20VC: $1BN ARR in 18 Months; The Untold Story of Higgsfield | Spending $4M Per Month on Models | Why Moats in AI are BS | Scaling a Content Team to 150 People with Alex Mashrabov

Alex, I am so excited for this, dude. We were talking downstairs, and I said, “I don’t think the Higgsfield journey’s been told before.” 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—how social media has become an elevator for you, an opportunity to create a fund, and so on.

It’s very kind of you to say. I do 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 take me back? How did you first find and fall in love with computers and become a programmer so early?

1. The Competitive Programming Escape

First, you need to understand where I come from. My father is from Uzbekistan, a country in Central Asia where, if a family of 5 people makes $1,000 a month, it’s considered wealthy. It’s not a very high standard of living, unfortunately. Both of my parents are professors of mechanical engineering, and as long as I can remember, from the age of 8, my parents told me that I must get to the United States because this is the place where technology matters.

My mother had to work 3 jobs because my education basically consisted of competing in programming competitions all the time and going to various educational camps where I could learn from the best data structures, algorithms, and so on.

Can I ask you a question? Did you feel pressure as a child competing and being pushed into these environments when you were so young?

Absolutely. I’m very grateful to my parents that they showed me the path from such an early age. Definitely, when you come from this part of the world—think about post-Soviet countries, India, and China—getting to the top of the rankings in any international competition is the only way to really break out.

By the age of 19, I was able to get into the top 3 in the world in competitive programming. But then, instead of pursuing an academic career, I decided to do startups.

I’m sure your parents were thrilled. Can you take me to that decision? This is the penultimate moment. You’ve worked 19 years for this. Your parents have told you this is 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.” What happens then?

2. From Neural Networks To Snap

Let me take you back to 2014. I was very fortunate to work on neural networks with pre-Transformer architectures, and I was primarily doing optimization: making them run faster, parallelizing across multiple machines, and so on. 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 surprised to learn, when I came to San Francisco for the first time, how quickly Uber had spread. I was thinking, if this app can take over the world so quickly and transform the whole industry, maybe 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 spend most of their time watching AI-generated videos on their phones.

Who else is going to produce videos for phones? Maybe that’s going to happen with AI.

Okay, and so that was the company that we built before that you sold to Snap?

The company was called AI Factory. I was fortunate to meet Mahi in 2018. He’s a co-founder of Higgsfield, a veteran of Silicon Valley who went through ups and downs and sold the company to Snap for $166 million. Then I was leading GenAI there.

Pause. No offense, dude. You come from a family of incredibly ambitious parents who pushed you to do well, and you just skip to the moment where you sold for $166 million. It’s a lot of money. How did that feel when you did it?

We both remember those times when capital for AI companies was not really that available, and AI multiples were not 200 times revenue as they are today, but closer to zero because AI was not a topic. So there was severe dilution, which we experienced.

But what was the round?

Back then, rounds of $1 million or $2 million—having $1 million or $2 million in investments—were considered to be really good. But it was still an opportunity for me to finally go to the United States.

After the acquisition, I permanently moved first to Los Angeles and then to Silicon Valley, and my dream simply came true.

Was it what you thought it would be?

That’s a good question. San Francisco is definitely a place where no one judges you by race, nationality, and so on. That’s truly phenomenal. There is definitely a meritocracy in the sense that it’s possible to meet anyone.

But at the same time, what I see among Silicon Valley investors is that they are extremely consensus-driven. I expected that last part 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 now we’ve sold to Snap. We’re in the United States. This is the moment that you wanted. How does Higgsfield come to be?

3. Higgsfield Finds Product Market Fit

Back then, in 2020, Snapchat was growing so quickly, and the face filters that my team had built were driving most of the daily new users. What was important is that we were able to run these face filters on mobile devices, so it was virtually free for Snapchat. It wasn’t like the current cost of LLM tokens.

It scaled to hundreds of millions of people throughout the world, and it was truly phenomenal to me to build a product that is still probably the most-used consumer media AI product.

But then I realized there were a lot of unmet needs in advertising. The average company cannot figure out how to be relevant on social media, and this is a major gap. 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 sell more. But at the same time, most companies in the world simply cannot do that.

Basically, no one can keep up with the pace of production for social media, as trends change pretty much every day.

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 better than nothing, but still pretty bad, right? Another solution was to take long-form video and cut it into short, vertically oriented videos. This was better, but still really not perfect.

It felt to me that, especially in 2023, it was absolutely clear that scaling laws finally work. It’s not just a concept from science; scaling laws work. Video just takes maybe 2 or 3 years longer than LLMs and coding, but it was clear that scaling laws should finally work in video as well, and I just decided to take a bet.

But I just want to go back. I get that, in terms of what we see, we want to empower these brands and companies to create amazing media for social media, but it wasn’t a hit from day one. I spoke to Amy at Menlo, who mentioned 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 that could work. We burned more than $10 million out of the $16 million raised in seed fundraising, so we felt we had just one attempt left.

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 optimizing for what was hype today, what the right narrative was, how we could hijack attention—all these things, really everything—instead of building a good product.

When we had less than $6 million left—I guess it was slightly less than $5 million, actually—I realized that the only thing we could focus on was leaning into product-led growth and finally settling on the belief that the best product was going to win. Then we started to talk to customers. We spoke to 8 creative directors about their experience with AI and what was missing. Everyone told us that camera control does not exist in AI, and camera control is so important to tell a story. So this was a very important bottleneck to solve.

