demo uh Monday of the new uh new the snap

spectacles

uh specs that are coming out next year that are dev only right now and uh we were running around everywhere here having having a lot of fun.

I mean the are you guys going to rock them on the show next year?

The augmented reality is rock they're very chunky. They're they're big like and [laughter] and I I I've actually thought about this like like

no the developer version is not what they're releasing this next year but um uh I like wearing augmented reality glasses. I mean I guess I could have a screen. It's like I could kind of use it as like a teleprompter like I I sometimes I will uh like text with the team on my laptop here. Um oh is the next guest you know on time or do I need to hang out in this interview longer? Not to break the fourth wall here, but um and I could imagine I could imagine glasses. I hope you're not right now.

I'm not. But uh but Logan from Google is actually a few minutes late, so we can hang out here and chat. Um but uh and then maybe it would be fun to have a camera on there that you could feed into the show. So you could be like, "Let's go to the John cam." Because sometimes we go to the the gong cam and imagine if you could watch me hit the gong in first person as Santa Claus. Like that sounds like entertainment. Um, of course I'm a sort of a niche use caser use uh for sort of a niche a niche user, but uh I don't know. I just came up with two pretty cool use cases. I I I think it's going to be a fun future.

Yeah. I think the fun part about this stuff I mean it reminds me of the models. It's like you put it out in the world and you can't begin to imagine the way that people are going to use it, right? And I think that once you have a product that uh meets the aesthetic bar of something you'd actually wear, people are going to use it in all sorts of ways that I'm sure anyone making these things couldn't possibly predict. In the same way that I love whenever the model companies release a model, it's like they have some idea, but they're always day one shocked at like all these, you know, uh things that that work about it or ways people use it. They had no idea.

Yeah. Um last question. Uh with uh with all the competition in the foundation model space, um it does seem like there's a lot of startups that you've probably invested in that are spending a lot of money with the foundation labs. Have you seen that manifest in gross profits? There was a data point from notion in the Wall Street Journal that their gross profits went from like 95 to 85% or something. It was like very slight gross margin compression. Um certainly not cause for concern, but have you do do you feel like you have a view on what the impact of of you know high usage LLM inference might be for growth stage software companies uh on gross margins? Yeah, I mean I think for uh the the thing that makes it challenging to uh to know where it all shakes out is just the price of these models changes so much every you know two three months. It just keeps going down. And so you know if we stay on this like treadmill of capability improvement then maybe you're always paying for uh for the most cutting edge thing. But I do think that uh for any thing that one of our companies was doing 6 months ago it's gotten 10 times cheaper. Now they may be doing a whole new set of things as well but take any set of capabilities. I think the gross margin for that set of capabilities uh will end up being you know being pretty good uh just given the competition that exists in the foundation model space and just the insane rate at which this stuff keeps getting cheaper and we're going to see the same stuff happen in video uh in other spaces and I think that'll be incredibly exciting too for for folks building on top.

That's good to hear. Well, thank you so much for coming on the show.

Thanks for having me day. Yeah, we're gonna have a great time. Uh well uh have a great rest of your day and we will talk to you soon. Yeah, great great hanging jacket.

Sounds good.

Good talk so cheers.

Bye.

Um, let me tell you about eight.com. Exceptional sleep without exception. Fall asleep faster, sleep deeper, wake up energized.

I had kind of a brutal night.

Do you have a rough night? That's too bad. I'm sorry. Well, maybe you'll need to

63 4 hours.

Maybe you'll need to book a wander in the meantime.

Book a wander. Inspiring views, hotel grooming, hotel grade amenities, dream beds, top tier cleaning, and 24/7 concier service.

I forgot to ask Jacob what he felt is an appropriate tip. Founders end of year holiday tip.

We didn't ask Sarah either. Uh, I got an 89, so let's go.

Wow. I have some breaking news.

Breaking news.

Uh, Jared Isaacman confirmed.

Let's go online. um

friendly brain together.

I'm I'm I'm very excited. You know, whether the data centers go to space or not, there's obviously a lot of interesting and important things to be done in space and um very very excited