Fal co-founder on generative media demand surge: startup customers generating billions in combined revenue

Uh let's bring in our next guest. We have the co-founder of Fall. Do we have to stream? How you doing? What's up? Good to see you in person. Welcome to the show. Welcome to the stream. What's up, Jody? What's happening? Introduce yourself for the stream really quickly. Um, welcome to YCWA. Great shirt, by the way.

Look at this. All right. We just talked to We just talked to a founder that that made their first uh money flipping flipping Supreme Box logo t-shirt. There you go. There you go. Yeah. I'm decked out in false fall swag. Yeah. But the GPU poor. Are you guys GPU poor anymore? Well, I have a GPU rich hats, too. Okay.

I got to give you guys some. Um, so I'm I'm one of the co-founders of F. Yeah. Yeah. Fantastic. Also, one of our latest sponsors makes this possible. Uh, so thank you. corporate media. Uh give us your reaction to YC demo day. How what's the mood?

I think you just rolled in here from uh but give us the vibe from literally like I just rolled in from our office. We're like 5 minute drive away. Um I mean it's amazing energy. I love it. Yeah. Uh I mean being in the Silicon Valley for I don't know 15 years, close to 15, I actually never been to a demo day.

There you go. Brought you. So this is my first time. Um and yeah, love the energy. That's great. So give us um Yeah. So, so I guess what's been h how much how much have you guys from a from a customer acquisition standpoint? It feels like there's like incredible pull from the market.

You guys have crazy logos from you know uh decacorns, public companies all the way down. I imagine a bunch of bunch of just companies are getting off the ground and signing up for you guys. But like what is that customer acquisition kind of motion look like?

Are you guys do you pay much attention to kind of the super early stage or do you just like let people find you? Yeah, we pay a lot of attention to early stage. Uh a lot of our customers are early stage. Um and it is I think it's one of the best times to build like on top of fall and you know other other LLMs as well.

But like if you want to build anything in consumer AI specifically, I think fall is probably like the best way to accelerate you know building something out there. There's incredible demand.

I think I was doing some math on the way here like um I think in our startup segment I think the revenue of like all of our startups com combined and what what I define startups is let's let's call it like 200 people or less. Yeah. Raised less than a billion dollars. Yeah. Sure.

I think the total combined revenue of our customers is in the billions. Wow. No way. So, so it's kind of it's kind of ridiculous like how much demand there is in the in the consumer AI market or or even like proumer, but you can build amazing applications that that you know that leverage the gen media technology.

Yeah, that proumer thing is is fascinating.

I mean we saw this with uh the the images in chat GPT the dolly moment where you know it's not like like it's not an enterprise tool but it's something that went so viral that everyone has to wake up even if you're just a marketing manager in a Fortune 500 company like you're going to see it you're going to think about it and be like okay I actually need a strategy here we recently saw this with Nano Banana again we saw a little spike in the Google trends chart uh what is yeah what have you guys seen on the nano banana era, how are things changing?

Just kind of give me your read. It's incredible. I mean, the demand the the way the way I look at it is like every time we have new capabilities, there is a jump in demand. Yeah. And it it doesn't seem like these things are eating into each other.

It seems like we're really seeing like the the building blocks coming coming together. Yep. So, for example, with Nano Banana and like all the editing models in general, they actually are a key ingredient to really good video models.

Meaning when you want to stretch like stick together a few clips start with a few of those. Yeah. You you you actually want to you can keep like the consistency between using these edit models, right? So like it actually solves the consistency problem in video. Yeah.

So that enabled video to go even further because now you can just generate like you know 10 different starting frames and ending frames and then combine them together. So so these things are really like building blocks and they're kind of compounding on top of each other.

So the way we see it in traffic and revenue is that anytime there's like a new model, new capabilities specifically, we we see a huge jump. It was pretty wild. There was a site that spun up. I don't know if you saw this the day of Nanoban's launch. It was just called nanobanana. ai.

It had full like it was a full looked like a Google product and it was just someone else just vending in Nanobanana under the hood, but I'm sure they got their API access. This happens all the time. Like if you actually search nanobanana on app store like just go to app store there will be literally 20 results.

Yeah this is so crazy still nanobanana.

aii is the number one result for nano banana people have nailed it down AI image editor edit photos I mean the app store was like that for so long where you search chat GPT and it would be like chat with GPT and it has like 5,000 fivestar reviews and you're like I know that's not from OpenAI because I can clock it as like clearly this is not an OpenAI branding.

It's slightly off and and all of these things are, believe it or not, they're getting traction. Yeah. So, it's insane. Yeah. Like, and like like Apple, I don't know, like they're not really doing much to stop this. Maybe they're trying, but like not really. It's a game of whack-a-ole. Yeah. It's a game of whack-a-ole.

And and I think they're benefiting from this. And and I think, you know, obviously like long term, what matters is like how these products differentiate, but I think it just shows you like how insane the demand is. Yeah. What keeps you up at night between the uh inference cost kind of stack?

So uh are you are you worried about GPU poor like are we going to run out of line time at TSMC? Are we going to see a chip shortage versus just large scale data center capacity versus energy?

Like if we play this out a few years like what what what do you want to get right that might be out of your control even but like you'd like to make it make sure that it goes right. Yes. I think we're very capacity constrainted and but at what level that's what I'm interested in.

So, so yeah, it's it's so it even this even this this Nebia steel yesterday, right? Like like shocked everyone, right?

Especially we we were hanging with a a friend who's a multi-stage, you know, has multi runs, you know, help helps run a multi-billion dollar fund and he was just like like a year ago this company like couldn't, you know, this company was not taken super seriously as a player and all of a sudden they have a$20 billion deal with with Microsoft.