We released our product on March 31 of last year, and since then, we have been riding this crazy wave.

Was it immediate product-market fit, then?

Yeah, it was immediate.

Is product-market fit like love? When you know, you know.

Yes, it’s definitely when you know, you know. For example, we don’t do any paid advertising. We have people on the team who scaled other businesses to over $1 billion and $2 billion in revenue with paid advertising. At Higgsfield, we decided to really make a bet.

You don’t do paid?

We don’t do paid.

Are influencers not paid?

That’s a good point. With influencers, there are different types, but typically there is some fee for video production and then some cost-per-click attribution, which works really well on YouTube.

You guys got into some controversy—I can’t remember what it was. You were paying people to promote for you or doing something rogue with influencers. Was that completely unfair? Was it, “My bad, we did do that”? How do you respond to that?

Look, I think the main takeaway from our experience is that it’s very important to own distribution. Distribution is now more important than ever. We basically outsourced it. We had just a team of 2 people on the creator and customer-success sides, and we outsourced to an agency. That was not a good experience.

I do just want to go back to part of the story. Where are you at revenue-wise today?

4. One Billion In Annualized Revenue

Today is actually an exciting day. As we record this, a Bloomberg article just went out, so we crossed $1 billion in annualized revenue.

If I had a gong here—

Mm.

—I’d be hitting the gong. A billion in revenue.

Yes. Actually, it took us 18 months to go from $1 million to $1 billion. For Cursor, it took 24 months. So we are probably the third after OpenAI and Anthropic.

18 months from $1 million to $1 billion?

Yes.

How do you calculate revenue? It’s a controversial topic. How do you 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.

Let me be transparent about that. What we do is look at revenue over the last 4 weeks and multiply it by 13. From what I know, OpenAI, Anthropic, and CloudBell all use the same methodology.

What’s very important is that we take revenue, not sales. So if that’s an annual subscription or an annual enterprise contract, we prorate it across 12 months and only take the piece that corresponds to 1 month. That’s 28 days, to be precise.

That’s the first piece. Second, it’s only live revenue. We are not taking 3-year enterprise deals and baking them into the $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?

Video AI is still relatively early. It is still probably 2 years behind coding in terms of adoption. 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 some significant revenue expansion. I always love to study the stories of the largest customers on the platform. One customer started 6 months ago spending just $99 a month on a subscription, and now we just signed a deal worth over $6 million.

$6 million?

$6 million a year. This level of acceleration is mind-blowing to me.

Dude, what are they getting for $6 million a year? That’s almost a Hollywood content team.

There are multiple trends, and all of them, frankly, are coming from Asia. 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 make hundreds of ads, if not thousands, a week, and can A/B test what performs well.

But we all know about short-form dramas, right? Short-form dramas today are an industry worth over $10 billion, owned primarily by Chinese companies, with a huge impact in China, the United States, and Europe—everywhere in the world. Most new shows there are made with AI end to end.

I think this adoption is obviously coming bottom-up, but it’s very difficult to refute this new reality.

What percent of revenue is consumer versus enterprise?

That’s a great question. Business revenue is slightly over 50%.

Wow.

Yeah. On the consumer side, it’s also important to break it down. Of that 50%, around 10% is pure consumer use cases. Pure consumer use cases are roughly people who use it on mobile, so our share of revenue from mobile is less than 10%. We are very different from many other companies.

But there are lots of aspiring creators—basically, people who are freelancers doing social media marketing projects and so on—who try to learn video AI so they can make more money. It’s true that their behavior is a little bit of a journey. 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 things out and become this new AI-native workforce.

It’s still important for us to educate them. That’s why we invest so much in Higgsfield Academy, the Higgsfield YouTube channel, and so on. But we are also fully cognizant that we will never be able to win in the market of $20-a-month subscriptions.

Why?

Because I think today Google and OpenAI are pursuing ads so much, but fundamentally, I think they are going to completely demolish their prosumer subscription markets, which are $20-a-month subscriptions.

Oh, so you’re saying that because they provide a horizontal product that’s very good, you’re just not going to pay for a lot of the verticalized products that you used to pay $20 or $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 already.

Are you seeing it cannibalize 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. I do believe this is just the most apparent example, but there are a couple more that are already happening. I believe that’s why, for Higgsfield, what really matters is how—even if we get someone on a $20-a-month subscription—we can show them value. How can we get them to upgrade and spend over $1,000 a year with us?

I can’t believe that’s $6 million a year from $99.

That's the best-ever slide on a fundraising deck.

Yeah.

All of our customers are going to do the same.

Exactly.

Can I ask—you mentioned their churn rates. When you look at 30-day retention rates for consumers and 90-day retention rates, what are yours, and what is good?

There is quite a massive drop within the first month simply because people don't fully realize the value. That's a core priority for us: to get better at showcasing the value.

Is it like half?

No, it's maybe like a 30% drop.

Okay.

But then it's really flat after that. We obviously look at logo retention. I wouldn't say it's great, because we all remember the B2B SaaS era: logo retention in month 1 was expected to be over 80%. Clearly, we have a lot of work to do on user education to get there.

But some things are truly phenomenal. When I look at the core business segments and NRR at month 12—obviously, you're going to argue, "It's an 18-month-old company. What are you talking about?"—when I look at the numbers I have today, NRR at month 12 is over 300%. It just never happens in B2B SaaS, right?

