A Billion Dollars With Zero Paid Ads

Alex Mashrabov, co-founder and CEO of Higgsfield, interviewed by Harry Stebbings (20VC)

Summary: Higgsfield went from $1M to $1B in annualized revenue in 18 months after nearly burning out its seed round chasing hype. Alex Mashrabov credits a return to customer conversations, owned distribution with no paid ads, and "tokenomics": routing customer work to cheap post-trained open models at 80%+ margins. He thinks most social media content will be AI-generated, and he's aiming at the sales and marketing budget, which at most public companies is bigger than R&D.

1. The Last-Chance Pivot. Higgsfield spent more than a year searching for a product and burned over $10M of a $16M seed. Mashrabov blames himself: he was optimizing for hype, narrative and how to "hijack the attention" while the product went unbuilt. With roughly $5M left, the team committed to product-led growth and interviewed 8 creative directors, and every one said AI video had no camera control. Higgsfield shipped camera control on March 31 last year and had product-market fit right away.
2. 18 Months to a Billion. Higgsfield crossed $1B in annualized revenue 18 months after hitting $1M, against 24 months for Cursor. Mashrabov thinks only OpenAI and Anthropic got there faster. The math is the last 4 weeks of revenue times 13, with annual contracts prorated to a single 28-day slice. Only live revenue counts, and the company keeps multi-year enterprise deals out of the figure.
3. The $99 Customer. Six months ago one customer paid $99 a month for a subscription, and Higgsfield just signed that customer to a deal worth more than $6M a year. Business net revenue retention at month 12 is over 300%, a number Mashrabov says "never happens in B2B SaaS." Consumers churn hard early, with about 30% gone in month 1, then retention goes flat. Business customers are slightly more than half of revenue, and pure consumer use on mobile is under 10%.
4. The $20 Subscription Trap. Mashrabov bets that Google and OpenAI will "completely demolish" the $20-a-month consumer subscription market with their horizontal products. He calls it a contrarian view, and Stebbings points to Canva's consumer design business as an early casualty. Higgsfield says it can never win at $20 a month. Its job is to show those users enough value that they move past $1,000 a year.
5. Owned Distribution. Higgsfield buys no paid ads, even though people on the team have scaled other companies past $1B and $2B with paid advertising. It employs more than 150 in-house creatives, close to half the company, who make launch videos and tutorials and open-sourced the first fully AI-generated movie. Outsourcing influencer work to an agency went badly and drew controversy. Mashrabov's takeaway is that "distribution now more important than ever," so you have to own it.
6. Benchmark Gaming. Mashrabov calls chasing benchmarks corporate theater. Researchers at big labs tell him colleagues put test data into training to hit numbers and collect bonuses. Video benchmarks mostly test text-to-video, while Higgsfield's open-sourced movie averaged prompts of over 3,000 words and at least 10 image references per scene. He describes video models as a modern rendering engine, comparable to Unreal or Unity with different inputs, and says you can't direct one through a short text prompt.
7. Models on Customer Demand. Higgsfield builds its own models only when customers ask for a specific capability. VFX and camera control took it from about $1M to $20M in ARR in roughly 3 months, and its own image model for photo shoots and product consistency took it from $20M to $100M. Mashrabov says most companies claiming their "own models" are post-training open weights, as Cursor did. The most valuable kind is reinforcement learning on customer decisions, which teaches a model to compress 10 steps into 1.
8. Tokenomics. Higgsfield earns over 80% gross margin when it runs its own or post-trained open-weight models, and 20% to 30% on closed models. The company chooses which model runs in more than 40% of customer jobs, so routing traffic is a core feature. A brand printing hundreds of ad creatives a week doesn't need PhD-level intelligence for a viral social video. Mashrabov cites OpenRouter data showing open-source share rising from below 30% to over 60% this year.
9. $10,000 Per Head. Higgsfield's roughly 400 employees spend more than $4M a month on models, over $10,000 per person. This month one creative spent $30,000 in a week on Astra, vibe coding an asset-organization tool over 5 straight nights. The tool wasn't production-ready, and Mashrabov still calls the experiment net positive. He expects top engineers and creatives to reach $50,000 to $100,000 a month in spend, and to ask for raises to match.
10. Two Moats Left. Mashrabov sees only 2 sources of defensibility today: delivering an outcome, which for Higgsfield means helping businesses sell more through AI ads, and network effects. He doesn't expect AI agents talking to each other to replace network effects within 5 years. Higgsfield's open-source community projects grew from about 10 seeded projects to over 10,000 in 8 weeks, built on the same forking habit that grew GitHub. He also wants to own the asset library as an AI-native system of record, since Adobe and Canva built for a "pixel first era."
11. The Biggest Budget. Excluding pharma and big tech, public companies spend more on sales and marketing than on R&D, a figure Mashrabov had 4 people on his team verify. He believes most content on social media will be AI-generated, while authentic shows like 20VC will earn 10x to 15x higher CPMs. His finance team's plan says $4.5B in revenue by the end of next year. Mashrabov's own number is over $10B, and the team is pushing for at least 30% month-over-month growth.
12. Momentum Expires. Mashrabov's core lesson from Snap is that "momentum doesn't last forever." His team built face filters there that drove most new daily users, when Snap was worth $80B and within 10x of Meta, and Snap is now valued under $15B. He blames a failure to tell an AI story to public markets. Higgsfield is raising and pushing while growth is strong, with the stated aim of becoming bigger than AppLovin and Shopify.

Watch the full video at https://www.youtube.com/watch?v=jszn8rFtxm4. Read the full transcript at https://www.usetranscribe.io/yt/jszn8rFtxm4/higgsfield-ai-costs.