Everyone said it was "just a wrapper." It hit $1B a year in 18 months.

You've probably seen Higgsfield everywhere. It doesn't make a single major AI model. And it now brings in a billion dollars a year.

higgsfield in old computer screen with ai videos around it

Higgsfield

$1B

Annual revenue run rateEstimated

Founder
Alex Mashrabov
Annual revenue run rate
$1,000,000,000

Estimated

Jump to the proof

September 24. Higgsfield announces it just crossed $1 billion a year in revenue. Eighteen months in.

The same day, OpenAI switches Sora off for good. According to Forbes, Higgsfield was Sora's biggest customer.

Your biggest supplier dies on your best day... and nobody at Higgsfield looks worried. Why?

Let them cook.

$1,000,000,000 a year in revenue · 18 months

Why was everyone laughing at it?

Because on paper, Higgsfield is just a middleman.

You open it and you get almost every image and video model that matters, from Google, from Kuaishou, from ByteDance (the company behind TikTok). It's one subscription with a unified credit system. Higgsfield built almost none of these models.

That's what we call a wrapper. An app that packages someone else's work basically.

And the math of a wrapper looks bad. The lab (Google, Kuaishou, ByteDance, ...) gets paid, the wrapper adds its cut, so at the end the customer pays more. And the worse part, the lab can turn into your biggest nightmare overnight. We've seen it happen.

So the comments all said the same thing. "Why not just go to the model directly? why pay the wrapper if I can get the same thing for cheaper?"

So why pay the wrapper?

Good question! (if you have been asking yourself that)

Short answer: you don't have to, that's the weak spot. Every one of these models is a few clicks away.

So why Higgsfield then? Because Higgsfield sells the part the model doesn't give you.

Type "a man running in the rain cinematic" into a raw video model. Technically it works. It also looks like a home insurance ad, I tested it, it's really bad, don't do it, save your money.

The model can do a tracking shot. You just have to know how to ask for one. And I'm pretty sure none of us is Spielberg.

So that brings us to spring 2025, when Higgsfield launched its web platform. It bet on that gap. It brought in around 200 people from Hollywood (directors, cinematographers, editors ... not Spielberg just yet but they managed just fine) to build about a hundred ready-made movie shots, camera moves already set. According to the founder, that's the moment everything took off.

How do you show people a trick they've never heard of?

Nobody wakes up in the morning thinking "I wanna do a dolly zoom in my AI video clip". Somebody has to show them.

That somebody calls himself "the Higgsfield guy" on X. His name is Adil Alimzhanov, and he's in almost everything they post. Tutorials. A full ad built live. A face-off against a professional VFX artist. In his own short film, he jumps off a cliff onto a dragon and flies through a waterfall, nearly all of it generated from his own face.

The channel puts out two or three videos a week. Some episodes pass 2 million views. This summer they went further: an AI feature film with the licensed faces and voices of MMA fighters, including former UFC champion Israel Adesanya.

Then the second effect kicks in.

You watch the video. You go home. You try the raw model, without any of Higgsfield's presets or templates. Your result is... shit tbh. You blame the tool, and you go back to this Higgsfield guy's method.

How do you market a product you don't even make?

You let the people who make it do your launches for you.

On February 4, Kuaishou released Kling 3.0, the best video model on the market that week. About twenty minutes later, Higgsfield announced it. Available now. Unlimited. Exclusive on day one.

Twenty minutes.

When Google shipped Nano Banana Pro, its best image model. Guess what? Higgsfield had it the same day.

This has a name: newsjacking. You plug into attention that already exists instead of building your own. And for Higgsfield, the news is every single model anyone releases. Their changelog this year averages around six new things a week!

It's kind of like Zara fast fashion for AI models. Only difference is that Zara has to design and sew all of it. Higgsfield gets it delivered, by other big AI companies like Google. Nice huh?

Newsjacking only buys you a few days, though. Then everyone else has the same model.

And how do you get everyone else to talk about you?

This is the crazy part.

Normally a brand finds creators one by one, haggles over a price, approves every video, and pays before knowing if any of it works. Higgsfield flipped it. They created Higgsfield Earn, where anyone can make a video with the tool, post it, and get paid by the view.

Three weeks in: more than 10,000 creators and 50,000 videos.

They only pay for what works. They don't need to guess which video will blow up, because out of 50,000, a few will. And every single one is a free demo of the product.

The founder calls creators their "natural distribution channel."

So you see Higgsfield in a YouTube tutorial on Monday. In a Reel on Wednesday (We've all seen Jean Philanthrope and his gang, now you know).

They become the default answer to "how do I make an AI video?"

Same trick with their film festival: more than 67,000 films, all made on Higgsfield. There's an audience award, so every entrant campaigns for their own film... to their own followers.

Then came the night of the billion-dollar news. Higgsfield posted an even bigger number: 100% cashback on its API models, from a $20,000,000 pot, first come, first served.

Now read the fine print. The cashback comes back as credits you can only spend at Higgsfield, and you spend a dollar to get one back. In practice, that's actually about 50% off your generations for about six days.

But nobody shares "50% off for six days." The $20,000,000 version passed a million views.

Why don't people cancel?

Remember Sora at the beginning of my talk?

A wrapper that only sold access to Sora would have lost a huge chunk of its business that day. A Higgsfield customer picks Kling, Veo or Seedance instead and moves on.

Companies stay for a duller reason. And lately the best video model changes every month. Without Higgsfield, that's a new account, a new card, a new security review, the whole team trained again and a new tool to learn. With it, nothing moves.

And for everyone else, a new thing drops every week. Cancel for a month and you'll miss the next model. Lowkey, that's the real retention plan.

The catch

"Cool. I'll wrap a few AI models and wait for the money."

It's actually a dangerous game.

Sell only the access, and every new model is a countdown.

Higgsfield sells the hundred shots, the guy on the dragon and 50,000 videos it didn't have to film. Sora closing didn't touch any of that.

The Secret Ingredient

Here's what I think nobody laughing at "just a wrapper" saw.

Most startups grow at the speed of their own team. Higgsfield grows at the speed of every AI lab on Earth.

Each time Google or ByteDance ships a better model, Higgsfield's product gets better overnight. Someone else paid for the research. Higgsfield's own team only builds the layer on top: the shots, the creators, the events.

That's two engines. It pays for one.

Look at Adobe. Firefly now offers more than 30 models Adobe didn't make. The giant is turning into a wrapper also. Adobe announced more than $500,000,000 a year in AI revenue at the end of 2025. Higgsfield claims double that, on its own.

So the people laughing were right. It is a wrapper. They just laughed at the wrong part.

Remember the morning Sora died? That's the test for any product built on someone else's model. If that model vanished tomorrow, or became free... what would you still be selling?

Proof of cook

Annual revenue run rate$1,000,000,000
PriceOne subscription, one credit system, every major image and video model
Built inWeb platform launched in spring 2025
StackOther people's models (Google, Kuaishou, ByteDance, OpenAI until Sora closed)
Main channelCreators paid by the view (Higgsfield Earn)
What made the differenceAnnouncing every new model the same day it comes out
ProofEstimatedOct 14, 2026