Higgsfield hits $1B revenue in 18 months, lands Nvidia investment
AI video is now a factory business, and Nvidia just bought into the production line to secure its future chip demand.
Business · Source: Bloomberg via Bitcomme
What happened
Higgsfield just crossed a $1 billion annualized revenue run rate only 18 months after launch. The San Francisco-based AI video startup announced a Series B extension backed by TCV and Nvidia. This builds on an August round led by DST Global that raised $400 million at a $5.4 billion valuation. The announcement coincided with Nvidia CEO Jensen Huang visiting Kazakhstan, where Higgsfield maintains deep engineering roots and is celebrated as the country's first AI unicorn.
The platform operates as a massive aggregation play for visual media. It lets creators and brands combine about 35 different AI video models alongside Higgsfield's own proprietary engine. Users can select different models for storyboarding, character creation, and final video rendering. This allows enterprise clients to generate up to 100 ad variants from a single campaign concept. The scale of production is staggering. The company processes 12 trillion tokens and 150 million generations every single month.
Enterprise adoption is driving this hypergrowth. Business contracts have grown tenfold since June, and enterprise revenue is growing by more than 100 percent month over month. The company now boasts over 40 million registered users globally, adding 10 million since the summer. They will use the fresh capital to expand their global compute infrastructure, hire top talent, and accelerate product development to capture more advertising budgets.
Key facts
- $1 billion — annualized revenue run rate reached 18 months after launch
- 40 million — registered users across 238 countries
- 12 trillion — tokens processed monthly across 150 million generations
- 35 — different AI video models available on the platform
- 100 — ad variants generated per campaign concept
Why it matters
The era of single-model loyalty is dead for enterprise video production. Builders need to understand that brands do not care about which foundational model generates their content. They care entirely about workflow, localization, and scale. Higgsfield proves that the real money sits at the application layer where multiple models are stitched together to solve specific business problems. Advertising agencies need mass variation for different markets, and they will pay a premium for platforms that abstract away the complexity of managing 35 different AI models.
The second-order effect is a massive shift in the underlying economics of compute. Generative video yields 30 to 150 dollars per million equivalent tokens, compared to mere pennies for text generation. Nvidia knows that video generation will consume unprecedented amounts of compute power in the coming years. By investing directly in the application layer, Nvidia ensures these high-yield token factories run exclusively on their silicon. This guarantees demand for their hardware, from current Hopper systems to the upcoming Vera Rubin architecture.
For builders
Build model-agnostic workflows for enterprises
Enterprises pay for reliable outputs, not specific models. Higgsfield routes workloads across 35 different video models to get the best result for storyboarding or rendering. You lose if you lock your product to a single provider.
Target high-yield token generation
Video tokens yield up to 150 dollars per million, dwarfing text generation margins. Builders who shift focus from text wrappers to visual media generation will capture massive budgets from advertising agencies. Text is cheap, but video pays the bills.
Localize creative assets at scale
Marketing teams are desperate to turn one concept into hundreds of localized variants. If you build tools that automate this variation process, Fortune 500 brands will pay you handsomely. Manual video editing is a bottleneck you can solve.
My take
I always say that infrastructure follows the application. Nvidia investing in Higgsfield proves they know the future of compute demand lies in video generation factories. If you want to build a unicorn today, stop training foundational models from scratch. Start building the assembly line that enterprises actually use to turn tokens into revenue.
Original reporting: Bloomberg via Bitcomme. This is my rewrite and opinion.