On average at Higgsfield, a person on the team spends over $10,000 a month on various models. So internal usage of models a month is over $4 million. Higgsfield. This is the story that no one has told in startups yet. The company has just hit a billion in revenue. It is the fastest growing company in consumer land to hit this milestone. It even surpassed Cursor. Alex, the founder, is an incredible genius. This is the story that you don't know that you need to know.
My parents told me that I must get to the United States because this is the place where technology matters. By the age of 19, I was able to get to top three in the world in competitive programming. I just caught a guy who spent over 30k in a week on Astra model. Many people spend over 10,000 in a week. Ready to go.
Alex, I am so excited for this, dude. We were talking downstairs and I said, I don't think the Higgsfield journey has been told before and it's an amazing journey. So, thank you so much for joining me today.
That's a very special opportunity for us. Thank you for having me. Obviously, your story is inspiring as well, like how social media has become like an elevator for you, an opportunity to create fun and so on.
Dude, it's very kind of you to say. I do just want to go back though because you're not the Stanford, Silicon Valley, born and bred engineer. You were a competitive programmer in Kazakhstan. Can you just take me back? How did you first find and fall in love with computers and become a programmer so early?
So, first you need to understand where I come from. So my father is from Uzbekistan. Uzbekistan is a country in Central Asia where like if a family of five people makes $1,000 a month, it's considered to be wealthy. So it's like not very high standards of living unfortunately. But both my parents are professors of mechanical engineering. Since I remember myself, since I was eight, my parents told me that I must get to the United States because this is the place where technology matters.
So my mother had to work three jobs because basically my education was to compete in programming competitions all the time and to go to various educational camps where I could learn from the best, like certain data structures, algorithms and so on.
Can I ask you a question? Did you feel pressure as a child competing, being pushed into these environments when you were so young?
Absolutely. But I'm very grateful to my parents that they showed me the path really from that early on. Definitely when you come from this part of the world, think about post-Soviet countries, India, China, like getting to the top of the rankings in any competition, in any international competition is the only way to really break out. So by the age of 19 I was able to get to top three in the world in competitive programming. But then instead of pursuing like an academic career, I decided to do startups.
I'm sure your parents were thrilled. Can you take me to that decision? Like this is like the penultimate moment. You've worked 19 years for your parents, have told you this is like the mother lode, this is the thing, and you're like, I'm going to go and do this really risky thing called a startup at this point. Like what happens then?
So let me take you back to 2014. I was very fortunate to work on pre-transformer architecture neural nets and I was primarily just doing optimization, make it run faster, parallel across multiple machines and so on. And we actually built a state-of-the-art system for language translation from English to Russian and Russian to English. Apparently talent wars were a real thing even back then. A lot of my teammates were hired by DeepMind and Meta. But my passion was actually different. I was very, very surprised to learn when I came to the US for the first time how quickly Uber actually spread out. And I was thinking if like this app can take over the world so quickly and transform the whole industry, maybe what's going to happen is that mobile phones are going to become the most used devices in the world. Maybe there is going to be a version of the future where everyone is going to be spending most of their time in their life watching AI-generated videos on the phones because I mean who else is going to produce videos for the phones? Maybe it's going to happen with AI.
Okay. And so that was the company that you built before that you sold to Snap.
So yeah, so the company was called AFactory. I was fortunate to meet Mahi in 2018. He's co-founder of Higgsfield and he is a veteran of Silicon Valley, went through ups and downs and sold it to Snap for $166 million. And then I was leading GenAI there.
Pause, no offense dude, you come from a family of incredibly ambitious parents who push you to do well and you just skipped the moment where you sell for $166 million. It's a lot of money. How did that feel when you did it?
We both remember these times where the capital for AI companies was not really that much available and AI multiples were not like 200 to revenue as they are today but closer to zero because AI was not a topic. So there was like severe dilution which we experienced.
So just to calibrate. So can you... okay, what was around?
No, look, I mean back then rounds like rounds of like $12 million, having like $12 million in investments was considered to be really good. But it was still an opportunity for me to finally go to the United States. So after the acquisition I permanently moved first to LA and then to Silicon Valley and my dream simply came true.
Was it what you thought it would be?
That's a good question. So as San Francisco is definitely a place where no one judges by race, nationality and so on and that's truly phenomenal. There is definitely a meritocracy in a sense that it's possible to meet anyone but at the same time what I see across Silicon Valley investors is it's extremely consensus-driven. So I mean I think that last part I expected to be different but then I read the book about the law of capital and I realized this is just how the world works.
So then tell me, we have sold to Snap, we're now in the US, this is the moment you wanted. How does Higgsfield come to be?