That's why I'm saying that while there is substantial churn in month 0 and we have to do a better job with user education to address that, expansion is unprecedented.

Can we actually unpack the 2 different go-to-markets? You've got consumer and you've got enterprise. I spoke to quite a few of your competitors, in all honesty, before this show, and I said, "Hey, you'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.

What I wanted to understand was, when you look at consumer growth, what worked, what didn't work, and how do you reflect on that?

That's a good question. First and foremost, 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.

We have an in-house team of over 150 creative professionals. 150. It's almost half of the whole workforce, frankly. Those people make product-launch videos and tutorials. For example, we made the first AI-generated movie, which is obviously a very sensitive topic. But what's important is that we open-sourced all of it.

What we learned is that for 90 minutes of TV-quality content, there were over 100 hours of AI-generated content. So creative decision-making—picking the right piece—is still very important. 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 subscriptions is through your 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've since walked that back. Can you talk me through why you chose your own models and why you walked it back?

5. Workflows Matter More Than Models

Obviously, this was my mistake. I'm going to do my best to be transparent. At some point, I really was thinking that chasing benchmarks was valuable, but I do believe this is just a sort of corporate psyop, frankly.

I was part of a large organization, so I know what happens. Everyone just thinks, "We need to show some progress, so we need to have some benchmark." But 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 leverage test data, use LLMs as judges for training the models, and use all the various tricks to basically game benchmarks and get quarterly bonuses and so on.

If I make my couple of million dollars a year in one of these labs, I can move to another lab easily. So that's unfortunately what's happening in larger organizations.

Can I just stay on that?

Yeah.

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 scores on benchmarks, which actually don't increase output efficiency.

Yeah. Look, I think let's just look at the outcomes we have today. Out of all the incumbents in the United States, when I look at OpenRouter data, the only relevant company is Google. When I look at China, where the obsession with benchmarks is probably less, we have Tencent, Xiaomi, and Alibaba—3 incumbents that are completely relevant—and, obviously, ByteDance trying to catch up as well.

What's your takeaway from that?

I do believe that, in the tech bubble, there is a strong obsession with benchmarks, which do not necessarily represent reality. But I can talk specifically about video. A lot of benchmarks for video today are really text-to-video, which does not represent actual workflows at all.

The way to think about video models today is as a modern rendering engine. Think about this as Unreal Engine or Unity, but with different types of inputs. It's virtually impossible to really define a visual output and direct the execution just through text.

If you go into our open-source projects, like the movie I mentioned, the average prompt length is over 3,000 words. That's the first thing. All these benchmarks we're talking about are not comprehensive in terms of the details of the prompts, and the people who are labeling them obviously cannot read 3,000-word-long prompts.

On average, there are also at least 10 image references for every scene. The reason that's important is because you need to define how the characters look, how the background looks, how the characters are positioned relative to each other in the scene, and so on. That's why prompting and the workflow are so complex. Benchmarks just don't represent that.

Going back to model selection, why did we decide we were going to do our own, and then why did we walk it back?

It's true that with VFX and camera control, we got very quickly from maybe 1 million to 20 million in ARR within the first 3 months. Then we released our own image model, which is really good at aesthetic photo shoots and product consistency. This allowed us to scale from 20 million to 100 million.

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 a specific use case, like these photo shoots, but only when that's what our customers want. It's all driven by customer feedback, not just by an ambition to conquer the world and build the best model in the world.

Do you think every company will have its own models? We're seeing Harvey, we're seeing Cognition, we're seeing Macaw, and we're seeing Ramp build their own models. Will every company have its own models with its own data, or will we all use a series of providers?

First of all, whenever someone says, "We built our own models," very likely what they mean is something similar to what's happened with Cursor. We remember, right? A lot of companies actually take an open-weights model and just post-train it on their own data.

Post-training can happen in 2 ways. The most important is whenever you have customer data around—like the decisions they make, the sequence of decisions—and you can teach the model to learn how to compress those 10 steps into 1 step. This type of reinforcement learning is the most valuable.

I think increasingly more and more companies will have to do that. We see this in the market as well. Most companies in the world today, most businesses, don't necessarily need Astra specifically. They don't necessarily need the newest Fable model. That's why OpenRouter reports that the share of open-source models went from below 30% to over 60% within this year.

What do you think the share of open models will be in 2 years' time?

Look, I do believe that, just because capitalism works, OpenAI and Anthropic are still going to have more than 50% of the market.

In terms of the dollars generated?

In terms of the dollars.

Yeah.

Right. Especially because coding still remains a very prolific use case, where coders are always jumping to the most recent model.

But for our market, we're seeing completely different dynamics. What's actually happening in social media marketing is that, as companies start to produce hundreds of ad creatives a week, they want to have the cheapest, most steerable models, because PhD-level intelligence is not necessarily needed to make a viral social media video.

That's where we've actually seen that we get 80%+ margins whenever we run open-source models—post-trained open-source models—but it can be much more cost-efficient for our end customer compared to proprietary models.

What's the comparison in margins between open and closed models for you?

The margin on our own models and open-weights models is over 80%, and then it almost doesn't matter.

For closed-source models, it’s probably between 20% and 30%. Then what becomes important is: can we actually steer the traffic? What makes me excited about Higgsfield is that agentic grows so quickly, and as companies start to create those agentic workflows to make more ads, we choose which model we can use. 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. There are certain amounts of work customers want to do, so how can we optimize the number of tokens required, and how can we pick the most efficient tokens for them? There are actually 2 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 a divergence of models. There are these state-of-the-art models that have to be really good at computer use, like Astra, or at coding, but they can be prohibitively expensive.

