Higgsfield announced a $400 million Series B on Monday morning at a $5.4 billion valuation, led by DST Global. The company’s own release says the round “more than quadruples” the $1.3 billion it was worth at its Series A and extension, and that it reached $700 million in annualised revenue this month.
Those are the numbers the coverage led with. They are not the interesting ones.
Buried a paragraph lower is the question of who is paying. Higgsfield spent its first year as a tool creators used to make social-media video. Founder Alex Mashrabov told the Financial Times that businesses now account for most of the company’s revenue — against less than a quarter in January. That is a customer base turning over inside seven months, and it is the reason Intel Capital, NTT DOCOMO Ventures and Goldman Sachs Alternatives are all on the same cap table.
What “annualised revenue” is not
$5.4 billion divided by $700 million is 7.7. By the standards of private AI rounds this year that is a restrained number — we have written about Cognition and Databricks pricing on far thinner arithmetic.
The restraint is doing work. “Annualised revenue” means a recent period multiplied out to twelve months. It is a run-rate, not a contract book, and for a product a marketing team can put on a corporate card it is the metric most sensitive to a bad month. Higgsfield did not publish net revenue retention, contract length, or the split between self-serve seats and signed enterprise agreements. A company converting from prosumer subscriptions to enterprise deals is exactly the company for whom that distinction decides whether 7.7x is cheap or generous.
The number under the number
The operational figure worth keeping is this: Higgsfield says users of its agentic products — the “Supercomputer” line it rolled out in May — grew 42-fold in three months, driving more than 20 million content generations a month. Thirty million users across 238 countries is a distribution stat. A 42x move on the agentic tier in a single quarter is a product stat, and it lines up neatly with the timing of the revenue-mix flip.
The company also says it “powers visual production for 390 of the Fortune 500.” Read that carefully. It is a logo-penetration figure, and logo penetration counts one designer expensing a seat exactly the same as a signed global agreement. It tells you the product got inside the building. It does not tell you what the building is paying.
Our take: Generative video spent 2025 as a novelty and is being repriced in 2026 as a line item in the advertising production budget — a much larger, much duller, much more durable market. Higgsfield’s round is priced like a company that knows the second story is the real one. The 7.7x multiple is not the market being squeamish about AI video. It is the market pricing the difference between a run-rate and a renewal.
What to watch
- The next disclosure. If Higgsfield publishes net revenue retention or a contracted ARR figure rather than an annualised one, the enterprise transition is real. If the next milestone is another user count, it isn’t.
- Model-cost exposure. Video generation is the most compute-hungry category in the application layer. Gross margin, not revenue, is where this business is decided — and nobody has published one.
- Who else moves. A $5.4 billion mark in AI video sets the comparable for every competitor raising this autumn, the same way OpenRouter’s reported $7 billion exit reset the price of AI infrastructure — from a $1.3 billion valuation, coincidentally the same number Higgsfield just quadrupled.
- Intel and NTT DOCOMO on the cap table. Strategic money in compute and distribution usually arrives with a commercial agreement attached. None was announced. That is the disclosure to wait for.
Higgsfield says the money goes to R&D, global infrastructure, AI hiring and go-to-market. Three of those four are the spending profile of a company chasing enterprise contracts. The fourth is the one that has to keep the 42x going.