Uh yeah, Microsoft.

Um so anyways, like that that just shows like the the fact that the fact that you know Microsoft is going I I I you know if you if you kind of like stack rank where Nebia sits in in the Neocloud space, it certainly wouldn't have been somebody's top choice, but I think that just shows like how capacity constrained um everyone is.

It's incredible. Yeah, I think I mean currently we're going through another crunch right now. Uh specifically for H100s. Uh so like we experienced this interesting and and you can you can kind of feel it coming and and you know like does it feel like that? It it it it feels pretty bad, right?

Like you're trying to you're trying to scale. So like most of our workloads are inference workloads, right? We don't do any training.

So our needs are like we don't just like buy you know 10,000 GPUs and just sit there for another for a whole year right that's not how we operate we're more dynamic we go and like expand our our fleet over time so we and we're multicloud so we talk to like all the you know new clouds and we have really great relationships with all of them we test them we benchmark them if if they work for us you know if they're reliable support is good we use them so and you know when you need when you need new capacity.

You literally like hit them up. You call all of them up in order old fashioned way and and literally I'm not even kidding like you just you just call them up and you know sometimes it's like you start hearing more nos, right? Like do you have do you have 512 GPUs? Do you have thousand GPUs?

And it's just like no no you're like oh crap. So so you you kind of want to like stock up a little bit. That's how we operate. We we like we like buy buy a little bit in bulk, you know, but but really prices hit hit a little bit of a bottom, which like hasn't happened with A100s.

I mean, if you look at it, like they just kept going down with H100s. They kept going down and they kind of hit the bottom. They're like now now kind of kind of picking back up again. How uh what can you tell me about how inference moves around the globe as a function of time?

I heard this story about midjourney potentially doing inference in Asia right where they're sleeping they're not using they're not running Netflix in Malaysia or Singapore and so when I generate an image or something you know it go it's inferenced over there is that something you're doing is that something that you're planning to do is that common is that happen naturally so so our traffic patterns are basically maxed out when US and Europe is up simultaneously so that's that's kind of like like the peak peak time, right?

And then we don't have like as much users like we still have customers in in Asia, but like it's it's much less. So, you know, we we do see a deep dip, okay, across, you know, when when US and EU is asleep, there's a slight dip. Sure. Um, we do we do a few things. Yeah.

Uh, we use like we basically like uh have a pool of GPUs that we like, you know, return and and and and bring back simultaneously. But what's really important for us is uptime. Yep.

So even if it means like we're going to have some idle GPUs, we we'd like to take that cost because if you can't take those GPUs back, like if you let them go, you can't take them back, you know, then then you you have a very bad time with your customers. So for us, like reliability is more important. Got it.

At at this time, you know, and over long term, I think if you might be able to move move the liquidity across the globe over time. Uh what any any key lessons? You were at Oracle for four years back in the day. any key lessons from that era that you've kind of brought in?

Obviously, Oracle's been on its tear um and really leaning in. So, Oracle was my first job out of college. So, you know, that's 2012, 2016. Um back then, they did not believe in cloud. They were like the anti- cloud company. Yeah. You remember that? Yeah.

So, so like Larry Ellison would go on stage literally be like, "Guys, like we don't believe, you know, he has like a famous quote, you know, it's someone's computer or something, you know, the cloud doesn't exist.

" um they've come a really long time long way since then and obviously like they're on a rip right now I think with with AI what we hear from the field is like um there's really good you know Nebus including Nebus is a very good like data center operator and Oracle is a very they had to do that because of running Yandex right yes yeah they actually had to I I get more worried about Neocloud products that are were doing you know like there there's one I think it's like Irene or something like that that's coming online that that you know just last year I think was doing like Bitcoin mining right so like fast pivot maybe a later pivot some of those guys are also good but so after Oracle I was at Coinbase so uh the crypto like some of the companies I including like they they had a lot of experience but like what's what's special about Oracle is that they they've got a lot of experience they know how to operate these machines at at you know really good cost so and and uptime is very important.

So, you know, all of these things, you know, the economics and and like operating the the data center play a huge role into this. I think they, you know, I'm I'm glad they chose going in this direction. You know, I I think there's only a few opportunities, right?

And this is this is one of them and they like they just went ahead and like capitalize on it, which is really great. Lightning round. Yeah, let's do it. We've been asking everyone, who is your favorite entrepreneur, someone you look up to throughout history? O. Um, could be anyone. Could be Larry.

Um, I really like Okay. Uh, could be anyone. Let's see. I like Brian. Brian Armstrong. There we go. It's a pretty good pick. That's a great pick. I love it. Yeah. Yeah. Legend. Uh, second question. How did you make your first dollar on the internet or in business? O. Okay. I had um Do you know these toys?

Karetta Fighter. No. No. Okay. It's like, is that too old? How big is the toy? This it's like it's like two fighters. You just like make them fight with each other. So when I was like six or seven, this is not on the internet. Yeah. Like almost rock and robot. Same thing. Same thing. I I rented those out.

You rented those rent. Flip the capex to opex. You don't need to you don't need to buy these things. You go to him, you rent them. Today he's renting. That's actually kids don't do renting enough. They're always transacting. Am I stretching? Let me draw a full circle analogy here. You rent you buy the servers.

You rent out the tokens. You rent out the He's going to the What's your allowance? Okay. 20 bucks a week. That's what That's the price. 20 bucks a week. There you go. Value based prices. Yeah, exactly. That's fantastic. Amazing. All right. Well, come back on the show, guys. Good to see you.

I got to I got to give you some hats. Please. Give them give them out to your favorites. Fantastic. Okay, we will. Perfect. Some rich hats. GPU. Oh, these are good hats. These are really Okay. Okay. Yeah.