On average at Higgsfield, a person on the team spends over $10,000 a month on various models. Remember, we’re split across the United States and Asia.

So how much do you spend on models per month?

Internal usage of models a month is over $4 million.

Wow. How many people do you have?

We have close to 400 people, and I 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 vibe coding.

This month, I just caught a guy who spent over $30,000 in a week on the Astra model. He was frankly frustrated that some asset organization workflow and, as you said, basically auto-editing, is still not very good in production. He said, “I’m just going to do this myself,” and went 5 nights straight on Astra.

And it works?

We learned a lot. I wouldn’t say it was production-ready, but we learned a lot.

$30,000 in a week.

Yeah, many people spend over $10,000 in a week.

Do you mind?

Yeah, my finance team will probably say that I’m being too stubborn and too relentless about controlling the spend. Sometimes I feel it really goes out of control. $30,000 in a week is quite a lot. But we learned. This was actually a net-positive experience.

Okay, so the internal spend is $4 million, about $10,000 per head. What will that be in 12 months’ time, do you reckon?

Across the top engineers and across the top creatives, it’s going to keep growing. I do believe we’re going to get to spending close to $50,000 and $100,000 a month for those we can call 10X engineers and 10X creatives.

Unfortunately, I also expect that these people will ask for comparable salary raises as well, so I think that’s just going to correlate at some point. But for a lot of other jobs, let’s say legal, finance, and so on, it really stabilizes around $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 used to take time. Do we just have dramatically smaller teams with those 10X engineers, 10X designers, and 10X finance leaders?

I can definitely say that I had a feeling legal and customer support were going to be mostly replaced. That’s obviously one of the main operational mistakes we’ve made in the company: we didn’t ramp these teams quickly.

What we’re seeing today is that our legal team is over 10 people, and our customer success team is over 40 people. All of them use AI heavily. In the professions I’m close to today, I definitely can say that 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 now has over a 92% resolution rate on customer support for consumers. Pretty good.

It’s pretty good, but obviously they did invest a lot into that.

A shit ton.

But also, the way Nick thinks about that is in terms of the playbooks. We launch new products pretty much every week. We have to keep agents updated with all the information and so on, and due to the high velocity, having an extremely smart, coordinated team is very important.

It’s really interesting how increasing product velocity leads to harder customer support for agents.

Of course, because the agents are only as good as the context and roles they have. If the context and roles change twice a week, it gets a little difficult.

When you look at your engineering team today, what are they on? Are they on Cursor? Are they on Codex? Are they on CoolCode?

From March to June, everyone really moved to Claude, including the creative team. That’s where we started to see the creative team vibe-code functionality that we don’t have in production.

Then we started to see all the coders quickly move from Claude to Codex as of mid-June. Over time, the creative team, especially 10X creatives, moved to Codex as well. But I do believe it’s cyclical.

It is so cyclical. My question to you is: will we continue to see the velocity of model releases that we’re seeing now? In 3 years’ time, will it be like, “Oh, Gemini this week. Oh, Claude from Anthropic this week. Oh, OpenAI this week”? Or will we see a reduction in the model release rate?

I don’t think that’s going to happen anytime soon. For example, recently OpenAI announced that they basically built OpenAI for Law, but that’s only v0.

Over time, they’re also going to try to build smaller, specialized models—not exactly smaller, but really specialized models for certain use cases. Clearly, Astra excels in long-horizon computer use.

Do you buy that? I look at that GPT for Law from Astra and I’m like, “I’m sorry, I think it’s complete bullshit,” with the greatest of respect. It’s a very deep functionality required to serve some of the biggest law firms in the world—very, very deep and specific functionality.

It’s also very specific according to the different types of law. Plus, if you want to sell into these law firms, it’s a multiyear sales cycle with some of the stodgy old lawyers and partnerships. You can’t just say, “I’m OpenAI. Yeah, we’ve just hacked into the Australian government, by the way, but we’re here to serve your law firm.” Okay.

First of all, I think the ability to switch internal use just for internal teams outside of law firms is definitely happening. I think we both invest in a company called Solve Intelligence.

Love it. Very specific.

Very specific. Let me bring up a couple of examples of why Solve Intelligence is so special and how we learn from this.

What can happen very often is that a company wants to control the patent workflow even if it outsources the work. It’s very valuable just to have one system of record. Whoever can create an AI-native system of record is going to win.

Going back to Higgsfield, this is why it’s so important. There are so many systems today that are used just to store assets. Some people use Dropbox, some people use Google Drive, some people are going to try to use Miro, and some people are going to try to use Frame.io. There are many solutions.

But let’s think about what people need. They want to be able to search content, and marketers especially want to make sure that content is on-brand in terms of the visual identity, but also that it adheres to certain brand guidelines. That’s where semantic understanding and semantic controls become finally possible. They never existed before.

In our space, there are definitely other companies like Adobe and Canva that build the best software for the pixel-first era, where everything was defined with pixels. But that’s clearly not how the world is going to work in the future.

What we’re envisioning, and what everyone wants, is to be able to search and really work through the library of assets and all the knowledge through natural interfaces. Being able to own this interface and build the analytics—this system of record—is important.

That’s why at Higgsfield we’ve invested so much in Harness, so that it improves over time. This Harness also allows it to learn visual style over time, which Claude and OpenAI cannot necessarily do.

