20VC with Harry StebbingsThe Untold Story of Higgsfield | Burning $4M a Month on AI Models | CEO, Alex Mashrabov
20VC with Harry Stebbings
Introducing Higgsfield and Its Scale 0:00
Higgsfield has just crossed a billion dollars in annualized revenue, making it the fastest-growing consumer AI company to hit that mark, ahead of Cursor. Internally, the company burns over 4 million dollars a month on AI model usage, with some employees personally spending more than 10,000 dollars a month, and in one case a person spent over 30,000 dollars in a single week testing an Astra model.
From Kazakhstan to Silicon Valley 1:31
Alex grew up in a family of academics with roots in Uzbekistan, where his parents pushed him into competitive programming from age eight, and his mother worked three jobs to fund his education. By 19 he reached the top three in the world in competitive programming. Instead of pursuing academia, he moved into startups, first working on pre-transformer neural networks for language translation, then founding a company inspired by Uber's rapid spread, betting that AI-generated video would dominate phones. That company, along with cofounder Mahi, sold to Snap for 166 million dollars, funding his permanent move to the US.
Building Higgsfield From Near Failure 7:30
At Snap, Alex built face filters that scaled to hundreds of millions of users nearly for free, but noticed most brands couldn't keep pace with social media trends. This gap led to Higgsfield, aimed at giving companies tools to produce fast, trend-relevant content. Early attempts, slideshow tools and video-cutting tools, failed to gain traction, and the team burned through 10 of 16 million dollars raised without product-market fit. With under 5 million left, Alex refocused on product-led growth, talked to creative directors, and learned camera control was the missing piece. Releasing that feature on March 31 last year produced immediate product-market fit.
Crossing a Billion in Revenue 14:01
Higgsfield hit a billion in annualized revenue 18 months after reaching its first million, faster than Cursor's 24 months. Revenue is calculated by taking the last four weeks, multiplying by 13, and counting only live, prorated revenue rather than multi-year contracts. One customer grew from a 99-dollar monthly subscription six months ago to a 6-million-dollar annual deal, driven by trends like AI-native e-commerce advertising and short-form dramas, an industry worth over 10 billion dollars, largely from Asia.
Revenue split and consumer behavior 18:02
Business revenue makes up slightly over half of Higgsfield's total, while pure consumer use on mobile accounts for less than 10 percent. A large chunk of the rest comes from aspiring creators, freelancers and social media marketers trying to learn video AI to earn more money. Their usage is churny at first, but most return within a year, and the company sees them becoming a new AI-native workforce, which is why it invests heavily in education through its Higgsfield Academy and YouTube channel.
Betting against $20 subscriptions 20:30
Alex argues that Google and OpenAI will eventually wipe out most consumer subscription products priced around 20 to 30 dollars a month simply by offering strong horizontal tools, something already visible in how such tools eat into Canva's easier design use cases. Because of this, Higgsfield focuses less on holding customers at a cheap monthly price and more on showing enough value to push them toward spending over 1,000 dollars a year, turning a 99 dollar entry point into meaningful annual revenue.
Retention numbers and enterprise expansion 21:00
Consumer retention drops about 30 percent in the first month before flattening out, which Alex attributes to users not yet grasping the product's value, an area he says still needs work. On the business side the picture is different: net revenue retention at month 12 is over 300 percent, a figure he calls almost unheard of in traditional B2B software, even for an 18-month-old company.
Influencer-driven growth strategy 23:01
Higgsfield's consumer growth has come entirely from its own content rather than paid ads. An in-house team of over 150 creative professionals, nearly half the company's workforce, produces product launch videos, tutorials, and demonstrations of professional-quality output, including an early AI-generated movie that was open sourced. Making that 90-minute film required generating over 100 hours of footage, underscoring how much creative curation matters even with capable models.
Walking back proprietary models 24:30
Alex admits it was a mistake to chase benchmark performance early on, calling it a form of corporate theater where researchers at major labs game test data and use tricks to hit quarterly targets. He points out that among US incumbents only Google remains broadly relevant by usage data, while in China, where benchmark obsession is weaker, three or four companies stay competitive. For video specifically, he says benchmarks mostly measure text-to-video tasks that don't reflect real workflows, where prompts often run over 3,000 words and scenes need roughly ten reference images to define characters and settings.