Back then like Snapchat 2020 was really growing so quickly and the face filters which my team has built was driving most of daily new users. What's important is that these face filters we were able to manage to run on mobile devices. So it was virtually for free for Snapchat. It's not like current LLM token costs. And it scaled to hundreds of millions of people throughout the world. And it was truly phenomenal to me to build a product which is still probably the most used consumer media AI product. But then what I realized is that there are a lot of unmet needs on advertising sites. Average company cannot figure out how to be relevant on social media. So this is a major gap like social media is the main media in the world. A lot of companies are actually able to build direct response advertising so that they can actually sell more. But at the same time, most of the companies in the world cannot simply do that. And basically no one simply can keep up with the pace of production for social media as trends change pretty much every day.
Mhm. And so you were like, hang on a minute, these big brands aren't able to have media houses and so we need to create a tool that lets them. That was the sell.
Yeah, absolutely. So where it all really started is that there was a tool to upload a set of images and transform them into a slideshow with music.
It's kind of better than nothing but still pretty bad, right? So another solution was to take long-form video and cut them to short vertically oriented videos. This was better but still really not perfect. And it felt to me that especially 2023 it was absolutely clear that scaling laws finally work. It's not just a concept from science that scaling laws work. Video just takes a couple, I mean maybe two, three years longer than LLMs and coding. But it was clear that actually finally scaling laws should work in video as well and I decided just to take a bet.
But I just want to go back. I get that in terms of what we see, which is, hey, we want to empower these brands and companies to create amazing media for social media, but it wasn't a hit from day one. And I spoke to Amy at Menlo who mentioned like a couple of pivots before and the meandering that we had. So what happened when we launched? Did we have immediate product market fit?
No, actually we spent more than a year in search of a product which could work. We burned more than 10 million out of 16 million raised in seed fundraising. So we felt we have just one attempt left. And frankly I feel I'm responsible because I was focusing on the wrong things. I think I just lost touch with reality back then. I was so much optimizing for what's hype today, what's the right narrative, how we can hijack the attention, all these things really like everything instead of building a good product. So when we had less than 6 million left, I guess it was slightly less than five actually, I realized that the only thing which we can be focused on is to lean into the product, PLG, and just finally set belief that the best product is going to win. And then we just started to talk to customers. We spoke to eight creative directors about their experience with AI and what's simply missing. Everyone told us that camera control does not exist in AI and camera control is so important to tell a story. So this is a very important bottleneck to solve. So we released our product March 31st last year and since then we are really riding this crazy wave.
Was it immediate product market fit then?
Like yeah it was immediate. Is product market fit like love? When you know, you know.
Yes, it's definitely when you know, you know. Like for example, we don't do any paid and like we have on the team people who scaled businesses to over like billion and two billion in revenue like other businesses with paid advertising. Like at Higgsfield, we decided to really make a bet that we don't do paid.
We don't do paid. Is influencers not paid?
That's a good point. So with influencers, there is typically, there are different types of influencers, but typically there is some fee for just video production and then like some cost per click like attribution which works really well on YouTube.
You guys got into some controversy for like, I can't remember what it was. You were like paying people to promote for you or doing something rogue with influencers. Was that completely unfair? Was it kind of my bad we did do that? How do you respond to that?
The main takeaway from like our experience is that it's very important to own distribution. Distribution now more important than ever. And like we basically did outsource, we had just a team of like two people on creator and customer success sides and we just did outsource to the agency and that was not a good experience. But we are still trying to find interesting opportunities to tell about new media formats. Some of them are rather controversial. So for example, recently we partnered with Neon, one of the largest streamers in the world, and launched like his own sort of AI-generated stream. Like no one else did this before because this is like a real creator making a replica of themselves. A lot of people start to question, start to question like is it really authentic content or not? But at the same time, those creators are under immense pressure. We all know about the story from MrBeast about like really how much like there is just pressure to constantly perform. And we also know through conversations with many talent agencies a lot of top stars actually want to be able to do more if they could create a digital replica. But so what's happening today very frequently is that those A-tier celebrities they simply come up for a recording on like let's say green screen and then there is just a lot of post-production which goes on top of it and it feels to me that we are naturally going to come to the point of time where AI digital replicas are going to become just one of the ways how creators can monetize.
Totally get that. I do just want to go back to part of the story. Where are you at revenue-wise today?
So today is actually an exciting day, like when we record, just a Bloomberg article went out so that we crossed 1 billion in annualized revenue. If I had a gong here I'd be like hitting the gong. A billion in revenue.