Do you believe in moats anymore? You’ve been around startups for a long time. We always talk about moats and defensibility. I largely think they’re bullshit.

We saw Lovable. When I invested, everyone was like, “Oh, it’s a wrapper. It’s a wrapper, you idiot, Harry.” Actually, it was a wrapper, but it’s about speed of decision-making, product execution, and building value over time very, very fast.

Instinct is a wrapper. Of course it is. It's not that difficult to build an AI assistant today, which is why there are so many, but they're building incredibly quickly, very valuable features, and you build it over time. Do you believe that moats actually exist, really?

6. The New AI Moats

I think it's very difficult to figure out where the value accrues in the supply chain. We do believe that there are only 2 ways of modern value creation, or moats, today. First is when you deliver the outcome, and for us, it's allowing businesses to sell more through AI ads. That's the first thing. The second thing is network effects.

Unfortunately, AI does not replace network effects, and when people talk about a swarm of AI agents talking to each other, I'm not sure this is happening in the next 5 years. That's why it's so exciting that within Higgsfield, we really wanted to empower the community to create more projects, open-source them, and really build a snowball where people can capitalize on each other's output. This is the reason why software grows so quickly, because it's so easy just to go and fork someone's project on GitHub. We were able to scale from, basically, I don't know, 10 seeded open-source projects 8 weeks ago to over 10,000 today. Seeing these types of network effects, I believe, can become a moat over time.

When we look at your growth, fundraising is a big part of it. It costs a lot of money to be able to spend $4 million on different aspects of inference burn. What was the best VC meeting you've ever had?

Obviously, Yuri Milner gets it.

How was that meeting? Was it in person?

Yeah, definitely in person, and Yuri definitely stays on top of all the trends.

How was it? Were you nervous?

I wouldn't say I was nervous. It was more to see how much of the market we saw the same way, and I was truly surprised that Yuri deeply understands this transformation of content, first and foremost. Obviously, it starts with direct-to-consumer AI ads. It starts with short-form dramas. All these trends come from Asia to the West.

Mm-hmm.

Fundamentally, we believe that most of the content on social media and in the world is going to be AI-assisted or AI-generated. This multi-trillion-dollar advertising industry—and contextual advertising is the main business model of the internet—is all going to be substantially disrupted with video AI. This industry is still going to be very valuable, but it's never going to be the same.

Did you know when you left the meeting with Yuri that he was going to write the check?

Sophisticated investors can play games. I've had so many scars. People really shook hands and said, "We do it at this price." The next day, what I learned was that they had called other investors and pulled the syndicates in to invest at a 30% lower valuation compared to what we discussed. These things just happen, so you can never be sure.

I think there's a discount placed on Higgsfield because you're not a Silicon Valley insider. Do you think that's fair? Let's be clear: you're at a $1 billion valuation now.

Yeah, yeah.

If you were a Silicon Valley company, that would easily be a $25 billion company growing at the rate that you're growing in 18 months.

Yeah, you could also argue that's what Cognition was valued at 50, right? So there is definitely an upside.

Okay. Upper bound even more. Yeah, 100%.

A couple of things that I believe are very important. First, we build for the long term. We have seen that the direct-to-consumer space, like e-commerce, can be disrupted. Shopify is a great example of how they have become infrastructure to build direct-to-consumer businesses, and we become infrastructure to essentially build distribution for direct-to-consumer businesses. That's one aspiration.

The second aspiration is obviously to be a plugin to Shopify. The company's worth over $200 billion. It's insane. As we think long term, these multiples don't matter that much. We know we're building for the long term; we're going to be over $100 billion. It's true that most people don't get the opportunity to go after the biggest industry in the world, but I wanted to drop another number.

When I looked at public companies—and I asked the team to double-check, so it's at least 4 people on the team who verified it; it's not some random fact—I found that when we look at public companies and exclude pharma and big tech, spending on sales and marketing is higher than spending on R&D. When it comes to sales and marketing, the goal is to deliver a personalized offering that converts the best. A lot of that is human work, of course, but a lot of that is going to be personalized videos in some shape or form. That's why I'm saying that it's good for us that many people don't understand the opportunity in this large market that we're going after.

You've mentioned Asian short-form dramas a lot. What percentage of revenue is from Asia versus the West?

The West makes up well over 70% of revenue.

Oh, wow.

Well over. It's just important to say that we learn a lot from trends coming from Asia. Higgsfield does not exist in China, for example, which is a massive market for AI. The largest city by usage is Seoul in South Korea, while the largest country is obviously the United States.

What's the biggest lesson from Asia that you've learned?

There is so much IP and so many products coming from Asia, and they all try to figure out direct-to-consumer distribution. That's why they lean into new tooling like video AI, which actually helps them achieve that. That's just a very different mindset. They feel that they could do way better if they could establish a direct relationship with the customer instead of having some other layer. That's why they go so much direct-to-consumer rather than using resale platforms and so on.

7. The Personal Cost Of Speed

I sacrificed a lot of life for the life and the career that I have, and I love it. Do you think you will one day regret not spending a day with your son in 3 and a half months?

This goes even beyond that, because from the age of 7 to 12, my mother had to work 3 jobs, so I didn't see her. My father spent all the time with me, going to all these camps with me. I was basically a minor, so he had to go to all these camps with me. I also played checkers. I was top 3 in the world, so we traveled throughout the world.