From custom models to customer-driven choices 28:33
Higgsfield's own VFX and camera-control models took annual recurring revenue from about 1 million to 20 million dollars in three months, and a later image model built for aesthetic photo shoots pushed growth from 20 to 100 million. Now the company only builds custom models when customer demand clearly calls for it, rather than chasing the idea of building the best model outright, and it lets customers choose which model to use in over 40 percent of cases, treating model routing as a core part of the product.
Open versus closed model economics 31:00
Alex expects open source models to keep gaining share, noting that on OpenRouter their usage rose from under 30 percent to over 60 percent within a year, even as he still expects OpenAI and Anthropic to hold more than half the market by dollars, especially for coding. For Higgsfield's own use, post-trained open source models bring over 80 percent margins, while closed source models bring only 20 to 30 percent, since many customers, like social media marketers producing hundreds of ad creatives a week, don't need top-tier intelligence, just cheap, reliable output.
Massive internal AI spending 33:00
Higgsfield spends over 4 million dollars a month internally on AI models across roughly 400 employees, averaging over 10,000 dollars per person. Alex describes one creative team member who spent over 30,000 dollars in a single week experimenting with the Astra model to solve an asset organization workflow problem, working five nights straight; the result wasn't production ready, but the team learned a lot. He admits his finance team likely sees him as too permissive with this kind of spending, even as he tries to keep it under control.
Costs for elite talent rising 35:32
Spending on top creative and engineering talent is likely to keep climbing toward fifty to a hundred thousand dollars a month for people who function as ten times engineers or ten times creatives, though Alex expects their salary demands to rise in step. Other functions, like legal and finance, tend to stabilize around five hundred thousand dollars a month in cost once AI is layered in. He admits Higgsfield made an operational mistake by not ramping up legal and customer support teams fast enough, even though those are areas many expected AI to replace. Today the legal team is over ten people and customer success is over forty, all using AI heavily, and while over sixty percent of first line customer support requests can be handled by AI, B2B support still resists automation. Revolut's ninety two percent AI resolution rate on consumer support is cited as a benchmark, achieved only with heavy investment, and Alex notes that fast product releases, sometimes weekly, make it harder to keep support agents updated since their usefulness depends entirely on current context and rules.
Coding tools shifting fast 38:02
The engineering team moved from Claude to Codex by mid June, with even the ten times creatives following that shift, and Alex describes the pattern as cyclical rather than permanent. Asked whether the pace of model releases will slow within three years, he doubts it, pointing to OpenAI's early move into law-specific AI as just a first version of a coming wave of specialized models built for particular industries. Harry pushes back hard on the idea that a generic AI product can serve big law firms, given how deep and specific legal functionality needs to be and how long law firm sales cycles run. Alex agrees more with narrow, specialized plays, citing their shared investment in Solve Intelligence, a company built around owning a patent workflow's system of records, which he says shows why owning a specific, valuable workflow beats generic tools.
Moats, network effects, and fundraising 41:32
For Higgsfield, the goal is to become the natural home for creative assets, letting marketers search content semantically and check brand adherence in ways tools like Dropbox, Adobe, or Canva were never built for. Alex argues real moats today come from only two places, delivering a concrete outcome or building network effects, since AI hasn't replaced the latter. Higgsfield's open source project count grew from about ten seed projects eight weeks earlier to over ten thousand, which he sees as an early network effect. On fundraising, he singles out Yuri Milner as the standout investor meeting, someone who deeply understood the shift toward AI generated content, short form drama trends moving from Asia to the West, and the coming disruption of the trillion dollar advertising industry. Harry argues Higgsfield is undervalued compared to Silicon Valley peers given its growth rate, comparing it to Cognition's valuation, but Alex insists the company is building for the long term, aiming past a hundred billion dollars, and notes that across public companies outside pharma and big tech, spending on sales and marketing already outpaces R&D, which is the opportunity Higgsfield is chasing. He adds that the West provides over seventy percent of revenue, even though Asia, especially Seoul, shapes many of the trends, and that Asian creators lean into direct to consumer distribution rather than resale platforms.
Family sacrifice and personal drive 50:02
Alex reflects on the personal cost of his career, recalling that his mother worked three jobs when he was young while his father devoted himself to Alex's competitive checkers career, once ranked top three in the world, traveling with him worldwide. His father has had Parkinson's disease since his twenties, and Alex says no amount of capital or business success can undo that kind of sacrifice. He admits he still misses family birthdays and weddings, and traces his drive to a mix of conviction about the technology and market opportunity, deep fear of missing out, and formative memories, including reading about Bill Gates as a six year old and his mother's belief that merit based success was possible through technology. He credits his parents with instilling the importance of a merit based environment, which is part of why moving to California mattered so much to him.