Actually it took us 18 months from 1 million to 1 billion. For Cursor it took 24 months. So we are probably like the third after OpenAI and Anthropic.
18 months from a million to a billion.
How do you calculate revenue? Like it's a controversial topic. How do you help calculate revenue?
Absolutely. By the way, your co-host Jason also asked this question in May. Luckily, the answer didn't change. So we are at least consistent. But let me be transparent on that. What we do is we look at revenue over the last four weeks and multiply it by 13. From what I know, OpenAI, all of them use the same methodology. What's very important is that we take revenue, not sales. So if that's like an annual subscription or annual enterprise contract we prorate this across 12 months and only take like a piece which corresponds to one month, to 28 days to be precise. That's the first piece. And second, it's only live revenue. It's only live revenue. We are not taking like three-year enterprise deals and baking into like a 1 billion figure. No, we don't do that.
If you were to break that billion up today into annual contracts, monthly subscriptions and then token spend, what would that be?
So video is still relatively early in my opinion. It is still probably two years behind coding in terms of adoption. So on-demand usage for leading coding companies could be over 50%. And I would be honest, for video it's substantially less than that. At the same time what's very interesting for us to observe in the business is that there is significant revenue expansion. I always love to study stories of the largest customers on the platform. So one customer started six months ago spending just a subscription of $99 a month. $99 a month. And now we just signed a deal over 6 million.
6 million a year. Right. So yeah. Like this level of acceleration is something which is really mind-blowing to me.
Dude, what are they getting for 6 million a year? That's like a Hollywood content team almost.
So there are multiple trends and all of them frankly coming from Asia. So first we're seeing a lot of direct-to-consumer e-commerce companies rebuilding their whole go-to-market to be AI-native where they just make hundreds of ads if not thousands a week where they can A/B test what performs well. But we all know about like short-form dramas, right? Like most short-form dramas today is an industry over 10 billion owned primarily by Chinese companies having huge impact both in China, United States, in Europe, everywhere in the world and most of new shows there are made with AI end to end. So look, I think like this adoption obviously is coming like bottom up, but that's very difficult to refute this new reality.
What percent of revenue is consumer versus enterprise?
So that's a great question. So B2B business revenue is slightly over 50%.
Thank you. On the consumer side, it's also very important to break it down. So on the consumer side, out of these 50, around like 10% is pure consumer use cases, pure consumer, and that's roughly people who use it on mobile. So share of our revenue from mobile is less than 10%. That's why we are very different from many other companies. But there are lots of aspiring creators, like basically those people who are freelancers doing social media marketing projects and so on, who try to learn video AI so that they can make more money. It's true that their behavior is a little churny. Within a year most of them actually come back to try again and we do believe that over time most of them are going to figure stuff out and they're just going to become this new AI-native workforce. So it's still important for us to educate them and that's why we invest so much in like Higgsfield Academy, YouTube channel and so on. But we also are fully cognizant that we will never be able to win in a market of subscriptions of $20 a month.
So why? Because like I think like today Google and OpenAI they pursue like ads so much but fundamentally I think they are going to completely demolish all the consumer subscription markets which is $20 a month subscriptions.
Oh, so you're saying that because they provide a horizontal product that's very good, you're just going to not pay for a lot of the verticalized products that you used to pay $20, $30 a month for.
Yeah, I do believe that. That's essentially what's going to happen over time. I know this is a very contrarian bet, but at least we can see some of that.
I think it cannibalizes Canva's growth if you're honest. A lot of the low-hanging fruit on the consumer design side that Canva used to serve can now be done in OpenAI in particular. Is that what you're talking about?
Yeah. And I do believe this is just the most apparent example, but there are a couple more which is already happening. And I do believe that that's why for at Higgsfield what really matters for us is how we, even if we get someone on like a $20 a month subscription, like how can we show them value, how can we make them upgrade to spend over $1,000 a year with us.
I can't believe that's 6 million a year from 99 bucks. That's the best ever slide on a fundraising deck.
And all of our customers are going to do the same. Exactly.
Can I ask you, you mentioned there kind of churn rates. When you look at 30-day retention rates for consumers and 90-day retention rates, what are yours and what is good?
So there is quite a massive drop within the first month just simply because people don't fully realize the value and that's a core priority for us to actually get better in that. So showcasing the value.
Is it like half or like...
No, it's maybe like a 30% drop.
But then it's really flat after that. We look obviously at like logo retention.