Then I did programming. He spent all the time with me, really dedicated his life to me—he did sacrifice. Since I was 21, he has had Parkinson's disease. Even having some access to capital and exits cannot fully change things, and this is something that is deeply personal, obviously.

But you don't need to do what you're doing now, Alex. I didn't need to anymore either. I still am. I still miss family birthdays. I still miss weddings, because mine is about a deep insecurity rooted in me being a fat kid. Why are you doing it?

Marc Andreessen described it really well. There are 5 archetypes. For me, it's just huge conviction about the technology, the markets, and the opportunity, and a huge fear of missing that.

Remember that my parents really taught me that there is a place in the world where good technology products matter. I remember when I was 6, there was this magazine about Bill Gates building Microsoft and not being very socially accepted everywhere back then. My mother told me, "These examples do happen in the world." I think she didn't fully understand that San Francisco and Seattle are different cities, but that's still deeply rooted in me.

Childhood shapes us a lot. What did your parents teach you?

For them, what was important was just to be in a merit-based environment, and that's why getting to California felt so important.

What's your biggest lesson on hiring, speaking of a merit-based environment? We see a lot of focus on that. You're the Cognition of the world; you hire math Olympiads.

Yeah.

What's your biggest lesson on hiring effectively?

I think one of the reasons Europe thrives so much—I know that you typically say otherwise, but let me just challenge you—is that the most relevant neoclouds today are Anscale, Airan, Nereus, and Crusoe. Crusoe, okay, is a Silicon Valley story. Airan is from Australia, Anscale is from the UK, and Nereus is from the UK and the Netherlands.

Let's talk about the companies on the application layer that matter. I know that you mentioned Mercor and Harvey, but Legora, ElevenLabs, and Lovable all deeply matter. If we just go outside of the model layer, because then I don't want to go into the Mistral topic, the usage numbers are very strong, but people for some reason don't believe in that. I don't know why, but public data shows that the usage is there.

On every other layer, Europe is extremely competitive. ASML, for example—without ASML, this whole thing just wouldn't happen. I think fundamentally what matters is whether Europe is going to figure out energy, but that goes outside of—that's above my pay grade.

What's very important to say here is that now there are more opportunities to create a company from different kinds of cities and different parts of the world, whereas before it all felt extremely centralized, and we are obviously excited about that.

Another thing about hiring is that in Silicon Valley, unfortunately, what I’m seeing is that people just jump between jobs every 2 years. That’s why I think Europe can be so competitive, because the sense of loyalty matters a lot, and that goes a little bit to the childhood we just discussed. Let’s say if you’re a Fulham fan, you’re not going to root for Arsenal just because they won or played in the Champions League final. But in the United States, if the Lakers are at the top, people are going to say, “Yeah, I’m a fan of the Lakers,” because it just makes it easier to start a conversation.

When you think about your own CEO style, what’s changed most?

In AI, it’s so important to look at actual signals and actual adoption and have access to raw information. I was obviously taught the corporate school of management in the United States, and when I look at the CEOs whom I learn from, it’s obviously Jensen, Elon, and Nick. Those 3 completely abandon all the management principles.

They aren’t necessarily fans of one-on-ones and soft feedback. All of them, I think, encourage being down to the point and knowing the details, whereas in corporate America that would be called micromanagement.

What management principle do you disregard that many people think is important?

I do believe that it’s as simple as hiring the best people to do the best work and figuring out how to retain them. Everything else is frankly secondary. People just create so much theory around that, and essentially there are so many fake rules that are disconnected from reality.

It’s really as simple as hiring the best people, empowering them to do the best work, and figuring out how to establish a relationship and retain them.

A lot of them do see dollar signs, and secondaries are a part of that. How do you think about doing annual tenders to retain people?

Across our team, roughly 50 are in California. We’re going to get to roughly 50 remote employees, and over 300 in Kazakhstan. I just hope we’re going to bring more dollar millionaires in Kazakhstan, in Central Asia, and in this part of the world than any other company.

What’s the labor arbitrage on cost between Kazakhstan and the US?

I know that a lot of people, when they look at Higgsfield, think about the arbitrage.

Is that not true?

Kazakhstan is top 5 in the world in physics. You look at the recent International Physics Olympiad for high schoolers, and they’re top 5 in the world, on par with the United States, China, and India. The core of our team is made up of people who won international competitions in math and physics. That’s the first part.

The second part about Kazakhstan is that they took the Soviet school of math and really upgraded it with Singaporean principles. The Singaporean education system is considered to be probably the best in the world. At least, many people in Silicon Valley believe that.

The government basically subsidizes thousands of high schoolers to study abroad, and many of these people come back. The density of talent is definitely there. It’s one of the 10 largest countries in the world, with a population of over 20 million, and we’re also actively hiring and bringing in talent from Europe and other countries in Asia. People enjoy some benefits, like a 15% personal income tax. It’s not bad.

And don’t even get me started on the fucking UK—

Yeah.

—we’ll tax you to breathe. Seriously, in the UK, you get your paycheck, and it’s, I don’t know, £100,000. Then you get to the end, and it’s kind of like £3,500.

Yeah, but it’s also English common law, so it’s not that bad as people think.

You move here. Let’s swap places. Do you have a mega-pad in Kazakhstan?

No, I don’t. I don’t own any property.

What? Why?

Remember that I come from an Asian family. Whenever we sold the company, I made over $1 million, and I spent all that money buying apartments for my parents, relatives, and my wife’s parents, because it’s just part of the culture.