Europe's competitive edge and talent 53:01
Alex points to companies like Legora, ElevenLabs, and Lovable as proof that Europe's application layer is stronger than people give it credit for, even if usage data doesn't get the same belief as Silicon Valley hype. He notes Europe's deep strength in hardware, citing ASML as irreplaceable to the entire industry, and argues the real ceiling for Europe is energy policy. He also sees more startups now emerging outside traditional hubs, and contrasts American job-hopping culture with the loyalty he associates with European and Central Asian talent, comparing it to sports fandom rooted in place rather than winning.
Management style and Kazakhstan talent pool 55:00
Alex says he draws his leadership approach from Jensen Huang, Elon Musk, and Nick, all of whom reject soft corporate feedback styles in favor of being detail-obsessed, which outsiders might call micromanagement. His own rule is simple: hire the best people, empower them, and find ways to retain them, dismissing most other management theory as disconnected from reality. On team makeup, he says roughly fifty are in California, fifty remote, and over three hundred in Kazakhstan, a country he says ranks top five in the world in physics olympiads and has absorbed Singaporean education methods on top of its Soviet math foundation, producing deep technical talent at a 15 percent personal income tax rate.
Family, work hours, and sacrifice 59:00
Alex says he owns no property himself, having spent over a million dollars from an earlier company sale buying homes for his parents and in-laws, which he frames as an Asian cultural obligation to give back. He now works 80 to 90 hours a week on Higgsfield, aiming for just 3 hours with his wife and 5 with his son, occasionally taking a full day off when travel allows, and insists there is no shortcut to hard work, pointing to product leaders he has known who all work relentlessly. He describes his wife's patience with him skipping events at the last minute due to urgent fires, including AI-driven fraud and hacking attempts the company has had to fight.
Quick fire round on predictions 1:04:30
Asked what he changed his mind on, Alex says he once expected HubSpot to become obsolete but now believes its familiarity keeps it sticky, especially for go-to-market hires used to it as a system of record. He predicts most social media content will soon be AI generated, making rare authentic content command far higher ad rates, and suggests creators should still use tools like Higgsfield for video overlays. He expects new creative-director-style jobs to emerge within five years where people direct AI to generate story variations in real time, a role with no name yet. He also names Frank Slootman and Chad Pete as leaders he admires for their direct, no-nonsense style, while citing Snap's fall from an 80 billion dollar valuation as proof that momentum never lasts forever.
Meta's AI Storytelling Advantage 1:10:00
Asked why Meta's market cap has pulled so far ahead of Snap's, Alex points to storytelling: many public companies have failed to articulate a convincing AI narrative, and Snap is one of them. He praises Mark Zuckerberg for succeeding in both private and public markets, and highlights Meta's move to bring in Alex Wang to effectively run Meta's AI efforts as a shrewd, low-cost talent acquisition.
Revenue Growth Projections 1:12:00
With current revenue around a billion, Alex's finance team projects 4.5 billion within a year, expecting some deceleration even while pushing 30 percent month over month growth. Pushed for his own honest estimate, Alex suggests the real number could be over 10 billion, driven by monetization in the creative AI space as direct to consumer brands produce more ads and aspirational cinematic AI content draws in new creators.
Hollywood's Shifting AI Sentiment 1:13:03
Alex describes Hollywood's embrace of AI mainly as a hybrid production tool, a new form of CGI. He notes sentiment has moved from strictly negative toward neutral, though many creative talents remain firmly anti-AI. Still, more people are asking whether AI can help tell stories that budget limits once prevented, a shift he sees as genuinely positive.
Valuation Ambitions and Public Future 1:14:31
If revenue reaches ten billion next year, Alex estimates the company could be valued around eighty billion at a conservative multiple. He says the goal isn't chasing valuation but building a sustainable company fit for public markets, with distribution mattering as much as infrastructure. He believes Higgsfield has potential to surpass companies like Applovin and Shopify, and confirms he wants Higgsfield to eventually go public.
AI-generated summary. It can be wrong or incomplete - check anything that matters against the original.
