USV's Nick Grossman: the 'Rebel Alliance' thesis — why generational companies will be built around AI labs, not just by them

subscriptions. They have great businesses. Some of them are enterprise. Lots of stickiness in the revenue, but uh would be interesting to see them at the helm of a true social network, potentially even a growing

one. Anyway,

our next guest is in the waiting room. We have Nick Gman from USV. He's a general partner and he's here to break down the Rebel Alliance thesis.

What's up, guys?

How you doing?

I'm doing great. How are you?

So, you're fighting against the Death Star. What? First, introduce yourself. Tell us a little bit about yourself, your general investing thesis, and then break down the Rebel Alliance thesis.

Yeah, you got it. It's great to be here. Um,

I'm Nick. I'm a partner at uh Union Square Ventures. I've been at USV for about 15 years. Um and we've over that time been investing

overnight success bingo. uh been investing up and down the stack and across you know moments in time when uh different sort of layers of technology have have emerged uh and you know open the door to lots of new stuff and it's I think if there's one thing that's consistent about the way we look at the world it's like every sort of wave of technology is either kind of expansive or consolidating and uh those moments where it starts to feel expansive are just incredible right so back in our history. The first one was like web two, you know, you had built the the core internet and now you could build anything on top of it.

Um

the other one, you know, for us it was, you know, Twitter was probably like the big investment from that era. If you look at like the crypto era,

wasn't USV also in like Tumblr and Etsy and like a bunch of other companies from that era. It was a great portfolio.

Yeah, that was back. So the firm was founded in 2004 by Fred Wilson and Brad Bham and the thesis back then was um there's been this epic infrastructure build in the you know fiber and and switches and routers a first pass at building you know mainstream consumer apps but like you know a big blow up with the dot you know pop and the thesis in 2003 which was pretty contrarian was there's going to be a big opportunity at the applications layer and there's going to be a moment you know it's time to build basically

and that was pretty

you know it wasn't the most popular view back then. Uh I think it created an opening to make some really great investments in a lot of early you know um application layer things. So we did Twitter, Tumblr, Etsy, SoundCloud, you know indeed you know a whole bunch from that era and it was all about you know the the tools are on the ground so what can we build with them? And I think those moments when the tools are on the ground and it's time for people to build with them and anybody can build anything in any direction. I think those are the moments that have always felt the most exciting to us and that we've been looking for you know for 20 some odd years.

Pros and cons of thesis driven investing. Most I would say investors now are humble enough to be like I just try to find the best possible founder and give them money.

Most thesis I hear is is this one buy low sell high.

Yeah, that's also that's also a good one. Um, but I just feel like, you know, you you guys have done so well like trying to really crystallize your vision of the future

and the opportunity in the present and then investing and like actually being like we know there's a company that should be built here and then taking the time to figure out what the right team is to go and attack that opportunity. But the I'm sure there's like the the other side of it was like sometimes you just like get a little high on your own supply and you know back back an opportunity that that doesn't exist and you kind of forced it because you wanted it to exist. But so so every good strategy has its downsides. But I'm curious.

Yeah, I would say that the two potential downsides of sort of thesis driven are you're wrong. Like that's the worst one. You're like you think something's going to happen and it doesn't. Um that's happened to us, don't get me wrong. Uh, and then the second flavor is like right idea, wrong team or, you know, right idea too early, you know, or whatever. And like we've had our share of those two. And so I think, you know, we're a small fund. Uh we don't need to play a coverage game. Uh we don't need to play momentum. That's just not how we've operated. And so we're sort of set up to be able to pick our moments and take shots and um you know have a point of view about what we think is going to happen and and hopefully if we're right we're contrarian and right you know and we we've been that way a bunch of times and hopefully if we're wrong you know we limit we limit the damage

y on the losses.

Yeah. So take us through the rebel alliance. Who's in the alliance? Who's out of the alliance? Break it all down for us.

Yeah. So the Rebel Alliance, we just put this up on the USB blog last week. Uh it's an idea that's been cooking uh for some time for us

and it's basically the idea that if you think about AI and now agentic AI the opportunity is so much bigger than any one company and that there is a a massive ecosystem exploding uh around these incredible magic core models. Um, and there are layers and components and modules and we're just beginning to start to see all of those get built out at an everinccreasing rate, especially now that model of capability is so good.

Yeah.