The extended family is not small by any means. But it’s just part of the culture to give back. When it comes to my family, especially my parents, they obviously sacrificed a lot, so I felt like I had to give back at least monetary things that I could provide.

I drive a Tesla Model 3, and that’s leased, so I’m not a guy who’s going to show up with a Lamborghini or Porsche.

Do you invest? We mentioned Solve Intelligence.

I did before, but now I spend roughly 80 to 90 hours a week on Higgsfield. I try to spend ideally at least 3 hours a week with my wife and at least 5 hours a week with my son. Sometimes I catch up because when I travel for 1 week, 2 weeks, or 3 weeks, I try to take Sunday off to spend the whole day with my son.

Over the last 3 months, I was able to find 1 day when I spent the entire day with my son without emails and without talking to team members.

I get in trouble for this, but I think there’s no shortcut to hard work. The harder I work, the luckier I get. I meet more founders, find more great companies, do more shows, and have more success. Do you buy the bullshit of balance, where it’s okay to leave at 5:00, be home for bath time, and still crush it?

This is a good question. Obviously, being an immigrant, I always have to prove that I belong. I feel like now people accept and recognize that Higgsfield is probably a top 10 applied AI company by revenue, and probably number 1.

When it comes to hard work, the people whom we know in common—we talked about, let’s say, Peter Sales, a legend in the consumer space, and obviously Jack—I spend a decent amount of time with them and the other product leaders at Stamp. The density of product talent at Snap was unprecedented.

All of them work really hard. All of them are smart. None of them just checks emails for 5 hours a day and calls it work. Each of them is deeply rooted in the recent trends in product, product design, and activation. They know data really well. So, yeah, I don’t believe that there’s any shortcut to hard work.

3 hours a week with your wife? I don’t know about you, dude, but mine would dump me for 3 hours a week. How do you make marriage work on 3 hours a week?

I’m very grateful to my wife for being patient. It’s also very different in Asian culture. It’s more natural to try to make sacrifices for each other, and I’m deeply grateful to her for supporting me.

Sometimes, at this scale, I get invited to parties. I always send her and don’t show up myself. I don’t know if I piss people off, but this happens very frequently.

So wait, you say yes and then she goes?

Yeah, I say maybe we can both come together. Then there’s always some urgent fire at the last minute, and my wife just goes without me.

What fire was most urgent? What was the “Oh, fuck”?

For all the things we touched on earlier, whenever we weren’t very good at communicating the features or we felt like—I mean, now it’s a team of 40, so life is much better—but in the early days, obviously, I was involved in all the fires.

Recently, all the different types of attacks on AI companies have been crazy. LLMs are being used to hack companies. There are new types of LLMs being used to commit fraud, like bots using credits and then doing automatic refunds.

Since I have a machine learning background and a data science background, I can still move the needle substantially when it comes to statistics and data. So, yeah, I have to be involved somehow. These LLMs amplify many types of behavior, including various types of attacks and fraud, but we have to fight against that.

We’re going to do a quick-fire round. I’ll say a short statement, and you give me your immediate thoughts. What have you changed your mind on most in the last 12 months?

I was thinking that HubSpot was going to become obsolete and that everyone was going to build their own CRM. But especially when we hire and scale a B2B go-to-market team, having a familiar interface matters a lot.

Wow. I would still say they’re going to get fucked. You think that stickiness is there with SMBs?

Yeah, I do think so, especially when I hire go-to-market talent.

Wow. Why? What is it about hiring them that makes you think that? They’re just so used to it?

People who are very good at understanding customers and talking to customers may not simply accept a new interface so quickly. Having HubSpot as a system of record and being able, if there’s any mismatch, to understand where the data flow went wrong—I think that’s still very valuable. It’s the familiarity.

What do you believe today that everyone else thinks is fucking crazy?

I think people still don’t fully appreciate that most of the content on social media is going to be AI-generated. There are going to be some shows, like yours, where the content is authentic.

It’s going to be a 10–50x higher CPM than AI-generated content, so there’s going to be way less content created, but it’s going to create way more value than AI-generated content. Even when I look at your content specifically, you’ve made multiple very successful Shorts—millions of views, better than anyone else in this space—and you do a lot of overlays.

Mm.

While the content is authentic, I think we should do a better job so that you use Higgsfield, at least for the overlays on top of existing videos.

Dude, I would love that. People don’t know this, but they take 3 hours. I spend 2 hours a day just doing Instagram now. We decided that Instagram short-form was going to be a big new push for us. That’s 2 hours a day just for me: I write the scripts, and then I record them. Then it’s 2 people, 6 hours per one, for those 3.

That’s extremely smart of you. Going back to some of the topics, clipping is a huge topic, and that has its own upsides and downsides. Obviously, everyone sees this opportunity to win and build a massive top of funnel—hundreds of millions of views with short-form content—as long as you can have downstream monetization or value creation like you do.

Totally agree with you. What job today doesn’t exist that will be big in 5 years?

In 5 years, especially in our space, creative directors are going to be talking to computers and generating stories in real time in video. AI is going to help create multiple variations. Today, there is no word to really describe that, because there are scriptwriters, then screenwriters—those who are going to break it down shot by shot. Then there are people who do the storyboarding. Then there’s the person who oversees all of that, like a movie director and so on. There are so many parts of that. Eventually, taste is going to matter a lot more than just having stories to tell, and there is no word to describe it today.

Who do you not have on your board that you would most like to have on your board?

