Higsfield AI CEO Alex Mashrabov: $200M ARR in two months, sub-$1 video production, and the rise of AI-native ad agencies

back to the AI world. We have Alex from Higsfield AI in the Reream waiting room. Let's bring him in to the TVP Ultra. Alex, how are you doing?

He's back.

Doing great. It's great to be back.

It's like we were just talking about the billionaire tax. You [laughter] guys something that uh uh fortunately your business is doing so well.

Yeah.

Uh

focus on growth. Focus on the good things. Tell us the tell us the best news in your world this year.

Give me that mallet.

Oh yeah.

Give me that mallet.

What's the Yeah. What's the latest what's the latest metric?

Absolutely. So what's exciting uh about Hicksfield is that we see emergence of AI native social media agencies. So really teams from maybe 10 to 30 people making commercial ads end to end with AI.

Yeah. We just talked to Yeah, we just talked to Sean Frank. He said he's a huge fan. He uses it for Ridge Wallet and has been able to generate custom ads for, you know, tons of different markets, local areas where it was just completely unfeasible to to cast so many actors to to to play so many different parts for every single different college team that he sells products for. Uh yeah, it's a it's a fascinating time. So this is mostly driven by agents or brands going direct. Uh what's the biggest driver of growth for you?

Absolutely. We are excited to see Fortune 500 brands really hiring primarily hiring really hiring smaller really hiring smaller agencies. Yeah.

So that they then use Hixel to make those ads. Um and I think although the most exciting segment is definitely direct to consumer especially when AI can deliver not just the efficiency not the cost efficiency but also higher level of personalization. So the example which you just brought up completely makes sense whenever someone needs to customize for 30 60 100 different uh user groups let's say sport teams um geni delivers against this goal and pixel is at forefront of that. What are the compute costs look like looking like these days? I mean uh the the big the big uh news is that you you jumped from a 100red million to 200 million topline in just two months. That's insane. Congratulations, Jordy. Hit the app level. Congratulations. Uh but but on the flip side, I mean if you I mean you are raising money. Obviously the capital markets are open but uh if you were really spending on inference that could be uh you know difficult. How is how are the margins developing? How are you uh optimizing for inference cost? Um what are you seeing on just inference cost trends broadly? Uh are you are you happy with the progress or what else can you tell us about uh the the the cost structure of this business?

I have good news and I have bad news. Okay,

look, I think on the good news is that video models gets more capable.

Mhm.

So that over time marketers can just upload all the brand assets in one place

and then the model just remembers all the context.

Mhm.

And this helps to make not just one video but multiple videos and uh extend existing marketing campaigns. I think this is going going to be a major unlock. Mhm.

Think about marketing agents who actually remember all the previous campaigns, all the performance of the previous campaigns and so on.

Yeah.

Although there is a bad thing like these models they gets bigger and bigger [laughter]

there is no other p there is no other path arounds. I think most of the video models this last year were maybe between 10 and 50 billion parameters.

Mhm. and it is going to scale well beyond 100 billion parameters.

Okay.

So I think um so the although the good news is that in the same time we're seeing customers spending thousands dollars a month on the platform.

Yeah.

Which tells us that slay likely make thousands of videos. Cost per one video is still quite below $1.

Yeah. This is this is very interesting times as we move from a concept of

well that's an insane that's an insane reduction in cost because uh

I'm sure

the most efficient video shoot on your iPhone you're going to spend a day even if you're paying a market intern $20 an hour

yeah when brands are doing trying to generate UGC content historically you know now they can use Higsfield and and

generate uh stuff on the fly but historically you'd be like okay I'm setting a budget of two a $100 an asset and then you'd have a bunch of people like actually record on their iPhone to generate it. And so if you're saying it's, you know, a dollar an asset or sub a dollar, it's like a 100x reduction in cost. It's pretty significant.

Absolutely. I think cost reduction has already I mean it has [clears throat] already happened compared to physical production. And I think this batch content generation meaning that we make not just one video but hundreds of videos simultaneously. I think this change happening this year. So the goal for Hicksfield is to lead those agent like this new wave of aentic behaviors and build courser for video.

What about uh getting the ingesting the data on performance? Do you have API integrations with all the big platforms where you're ingesting uh ad data from Meta and Apploven or whoever else you're distributing video content through or uh are is it PDF upload, CSV upload like how are you thinking about actually creating that feedback loop because that seems extremely powerful.

Yeah, absolutely. On that angle I can tell that probably 2027 is going to be a year of major revolution in adtech worlds as connecting generation with performance data is going to become a major points.

Got it.

Uh this year we collect a lot of these data from social media platforms. We build most of these integrations.

Sure.

Um and establishing direct relationships with the platforms. uh very we collect a lots of very valuable data points directly as we see which generations results in an actual ad.

Mhm.

So that we can actually run reinforcement learning to streamline content creation process. I think this is the main priority for 26 and 27 is going to be a year of full revolution in advertising technology space.