And and that means that we believe there are going to be generational companies built up and down the stack. Um, not just, you know, and really the, you know, I don't think that OpenAI or Anthropic are the Devastar. I do think they are these really powerful companies with that that have sort of a vertical vertically integrated approach and I think the meme in the market is like maybe you know Claude becomes the world's only employee and and all value gets sucked into those companies. Our thesis, the Rebel Alliance thesis is that the opportunity is so big and so massive and so expansive and there are so many forces pushing for this expansive motion that there are going to be companies, you know, in in all these components of the stack building around including the models themselves, but also so many pieces around them. We're already starting to see it. We're making lots of investments around them. Um and to us it feels like you know the beginning of the whole economy but maybe the whole internet getting rewritten on the agentic stack and with an agentic approach. Um and so really the idea is that this what's happening is is is bigger is so freaking big. Um and that the opportunity it's like a great time to be building. It's a great time to be building for builders. Um and there's a lot of big companies are going to get built. That's really the core thesis. No, I I I love the thesis and uh we see it every day on the show where someone's coming on, not just with a big fundraising announcement, but a ton of revenue traction and a ton of customer love. Uh I'm interested in uh the the down the stack feels maybe less contrarian to me just because uh I'll like you see announcements from the big labs where they're using Salesforce or they're storing all their data in Clickhouse and uh and and or or even going deeper into the semis and the memory and the energy. All of that feels uh really obvious, but uh the the bigger meme from a year ago was like, are you going to get steamrololled by the next model? And there were some companies that were just sort of thin rappers with some prompt engineering, but then you have companies like 11 Labs and Sunno and these specific midjourneys doing really well. There's so many companies that have gone and carved out different niches and done really well. Yeah, Sunno is one of my favorite examples because it's it's, you know, clearly a massive opportunity already a multi-billion dollar business. Millions of people love the product and yet

it's not a good fit for the labs to prioritize because the size of the prize is just like, you know, I don't, you know, Mikey's extremely ambitious, but like, you know, maybe it's a$20 billion business over time, right? I mean the Gemini app say the Gemini app has a make music button inside of it and Sunno has still carved out like this incredible niche anyway

and I think there's two there's two ideas like just looking at example like sununo one is uh user experience right and I think as we cover this we rewrite the world of AI whether it's in media or industrials or robotics or health or whatever these experiences are going to need to be specific and tailored and you know they're they're maybe you could call them niches but they're really big and I think that all doesn't collapse into a single app and then the other thing or or an experience and the other thing that we're seeing from everybody we talk to building it at the application layer whatever you however you describe the application layer is uh model competition there's so many good models today and as we move um you know as more things move into production we start being aware of the trade-offs between model capability and model cost. Everybody who's building is thinking about how to build an abstraction layer to make those those best trade-offs to deliver the best experience and to have margins. If you're building a product like Sunno, you know, you need to have margins. If you're an enterprise adopting this stuff, you're thinking about your token cost. And I think both of the those sort of forces are uh you know, are good for the Rebel Alliance. And at the at the end of the day, it's a core theme is competition among the models, but it's not the only one. All right, let's live workshop an idea that I have that I want someone I want someone to build.

Uh I over the weekend I was uh messing around with uh in chat with the image model on some furniture ideas that I had. I managed to make some stuff that I think is cool. I would happily pay like for this like chair that I made. I'd happily pay a couple thousand to get a very great version of this chair. And I was thinking like it'd be great if I could upload it and then have like some combination of like an Etsy or an Upwork style thing where furniture fabricators could say, "I will make it for this price. Here's the work that I've done."

And basically like a marketplace to kind of manage that.

Uh theoretically, if I made it and it's great, then other people could just buy that product and the order would go to the fabricator and maybe I could get a little bit of a cut. And I thought it was an interesting idea because it sort of like rides the tailwinds of like anybody can imagine and prompt a product into digital form and then like you need this network of people globally that can actually go make the products. I'm not sure if it's just like too small of a TAM, you know, like how many people really want to make furniture. Um but uh what's your immediate reaction?

Well, there was a company uh maybe a decade ago that did this for physical products. Um, I'm blanking on the name right now, but you know, this idea of invent your thing and make it, you know, from the 3D printing area era to the, you know, furniture or other things, I think is even more possible now because part of what we have is the ability to design, you know, amazing things, you know, overnight. And I think where kind of where you're going with this is this isn't really part of the Rebel Alliance idea, but it's another theme that we've been really focused on is uh services that plug into agentic experiences or I think with unique uh data or networks or sort of systems attached to them, right? So in this case like product fabrication, distribution, so on so forth. It's like you know these are durable types of networks that need to exist with a very different architecture that can plug into the agentic future. Um, and so I think things like this are going to be successful. And then there's a question of like at the end of the day, who owns the interface? You know, what's the consumer experience for that? You know, does it all roll into a chat dbt or a cloud or, you know, is it something else? Kind of a separate question, but I think like there another theme is like durable networks uh as APIs into aic experiences or MCPS or whatever. Um, I I think we're going to see tons and tons of those.

Should I pitch my idea? the agentic AI for doctor billing.

No,

no, definitely not. Definitely not.

For another time. For another time.

For another time. Uh, how are you?