Maybe among more professional CEOs, definitely Frank Slootman. Going back to the point, I was curious all the time: Does a no-bullshit culture exist in California or not? Can it allow companies to scale so quickly? Is it possible to build a successful enterprise go-to-market motion with a no-bullshit culture? When I read his book, Amp It Up, I realized it was possible, so I’m a huge fan. I’ve watched all his interviews.

The challenge with him is that he’s amazing—he’s the best leader by far—but sometimes you can do it at the sacrifice of product advancement. He built a go-to-market machine at Snowflake, but Databricks wiped the floor because they made product the priority, not go-to-market. That was dangerous. I prefer Chad Peets. Do you know Chad Peets?

No.

Oh, dude, this guy is no bullshit. I’ll introduce you afterward. He’s the best sales leader in the world, and he is no fucking bullshit. Unbelievable.

Then we probably should have him on the board.

Oh my God.

It’s possible.

Find any way to have him on the board. He is terrifyingly good. So, what’s the biggest lesson from Snap?

The momentum doesn’t last forever. Today, Snap’s market cap is below $15 billion. There are lots of memes on the internet, but this is a great company. It cares so much about trust and safety and the experience, and it puts those first.

Do you think it is a great company? No offense, but it’s been mismanaged to shit. Its SBC is through the roof.

Ah.

It’s tough to say it’s a good company.

Yeah. That’s why I say that momentum doesn’t last forever. When Snapchat was worth $80 billion and the gap with Meta was less than 10x, it felt like, “Oh, we could just go explore.” We just really massively invested. But momentum doesn’t last forever, and that’s my core learning. While we do have positive momentum, we don’t take this for granted. Clearly, the nature of capitalism is that there are ups and downs, and since we’re building for the long term, we should capitalize on the opportunity, like with the fundraising, and just keep pushing progress every day.

What is the reason the divergence between Meta’s market cap and Snap’s market cap has increased so significantly, if there was one reason?

Just to say this straight, a lot of public companies did not figure out their AI story. Snap, unfortunately, is part of that. We’ve seen other great companies like Figma trying to tell their story. You mentioned Canva. It’s not necessarily easy to be successful in private markets and public markets, and Zach is one of the best CEOs of all time because he managed that.

He’s such a fucking beast. He’s such a beast. You watched him last night at the event, and you’re just like, “Ah, now I get it.” That totally makes sense. And you know what? With Scale and Alexandr Wang, I was one who was like, “Really?” He basically acquired a second CEO. You know, Alex is now the CEO of Muse, and he’s crushed it. Crushed it. What an effective buy for 0.5% of your market cap. Do you know what I mean?

Yeah. Look, this happens with Instagram and WhatsApp. That’s why I’m saying that we should maybe put Meta—

Yeah.

—a little bit in its own league.

Yeah.

In its own league.

But he got rid of Systrom and Krieger. Here, he’s been like, “No, no, no. You, Alexandr Wang, are my guy.” Do you see what I mean?

Yeah. I do believe that it’s a little bit early to look at the whole Meta AI initiative. We probably need to see a year of successful launches and so on, and then we can look back and see what was good and what was not good. But at least the consistency of the storytelling and explaining what he is doing to public investors, being able to articulate why Muse is so different, is phenomenal.

Okay. Revenues, say, are a billion. What are the revenues in 12 months’ time?

Our current business model projects 4.5 by the end of next year, but this basically involves substantial deceleration. That’s what my finance team told me—there are a couple of strong quant people—and that’s just how the business works. But we’re still pushing to grow at least 30% month over month.

What do you think it is? They said $4.5 million. What do you think it is? This is me asking you, not your finance team.

Over $10 million.

Over $10 million?

Let me tell you why. A lot of adoption in the creative AI space is driven by monetization, with all these direct-to-consumer brands making more ads and also creating aspirational cinematic AI content, as this inspires creatives to explore the tooling. It feels to me that Hollywood is starting to embrace AI today, mostly as a tool for hybrid production, just as a new form of CGI. The sentiment has really shifted from strictly negative to neutral to slightly negative. In private conversations, yes, there are maybe more than half of S-tier talent who are going to say, “We’re anti-AI forever.” But increasingly, more and more people are asking, “Can we sell more stories with AI? Can we overcome certain budget limitations that existed before? Maybe AI can help tell new stories that we couldn’t sell before.” I do believe this change in perception, at least from my conversations, is extremely positive.

If you’re at $1 billion today and $10 billion in 12 months, where do you peg the next fundraise? If you’re at $1 billion, say, on a conservative multiple, you’d be at, you know, $15 billion. But if you’re hitting $10 billion next year, you’re, like, pegging the end of the year at $80 billion?

We’re not chasing just the valuation, because the goal is to make sure that the company can be sustainable over time in the public market. There is a lot of company building to be done beyond just chasing the revenue.

Do you want to be public at some point?

Yeah. I do believe that Hexcell has great potential to be bigger than Amplivine and Shopify. Fundamentally, building is one part of that. Shopify is one layer of infrastructure. Then for coding, there is obviously Claude. There is Codex. But what matters is distribution over time. Distribution matters. You know that better than any other VC.

Dude, it’s my business. That’s why we do what we do. Yeah.

Exactly.

Dude, I cannot thank you enough for being so amazing on the show. You’ve been fantastic. I’ve loved doing it—you can tell—and you’ve been an amazing guest, so I really appreciate you joining me today.

Thank you so much. It’s a pleasure.