I wouldn't say it's great but because like we all remember like B2B SaaS era like retention was expected to be, logo retention month one was expected to be over 80%. So clearly we have lots of work to do on user education to get there but some things are truly phenomenal like when I look at the business segments and NRR at month 12 obviously like you're going to argue it's like an 18-month-old company like what are you talking about but still when I look at the numbers which I have today, NRR at month 12 is over 300%. It just never happens in B2B SaaS right so that's why I'm saying that while there is substantial churn in month zero and we have to do a better job with user education to address that, expansion is unprecedented.
Can we actually just unpack the two different go-to-markets because you got consumer and you got enterprise and I spoke to quite a few of your competitors in all honesty before this show and I said hey, we've got Alex coming on, what should we ask him? Everyone said the same thing which was an admission of their respect for this particular kind of GTM. They said you've executed the most impressive influencer campaign in tech and what I wanted to understand was when you look at the consumer growth, what worked, what didn't work and how do you reflect on that?
First and foremost like the goal is to make sure that the best commercial video content is generated on Higgsfield and we show all the workflows of how to make such professional-looking videos and we have an in-house team of over 150 creative professionals. 150. It's almost half of the whole workforce frankly. And those people, they make product launch videos, they make tutorials like for example we made the first generated movie which is also like obviously a very sensitive topic but what's important we open-sourced all of it and what we learned is that for 90 minutes of like let's say TV-quality content it was over 100 hours of AI-generated content. So creative decisioning like picking the right piece is still very important. So that's really what we are focused on and that's what's driving most of the revenue.
So you're saying the growth in consumer subscription is through own content and distribution.
Yes. We don't do any paid.
Early on you made an interesting architectural decision to have your own models and then you since walked that back. Can you talk me through why did you choose own models and why the walk back?
Oh, yeah, obviously this was my mistake. I'm going to do my best to be transparent. What I need to admit, we really tried. At some point of time I really was thinking that chasing benchmarks is valuable but I don't believe this is just sort of corporate scops frankly. And I was part of a large organization so I know what happens. What happens is that everyone just thinks like we need to show some progress. So we need to have some benchmark but then when I talk to the top researchers from these labs especially larger companies what happens is that they start to put test data into the training. They start to kind of leverage test data to use LLM as a judge for training of the models. Use all the various tricks to basically gain benchmarks, get quarterly bonuses and so on because like who cares, right? So if I make my couple million dollars a year inside one of these labs, I can move to another lab easily. So that's unfortunately what's happening in larger organizations.
What do you mean? You're saying that they are incentivized by benchmarks and so because of that they are doing artificial things to improve their scoring in benchmarks which actually don't increase output efficiently.
Yeah. Look, I think let's just look at the outcomes which we have today. Out of all the incumbents in the United States, when I look at OpenRouter data, the only company which is relevant is Google. Out of all the incumbents when I look in China where probably obsession with benchmarks probably is less, we have Tencent, Xiaomi, Alibaba, like three incumbents being completely relevant and obviously ByteDance obviously trying to catch up as well.
What's your takeaway from that?
I just do believe that there is just obviously in the tech bubble there is a strong obsession over the benchmarks which do not necessarily represent the reality. But I can talk specifically for video. A lot of benchmarks today for video is really text-to-video which does not represent actual workflows at all. The way to think about video models today, it's just a modern rendering engine. Think about this as like Unreal Engine or Unity but just different types of inputs. And it's virtually impossible to really define a visual output and direct the execution just through text. If you just go to our open-source projects like this movie which I mentioned, average prompt length is over 3,000 words. That's the first thing and like look all these benchmarks which we are talking about they are not as comprehensive in terms of the details of prompts and people who are labeling they obviously cannot read like 3,000-word long prompts but also on average there are at least 10 image references for every scene. The reason why it's important because it's important to define how the characters look like, how the background looks like, like how actually characters are located to each other in the scene and so on. And so that's why like prompting and like just the workflow is so complex. Benchmarks just don't represent that.
So going back to the model selection, why did we decide we're going to do our own and then why walk it back?
It's true that like with VFX and camera control, we got very very quickly from like maybe 1 million to 20 million in ARR within maybe the first three months. Then we released our own image model which is really good at aesthetic photo shoots and product consistency. This is what allowed us to scale then from 20 to 100 million.
So help me understand, Alex, why did you decide that you were going to do your own models and why did you abandon them?
We still do them whenever we see like a specific use case like these photo shoots. But as soon as this is what our customers want. So it's all driven based on the customer feedback, not just by ambition to conquer the world and build the best model in the world.
Do you think every company will have their own models like we're seeing Harvey, we're seeing Cognition, we're seeing Mccor, Ramp build their own models and we'll see every company have their own models with their own data or we actually all use a series of providers?