What uh what's going on in the hardware side? what would be most valuable to you? We saw Nvidia uh acquire Grock with a Q. Uh Grock's whole pitch has been very fast inference uh usually for textbased LLM models. When I think about the work of a marketer generating a bunch of different video variations, if they fire that off and they come back an hour later, that doesn't seem like a terrible experience. Um, so are you more focused on higher fidelity, larger parameter counts, uh, over speed? Are you interested in what's happening in custom silicon in a particular area or do you just want more and more Nvidia GPUs at your disposal? Like what's what's important on the infrastructure side for your plans over the next few years? This is this is a great question and I think we have seen that custom silicon can perform really well.

Uh Gemini 3 is the first multimodel LLM which has these emerging capabilities like which is not a banana pro model.

Yeah.

Um this model essentially made Adobe obsolete. [laughter]

Yeah. Who needs to Yeah. Cuz I mean who needs to go to Photoshop?

Yeah.

Yeah. Look, I mean I I'm not saying like especially for social media marketing. Yeah. If we're being serious. So cuz I mean going and if let's say uh Jordy you assign me to go and make uh five ads for you a day.

Yeah.

And try completely different stuff. Uh I don't have time to go to Photoshop and push every pixel to make it perfect. I better use model like Nana Banana Pro to really

deliver against the vision. And it's semantic based editing, not just like pixel pushing. Yeah.

So, um

I kind of miss I kind of miss when you you used to be able you used to see a like a like really high quality Photoshop on the timeline and part of why you appreciated it is like somebody really needed to put

the effort in or they needed even if they're not great at Photoshop, they hired somebody and it's like you and nowadays it's

you know they got the lasso tool and they cut that character out and slapped that text over there, threw a drop shadow on it. Now it's just a prompt. It's fascinating.

Yeah, now it's just a prompt and it's done with custom silicon. So I think we're gonna see a renaissance of custom silicon for sure. I mean Gemini and Nana Banana Pro are amazing model with this emerging capabilities and we we really embrace those create those creators and social media marketers who creates ad with AI end to ends. So they don't have to go to Adobe toolink at all.

Mhm. uh how how are you tracking the uh the problem of understanding what's what's AI generated what's real this is now a daily

I feel like it used to be like even like 12 months ago it was like you know once once a month you'd see something and you're like oh is that AI or is that real now it's like almost every single day

oh I see stuff on Instagram all the time where I know it's AI but then I go to the comments and and everyone's happy no one's it used to be people would clock it and then even if they couldn't tell, they would key off of the comments and and then kind of pile on. Now people are just like, "Yeah, this is good content." Uh yeah, what's your take on on AI detection uh how it's working through its way through the internet?

Look, I think that we're we're going to see more standards emerging. I think Europe in European Union, it's a big thing.

We're going to see something by the end of the year coming on that front, I guess. Um, I can just tell you that direct to consumer brands, they primarily don't care if that looks like AI generated content or non AI generated content as long as it catches eyeballs and it sells.

Um, they just keep coming to Hixfield and making more of such videos. So the best the best performing AI generated ads, they look like AI ads and they don't look like real ads. I think that's that's my main observation.

Yeah. Yeah. Yeah. I mean, there's a certain element where, you know, the the iPhone, the selfie video, it never actually wound up looking like an HBO show or a James Cameron film. It never wound up looking like, you know, it was shot with a $100,000 camera, but everyone just got used to it and they like the aesthetic and we grew accustomed to that aesthetic and it has its own aesthetic. And so if AI content, even if it retains its aesthetic for a very long time, even though I think it's going to be indistinguishable, already sort of is, but even if it does, people [clears throat] could just become fans of that and be like, "Yeah, I like I like the way it looks."

Yeah.

Yeah. Interesting.

Uh,

anything else?

Very cool. Well, I also my buddy texted me. It's Alex Kassen. You guys just had lunch?

Oh, no way. [laughter]

Yeah, absolutely. We're just chatting about that. Um, small bolts. Yeah, he he's a great guy to know uh in this in this town. So, are you in LA right now?

Yes, I'm in LA. I love

you. Should have come you should have come by the studio.

Yeah, next time.

Could have been sitting here. You could have been hitting the gong.

Could have been moving the goalpost

next time. U

absolutely.

Anyway, thank you so much for taking congrats on the progress. The growth is absolutely insane. Uh and look forward to talking more this year.

Yeah, we'll talk to you soon.

Cheers.

Thank you guys.

Have a great weekend. Goodbye.

All right. We can close with one of the best ideas that I've heard recently. It's from Ryan Peterson. He says we should have an AI transfer portal like in the NCAA. Couldn't agree more. Well, that's our show. The bomb has been planted. Thank you for tuning in today. Thanks for watching. Leave us five stars on Apple Podcasts and Spotify. And sign up for our daily newsletter at tbp.com. We send out our newsletter every weekday.

Enjoy the weekend.

We will be live Monday.

We will be live Monday. We'll be back at 11 a.m. Pacific. So, tune in. We have a couple live inerson guests. A couple go touch some grass.

It's going to be a lot of fun.

Well, hope you have a great for being here with us

and goodbye. Nice work,