20VC Host Harry Stebbings: Higgsfield's Content Machine Outpaces Cursor
20VC host Harry Stebbings pointed out that AI video company Higgsfield achieved an annual recurring revenue of $1 billion in just 18 months, faster than the pace at which code editor Cursor reached the same milestone. He also noted another statistic: the content production side alone has over 150 people, with each product launch and funding announcement pushed out by the same content machine.
The company's stated $1 billion is an annualized run rate, not revenue already booked in the past 12 months. Bloomberg quoted CEO Alex Mashrabov's algorithm: take the revenue from the last four weeks and multiply by 13. The curve itself is: the browser product launches in March 2025, with an annualized revenue of about $200 million by the end of that year, about $500 million by June 2026, about $700 million by August, and in September the company told the media it had surpassed $1 billion annualized. Corporate contract revenue expanded about tenfold since June, with net revenue retention around 300%, and corporate adoption at about 115%, with a sales team of less than 20 people.
The personnel structure treats "content" as a capacity rather than a subsidiary of marketing. In a SaaStr discussion, the company had about 150 total staff, with about 60 in engineering and product, and the rest mostly in creative roles that have worked on commercials, rather than prompt operators. Stebbings later forecasted in the same episode that the content side alone exceeds 150 people; the company spends about $4 million monthly on different models, averaging over $10,000 per person; Mashrabov expects top employees to spend $50,000 to $100,000 monthly on tokens, with engineers burning $30,000 a week on atmosphere programming at Astra.
The revenue structure has shifted from creators to corporate advertising. Commercial marketing accounts for about 70% of platform activity, with clients including about 390 Fortune 500 companies, around 30 million users, covering over 200 countries and regions. Gross margins turned positive in early 2026, with positive cash flow reported in May. The product line is expanding into advertising studios, film studios, and agency-style multi-camera pipelines, with a single advertising concept being broken down into hundreds of channel variants. On the competitor side, Runway, Pika, and Luma still focus on consumer short films, while OpenAI's Sora has contracted its commercialization path, leading advertising budgets to flow more towards tools that can produce films, revise them, and enter corporate procurement.
Funding has locked this curve into valuation. In April 2024, Menlo led an $8 million seed round, in September 2025, GFT led a $50 million Series A round with a valuation of about $1 billion, in January 2026, Accel participated in a Series A extension of $80 million raising the valuation to about $1.3 billion, and in August, DST Global led a $400 million Series B round, with follow-on investments from Goldman Sachs Alternatives, Intel Capital, Liberty Global, etc., raising the valuation to $5.4 billion, quadrupling in eight months. The funds are clearly used to lock in computing power, enhance corporate security, and drive upselling. Cursor serves as a comparison: the product was made public in 2023, reaching about $100 million annualized by January 2025, about $500 million by June, surpassing $1 billion by November, and then continuing to surge to tens of billions in annualized revenue before being acquired by SpaceX for about $60 billion in stock; Stebbings compares the "months taken to reach $1 billion annualized," not the final scale.
In market mechanics, buyers are direct-to-consumer brands and agencies that need to produce ads in bulk and deploy multiple versions, while sellers like Higgsfield package generation counts, seats, and agency workflows for sale; the event-driven aspect comes from advertisers rewriting shooting budgets into computing budgets, and model vendors selling interfaces to the application layer. Funds flow from DST, Goldman, and Accel to secure GPUs and content production, then back from Fortune 500 marketing budgets to the platform. Beneficiaries are those who can write "production speed" into retention at the application layer and model vendors selling tokens into high-frequency workflows; under pressure are traditional production companies that charge by the day and peers that only generate demo-level videos without corporate procurement interfaces.
Source: Public Information
ABAB AI Insight
Alex Mashrabov's path is not the first time generative capabilities have been turned into a sellable product. He first sold AI Factory to Snap for $166 million, managing generative AI within Snap, and then founded Higgsfield in 2023, launching the browser product in March 2025. The former company proved that "models can enter consumer lenses," while the latter proves that "lenses can enter advertising procurement orders." The same team embeds creative roles directly into the product cycle: tutorials and sample clips are generated using their own platform, and creatives inform engineers where the model fails, thus turning the release rhythm into a customer acquisition rhythm.
Capital flows through the application layer rather than retraining a foundational large model. The total from seed to Series B is about $550 million, with valuation jumping from about $1 billion to $5.4 billion, and the buyer list shifting from Menlo and Accel to DST and Goldman for growth equity. The money primarily locks in scarce computing power and shifts revenue from individual creators to corporate subscriptions: in January 2026, corporate revenue accounted for less than a quarter, but by Series B, "most revenue comes from corporations." The strategy is narrow—avoiding competition with OpenAI for general video myths, instead targeting the existing shooting and revision budgets of marketing departments, turning one shoot into a hundred channel materials.
Comparative objects are Cursor and Lovable, not traditional film industries. Cursor established the editor as an enterprise standard through bottom-up developer subscriptions, later being folded into SpaceXAI; Lovable created a high-speed annualized curve with natural language generation software. Higgsfield aims to be the "Cursor of video": lowering the barrier for non-technical users and moving production studios into the browser through agency workflows. The industry phase has shifted from demo generation to controlling advertising inventory: whoever can enter the creative systems of 390 Fortune 500 companies transforms from a tool into a budget item. Runway still leans towards world models and short film aesthetics, while Higgsfield focuses on billable marketing pipelines.
Structural judgment involves technological replacement combined with a shift in pricing power. What is being replaced are daily billed shooting, editing, and multi-version revisions; pricing power shifts from production companies and media buying agencies to software companies that can charge by generation counts and seats, embedding content teams into their structure. The mechanism is: advertisers primarily want not "like a movie," but "the same concept instantly turned into a hundred deployable materials"; once this demand is captured by software, production hours are no longer a bottleneck, and computing power and content production become the constraints. The next generation of companies will write content into DNA, not just hire more PR, but turn publishing itself into product testing and sales funnels.
ABAB News · Cognitive Laws
The fastest-growing products often come with a media factory.
Computing bills determine who can publish daily, and publishing determines who can charge daily.
In the next round of company structures, content roles will no longer be cost centers.