So first of all, whenever just to be honest, whenever someone says we build our own models very likely what they mean is something what we see what's happened with Cursor. We do remember right, a lot of companies they actually take open-weights model and just post-train on own data. And post-training can happen in two ways. Most importantly is whenever you have customer data around like decisions they make like sequence of decisions and you can teach the model to actually learn how to compress these 10 steps into one step. Like this type of reinforcement learning is the most valuable. So and I think like increasingly more and more companies will have to do that frankly just we see this in the market as well. So most of the companies in the world today, most of the businesses they don't necessarily need Astra specifically, they don't necessarily need the newest Fable model and that's why like OpenRouter reports that share of open-source models went from below 30 to over 60 within this year.
What do you think share of open models will be in two years time?
Look, I do believe that just because capitalism works, I mean OpenAI and Anthropic still are going to have more than 50% of the market in terms of the dollars, right? And especially because for coding still remains to be a very very prolific use case where coders are always jumping to the recent model. But for our markets we're seeing completely different dynamics. What's actually happening in social media marketing as companies start to print hundreds of ad creatives a week, they want to have maybe cheaper, more stable models because like PhD-level intelligence is not necessarily needed to make viral social media video. So and that's where we actually have seen that we get like 80% plus margin whenever we run open-source models like post-trained open-source models. But it can be way more cost-efficient for our end customer compared to the proprietary models.
What's the comparison on margins between open versus closed for you?
The margin on own models and open-weights models is over 80%. And then it almost doesn't matter. And for closed-source models, it's probably between 20 and 30%. And then what becomes important is can we actually steer the traffic. What makes me excited about Higgsfield is that it grows so quickly and actually for us as companies start to actually create those agentic workflows to make more ads, we choose which model we can use. So like we choose what model to use in over 40% of cases.
In a way model routing becomes a core feature of the business. No?
Yeah. We call it tokenomics essentially right as like there is a certain amount of work customers want to do, how can we optimize number of tokens which requires and how we can pick the most efficient tokens for them.
There are actually two incumbents in the United States who figured out models. It's not just Google, it's also Nvidia. Why do you think that is?
What I'm constantly seeing is that there is the versions of models. So there are these state-of-the-art models which have to be really good in computer use like Astra or in coding. But they can be prohibitively expensive and we were chatting about that like on average at Higgsfield a person on the team spends over $10,000 a month on various models and remember like we are split across United States and Asia across.
So how much do you spend on models per month?
So internal usage of models a month is over four million.
Wow. How many people do you have?
We have close to 400 people and just want to make sure that the math adds up. Yes, it's definitely over $10,000 per person.
How has that changed over time?
That's the best question of the whole show, by the way. That's the best question. What actually started to happen is the creative team started to do VIP coding. Like this month I just caught a guy who spent over 30k in a week on Astra model because he was frankly frustrated that some like asset organization workflow and as you said like basically auto-editing is still not very good in production and he said, 'Oh, I'm just going to do this myself.' And just went like five nights, five nights straight on Astra and it works. We learned a lot. I wouldn't say it was production ready, but we learned a lot.
Yeah. Yeah. Many people spend over 10,000 in a week.
Yeah. My finance team will probably say, I don't know if you ask any of them, but they will probably say that I'm like being too stubborn, too relentless to control the spend because sometimes I feel it really goes out of control like 30k in a week is quite a lot. But we learned this. So this was actually a net positive experience.
Okay. So the internal spend 4 million, about 10,000 per head. What will that be in 12 months time do you reckon?
So that's very interesting. So across the top engineers and across top creatives I think it's going to keep growing and I do believe we are going to get to spend close to 50k and 100k a month for those who can call 10x engineers, 10x creatives. Unfortunately I also expect that these people will ask for comparable salary raise as well so I think that's just going to correlate at some points. But also for a lot of other jobs. Let's say we take legal, finance and so on. I think it really stabilizes around like $500,000 a month very very quickly.
With those 10x engineers, the idea is they have thousands of agents running below them doing a lot of the difficult execution work that took time. Do we just have dramatically smaller teams with those 10x engineers, 10x designers, 10x finance leaders?
I can definitely say that I had sort of a feeling that legal, customer support is going to be mostly replaced and that's obviously one of the main mistakes operationally which we have done in the company that we didn't ramp these teams quickly. What we are seeing today is that like let's say our legal team is like over 10 people, our customer success team is over 40 people, all of them use AI heavily. At these professions where I say quite close today I definitely can say that there is, I don't see any elimination. It's true that probably over 60% of customer support requests especially the first line of defense can be handled with AI but when it especially comes to B2B, AI just doesn't work.
Revolut has now over 92% resolution rate on customer support for consumers.
