Higgsfield has about 400 employees, monthly model costs exceed $4 million, over $10,000 per person, with some employees burning about $30,000 a week
Harry Stebbings shared the perspective of Higgsfield CEO Alex Mashrabov: annualized revenue is calculated by taking the last 28 days and multiplying by 13; the key point is that revenue is not sales, annual subscriptions and enterprise contracts are amortized over 12 months, only counting active revenue and not lumping three-year enterprise contracts into the billion-dollar figure.
The company claims it took about 18 months to grow from $1 million annualized to $1 billion, reportedly faster than Cursor's approximately 24 months, ranking behind OpenAI and Anthropic among AI-native companies. The projected path is approximately $200 million by the end of 2025, about $500 million by June 2026, around $700 million by August, and completing a $400 million Series B round with a valuation of about $5.4 billion, led by DST Global, with participation from Goldman Sachs Alternative and Intel Capital; 20VC mentioned subsequent valuation rumors could reach about $8 billion. In June, the Chief Strategy Officer stated that cash flow had turned positive. The company has seen a month-on-month growth of about 115% since June, with net revenue retention around 300%, a first-month churn of about 30%, and over 300% net retention by the 12th month. One customer expanded from $99 per month to about $6 million annually. Business revenue is slightly over half, pure consumer mobile is below 10%, and Western revenue exceeds 70%. They claim to cover about 78% of the Fortune 500 and over 32 million creators.
There are about 400 employees, with monthly model costs exceeding $4 million, averaging over $10,000 per person, and some employees burning about $30,000 a week. The gross margin for open-source weights is claimed to exceed 80%, while closed-source is about 20% to 30%, with over 40% of scenarios using the company's selected models. The content team has expanded to about 150 people. Mashrabov, who ranked among the top three in competitive programming in his early years, sold his previous company for about $166 million to Snap, after burning about $10 million and then pivoting. He claims the moat is results and network, not model ownership, and boldly stated that they aim to exceed $10 billion in revenue within 12 months.
Multiplying by 13 means treating the last four weeks as a steady state for the year. If usage fluctuates in a week, the annualized figure will swing significantly; the annual figure only includes the current month's slice, avoiding counting unfulfilled contracts as realized scale. Stebbings posed the question to several financial and venture capital commentators, explaining that the controversy over the billion-dollar headline lies in the denominator, not in the adjectives.
In market mechanisms, this is a valuation narrative battle. Buyers want to purchase growth funds for the fastest video application layers; sellers use verifiable active revenue slices to counter "counting three-year contracts as one-time revenue." Funds have already entered computing power and content teams from DST and others. Beneficiaries are companies that can repeat nearly four weeks of billing week after week; those under pressure are peers who report annualized figures based on total contract amounts but are contradicted by this methodology. Once the method is public, peers must explain whether they have included unfulfilled sales in their calculations.
Supplementary data: In a June SaaStr interview, he previously used "recent four weeks of usage plus monthly subscriptions plus annual subscriptions divided by 12, then multiplied by 12." The company's advertising videos are referred to as a snowball: once there is a positive return, they double down. Gross margins turning positive will start in early 2026, indicating that previously each video may have incurred a loss on computing power.
Source: Public information
ABAB AI Insight
The billion-dollar debate is not about the camera model, but about accounting multiplication. Multiplying by 13 freezes the high-growth range into an annual monument, while amortizing the year prevents unfulfilled contracts from being stuffed into the monument. Mashrabov applies this method to his lab peers, indicating: we use industry algorithms, not unconventional methods. The growth from $99 to $6 million and a retention rate of 300% demonstrates that there is expansion beneath the multiplication, not just coincidentally high figures over four weeks. The lesson from Snap is that momentum will fade, so the numbers must be recalculated weekly.
The capital path is to first open up a $5.4 billion valuation with sliced annualization, then fill the slices with corporate advertising budgets. An 80% gross margin on open-source weights subsidizes the 20% to 30% of closed-source shelves, and the company selects models in over 40% of scenarios, treating routing rights as a profit switch. The $4 million monthly model bill is a tax, and the 150-person content team is for distribution. DST and Goldman Sachs are buying video workflows, not equity in foundational models. If the four-week figures decline, multiplying by 13 will similarly and brutally revise downwards.
Comparable objects are annualized revenue in consumer internet and recurring revenue in SaaS: the former amplifies peaks, while the latter requires renewals. Higgsfield annualizes on-demand usage in a SaaS manner, but lacks the contract rigidity of SaaS. The industry stage is that the application layer uses laboratory-style annualization language, competing for pricing power that is "not just a shell." The moat is described as results and network, which acknowledges that models can be swapped at any time, but the billing cannot.
Structural judgment pertains to the transfer of pricing power. The mechanism is: whoever defines revenue, defines the myth of speed. The 28-day window shifts pricing power from audited annual reports to multiplication that can be recounted on podcasts. If limited partners accept this denominator, valuations will follow the four-week billing; if not, the billion dollars is just a headline. Transparent multiplication is to ensure that the next round does not need to explain where the numbers come from.
ABAB News · Law of Cognition
Annualization is multiplication, the denominator is more important than adjectives
Active revenue can be multiplied, total contract amounts can only be amortized
If the four weeks change, the billion dollars will also change