What actually happened
On August 17, 2026, Higgsfield, a text-to-video platform built for marketing teams, raised a 400 million dollar Series B at a 5.4 billion dollar valuation, with Goldman Sachs and Intel joining the round. The valuation grabbed the headlines. The number that matters sits underneath it. The company says annualized revenue jumped from roughly 20 million dollars a year ago to about 700 million dollars, a 35x climb in twelve months, serving more than 30 million users (TechCrunch, PR Newswire).
Now set that against the other AI video story of the season. OpenAI discontinued the consumer Sora app on April 26, 2026, after it reportedly earned about 2.1 million dollars in lifetime revenue, and it switches off the Sora API on September 24 with no named replacement (OpenAI). One AI video product is being valued like infrastructure. Another just died at two million in lifetime revenue. The gap between them is not model quality. It is what each one was actually being paid to do.
Why the revenue mix is the whole story
The easy read is "AI video is winning, so put your brand on it." That is the expensive misread, and the 5.4 billion dollar number is the reason it is tempting.
Follow the money and you find the same customer wherever the revenue is real: performance and social advertising. Platforms in this tier monetize volume and velocity, thousands of short, disposable clips that live for 48 hours inside an ad account and get judged on cost per result. That is a large, legitimate business. It is also the precise opposite of what B2B trust video is for.
Your brand and pipeline content has a different job. It has to make a named human credible to a skeptical buyer across a months-long cycle. It gets watched by a champion building an internal case, by procurement checking whether you are real, by a CFO deciding whether to sign. Disposable is a feature in an ad account and a liability on a category page. The market just put a 5.4 billion dollar price on the disposable use case. Read that as a definition of what generative video is economically good at, not as permission to route your credibility through it.
The data
Start with ours. In our retainer book, across more than 20 active B2B engagements, the assets clients rate highest for pipeline influence (customer interviews, founder point of view, product proof) run north of 90 percent real footage. The generative clips we do use are almost all disposable social hooks: the exact category the market just valued at 5.4 billion dollars. In a typical monthly deliverable, generative AI footage accounts for under 8 percent of finished runtime, and it never carries the trust moment.
Our intake audit says the same thing from the buyer side. Of the last dozen prospects who arrived asking for a specific AI model by name, nine could not name a single business outcome the model was supposed to produce. They were buying the headline, not a result. That is what a 35x revenue year does to a category: it makes the tool feel like the strategy.
The third-party record lines up cleanly. Higgsfield's raise and revenue are documented in its funding announcement and press coverage (TechCrunch). The Sora shutdown, and its 2.1 million dollar lifetime revenue, are confirmed by OpenAI's own discontinuation notice (OpenAI). And the EU AI Act's Article 50 transparency rules, in force since August 2, 2026, now require deployers to disclose AI-generated or manipulated video at the point a viewer encounters it (European Commission). Read together, they draw a hard line between two kinds of video: the disposable kind the market pays for, and the trust kind that now has to be labeled the moment it is synthetic.
The counter-argument, steelmanned
Here is the strongest case against me. Higgsfield did not reach 700 million dollars in annualized revenue by selling a toy. Generation is improving faster than any other part of the stack, it is obviously monetizable, and a B2B team that refuses to touch it will get outshipped on volume and cost by competitors producing ten times the output. If the models are this good and this cheap, why would a serious marketer keep them at arm's length from the brand?
That case is right about the tool and wrong about the bet. Generative video is genuinely useful, and we use it every week. The error is confusing a useful input with a durable foundation. The revenue that just got priced at 5.4 billion dollars comes overwhelmingly from advertising, where a clip is meant to be forgotten. B2B trust video is meant to be remembered and, increasingly, to be labeled when it is synthetic. You can absolutely use models to move faster inside a system that owns the trust layer. What you cannot do is let the model be the trust layer, because the market has already told you it prices that model as disposable.
What to do Monday
Split your video into two buckets before you spend another dollar. One is disposable: ad hooks, social cutdowns, thumbnail tests, things designed to live for 48 hours. The other is durable: brand, founder point of view, customer proof, the assets a buyer studies before signing. Use generative video freely in the first bucket and sparingly, and disclosed, in the second.
Read the receipt on any AI video tool you are evaluating. Ask what its largest customers actually use it for. If the honest answer is ad volume, do not expect it to build brand, no matter what the valuation implies.
Fund your real-footage capture cadence. Interviews, product capture, and event coverage are the one input no valuation can commoditize, because they require your actual people and your actual product. That library is the asset. The model is a filter you run it through.
Write a one-line synthetic-media standard and tie it to Article 50. Decide what you will and will not fake on camera, then match the disclosure rule already in force. Trust is the product you are selling. Put the boundary in writing before a cheap tool tempts you to spend it.
Budget by system, not by tool. A 5.4 billion dollar model is still an input, not a strategy. The teams that win the next year are not the ones with the best generator. They are the ones with the standards, the cadence, and the owned library that make any generator interchangeable.