Yeah. How are you? Um, it's interesting the like I I I still have some weariness around this like Rebel Alliance style plays because of the first generation of companies that were leveraging uh GPT models. They grew like crazy

and then it just was a matter of like building an interface and an interface is is straightforward. How are you like you clearly have have are overcoming that fear and are willing uh are investing a lot against it. Um part of it is like I just feel like this dynamic now where hiring an agent is like quite different than hire you know just sign you know using software and like you just assume that agents are going to get better and um one of the challenges is it's like an agent that is generally intelligent just like a human you can kind of plug into an organization and it can start just doing things and you don't really care that much how it how it does things and so I just feel like the process of like hiring agents for companies um is going to be like quite a bit different than like the last generation of software where you had to build out this massive sales force and you know you're going to conferences and it's like it's it's it's just a very different motion and I think that like the growth of of claude code specifically is like proof that like software has changed to some degree and you can get adoption of something at a speed that is entirely unprecedented. Um, so I'm curious how you're thinking about like how how just like the procurement and sales is going to change because like again we have we have companies on every single day that have that have grown revenue incredibly rapidly. They clearly are solving customer problems. Um, you know, we had a healthcare agent company on today started in 2023. It's a billion dollar company now. Um but uh but yeah the question is like are are you still betting against when you invest in these companies are you just betting against general purpose agent progress? Yeah, that's a really good question. And I think um one important idea here is we're not just talking about will there be other agents that compete with Claude or Chad GBT like will people build sort of like personified agents and how will those get to market. Um I think that's one version of what agents look like. Another version of what agents look like is like cloud infrastructure where every, you know, company that's building any system is going to have agents and the orchestration of hundreds or thousands of agents built into their underlying, you know, platform, whether that's an app or whether that's an enterprise or or anything else. And so if you I think part of the Rebel Alliance thesis for us is uh agents are moving from sort of people you like personified people you hire to cloud infrastructure

and uh and it and when we look at how a agents are getting deployed and orchestrated today it's not just I'm hiring an agent and it's showing up in my Slack. It's like I'm programmatically spawning you know 50 agents you know to go sort of be part of whatever system I'm building. And so if you think of agents as infrastructure as opposed to agents as employees, I think that's where a lot of these forces come into play where every every company, every enterprise who's building in an agentic style is going to need model routing, uh model choice, orchestration, um you know, uh memory, um you know, some sort of harness, and there's there's different options at each of these as you're building up a systems. So, so that's like really a big driver of this is the move from agents as people to agentic systems that have all of these layers in place and have more of like um corporate's the wrong word, but like a more of a production vibe to them where they're going to get built by professionals who want to have control and who are going to want to lay them out in increasingly sophisticated ways. And so, so I feel deeply like like very very strong that that's happening. I see it everywhere. And I think a big like a lot of the stuff that we're looking at right now in the agentic stack is like kind of low-level like networking and identity and payments and all these like pieces you need in order to compose and orchestrate agents. We're moving from like agent as employee to agent as infrastructure and and I think that also is like a big driver of how why we believe so much in the Rebel Alliance thesis.

Last question. Can you take me through the financials, what they might look like at a Rebel Alliance company versus a non-rebel alliance company? When I think about a lab, they might spend a billion dollars on a training run, amvertise it over months, and there might be all these GPU commits and and they're spending money like an industrial company sometimes. Uh, if you're building Rebel Alliance company, does it look different? Is it all talent? Are you subsidizing tokens at some point? Is it cost of customer acquisition? What are you pattern matching to in terms of like the shape of the business? Are we doing 12 to 18month runway here? Like how are you thinking about uh deploying capital and watching that capital flow through the business at the early midstage of a company?

Yeah, it's good that you brought up like capital intensity, right? So I think because I think an idea here is there's this category of hyper capital intensive AI companies which is the big labs. Yeah.

Um, you know, we have a portfolio company called Plurales Research that does modeling but in a decentralized network using, you know, compute that's scattered around the internet. So, you can still do training in a capitalite way. I think that's possible. So, I don't want to make it seem like the all the models must be capital intensive in the form of a single company, but I think

generally speaking, you're looking for things that are more capital light. um a lot of the things in the middle like harnesses and memory and you know identity and payments I think a lot of that looks like kind of regular tech companies and then anything that sits at the sort of user level that's passing through token spend you know there's a category of margin profile there that's that's new in the AI era right you're either taking a loss on on token spend passed through to the big labs or you're like ekking out margin by swapping out models and so I do think there's actually it's almost like this barbell of capital intensity where at the at the model layer there's a lot of capital required and at the application layer there can be if you're subsidizing a lot of token spend um certainly on a free tier or or something like that but there's a lot in the middle that I think has more of a traditional capital

it's just all wildly different than web 2 where uh a dollar of revenue was usually a dollar of gross profit basically and now you you might see 60% gross margins you might see 20 you might see 90 it Just depends on how the entrepreneur has positioned the business and how what the

and like 20% gross margins for some application companies is like great.

Yeah. Yeah. It might be negative 50% for them if they're growing really fast.