Fal courts a doubling of its $4.5 billion mark in the week OpenAI halved inference prices

A nine-month-old mark, up for doubling

Fal, the generative-media host behind the FLUX image and video models, is exploring a round that would roughly double the $4.5 billion mark Sequoia led in December ¹. A 25 September report↗ puts Fireworks AI on the same path at larger scale ¹. Fal was targeting $300 million to $350 million at roughly $8 billion back in March, and its annual recurring revenue has reportedly doubled since, to around $400 million ¹. Studio video editing is now its fastest-growing segment, sold on usage-based pricing ².

Fireworks ran this trade first

Fireworks raised $250 million at a $4 billion post-money valuation in October 2025 ¹³, then closed the template in July: a $1.505 billion Series D at $17.5 billion↗, priced above the roughly $15 billion it initially sought ³. Annualized revenue passed $1 billion, up fivefold in a year; daily token volume rose from 15 trillion to more than 40 trillion, and 95 percent of those tokens run models customers customized on their own data ³.

The same week, the meter got cheaper

Both businesses sell metered usage: fal bills per generated output on its serverless product, Fireworks bills per token ⁴⁵. The same week as the report, OpenAI made a 50 percent API price cut permanent↗, ninety minutes after Anthropic released Opus 5.5 and one day after xAI undercut the old frontier tier by 80 percent ⁶.

DocuSign showed the customer's version of the same move: more than a million contracts a day routed off frontier models onto small purpose-built ones, cutting AI processing costs 90 percent and per-document cost fiftyfold ⁷. Per-token prices are down about 98 percent since 2022, and enterprise AI bills still rose an estimated 320 percent, because usage grew faster than prices fell ⁷.

Volume has outrun price, so far

The bull case is volume, and it is real. Token counts nearly tripled at Fireworks ³, and its newest product is itself a price cut, routing each task to the cheapest capable model for 50 to 75 percent off ⁵. ElevenLabs holds a $22 billion mark on $600 million of ARR↗ ⁸, and Baseten and Modal Labs are in new funding talks ⁹. If volume keeps compounding faster than unit prices fall, every mark above holds.

The uncovered part is arithmetic: a permanent halving needs double the volume just to stand still, and these marks were set on revenue counted at the old prices. At fal the meter runs per generated image and video ⁴, so that test arrives per output, not per token.

The terms fal prints next are the test

Fal's round has not closed, and Fireworks' did, on revenue that had already lived through a year of falling token prices ¹³. What matters is the valuation fal prints next to the ARR it prints with it, and whether that ARR survives the next 50 percent cut. The valuation is a bet on volume; the volume is counted in a unit price the customers are learning to cut ⁶⁷.