Burning $4M a month: Inside Higgsfield's dash to a $1B ARR
Key points
Key takeaways from 20VC interview with Higgsfield CEO Alex Mashrabov (September 2026):
$1B ARR milestone. Higgsfield reached a $1 billion annualised revenue run rate on the day of recording, as Bloomberg reported the milestone. It went from $1 million to $1 billion in about 18 months, faster than Cursor, which took 24 months, and faster than any other consumer AI company. Net revenue retention at month 12 exceeded 300%, and one customer grew from a $99-a-month subscription to a $6-million-a-year deal within six months.
Revenue methodology. Higgsfield computes annualised revenue from the trailing four weeks multiplied by 13, taking revenue rather than sales. Annual subscriptions are prorated across 12 months, and only live revenue counts. B2B revenue is slightly over half of total, while mobile consumer usage is under a tenth.
$4M-a-month model bill. Internal spend on AI models exceeds $4 million a month, or over $10,000 per person across the ~400-person team. One frustrated creative engineer alone burned $30,000 in a week on OpenAI's Astra model. Mashrabov expects 10x engineers, and creatives to run closer to $50,000 to $100,000 a month each, while finance, and legal stabilise around $500,000 a month in total.
Model strategy. Higgsfield walked back its early bet on building its own foundation models, deciding that chasing benchmarks was the wrong game. It still trains models for specific customer needs, such as photorealistic shoots, and product consistency, and reports margins above 80% on its own, and open-weight models, versus 20-30% on closed-source ones. It picks which model to use itself in over 40% of cases, a practice it calls tokenomics, while open-source models' share of usage rose from below 30% to over 60% within a year.
Team and distribution. Higgsfield runs no paid advertising, and owns its distribution, backed by an in-house creative team of over 150 professionals, close to half of the workforce. That team produced, and open-sourced its first fully AI-generated movie, needing over 100 hours of generated content for 90 minutes of TV-quality output. Around 300 of its ~400 staff are based in Kazakhstan, where a 15% personal income tax, and a strong maths-and-physics education pipeline help it hire. Mashrabov hopes the company will mint more dollar millionaires there than any other company.
Product-market fit. Higgsfield spent more than a year hunting for product-market fit, burning over $10 million of its $16 million seed, until eight creative directors all made the same point: camera control did not exist in AI. Rebuilt around that insight, the web platform launched on March 31 2025, and product-market fit was immediate.
Asia-driven trends and moats. Mashrabov argues most social-media content will soon be AI-assisted or AI-generated, and authentic content will command 10-15x higher CPM. Short-form drama is now a $10-billion-plus industry led by Chinese companies, and direct-to-consumer e-commerce brands are rebuilding go-to-market around AI-native ad production. The west still generates well over 70% of Higgsfield's revenue,Seoul is its largest city by usage, and the company partnered with streamer Neon to launch an AI-generated stream. He sees only two durable moats: delivering an outcome, and network effects, and AI does not replace network effects. Higgsfield's open-source push grew from 10 seeded projects to over 10,000 in eight weeks, which it hopes will recreate the snowball effect GitHub-style forking gave software.
Read more: 20VC on YouTube · Harry Stebbings on X
