AppLovin Said Its AI Gains Wouldn't Slow. The Stock Fell 55% Instead.

A 'virtuous cycle' pitch is now a federal fraud case

Pomerantz filed Talbot v. AppLovin, No. 3:26-cv-10584, in the Northern District of California↗ on 17 September, accusing the company and senior officers of overstating their AI models' strength and hiding delays to a promised generative AI video tool ¹². The complaint's target is the pitch itself: models "constantly improving" inside a "virtuous cycle" of better ads, more advertiser spend, more data ².

The class covers buyers from 12 February to 5 August 2026, and at least four firms, Rosen and Levi & Korsinsky among them, are soliciting the same investors ahead of a 16 November lead plaintiff deadline↗ ²³⁴⁷.

The suit quotes management's own cadence promises

In May 2026, the suit says, management had "100% seen faster improvements to the models" and didn't "really see a reason why that's going to slow down" ⁴.

Two selloffs erased a third of the stock in a month

The first crack came from outside. On 13 July, a Bank of America note found AppLovin's e-commerce footprint "expanded at a slower pace in June" and called the 22 June self-service launch a "muted" start; the stock fell 12.65% to $442.85 ².

The second landed on 5 August. Revenue of $1.92 billion missed $1.94 billion consensus↗, the "pace of meaningful model improvement" was "lighter than normal", and the video tool was "still work in progress". Shares fell 19.66% the next day, to $335.67, and RBC, Piper Sandler, Wells Fargo and Needham cut in one session ²⁴⁵.

The stock still sits roughly 55% below its $745.61 high↗ ⁶. The complaint also counts the executives' own exits: 260,065 shares sold by the CEO and CFO for more than $109.1 million during the class period, through preset 10b5-1 plans rather than discretionary calls on the price ⁶⁷.

AppLovin's defense: model lifts are R&D, not a calendar

AppLovin's defense was delivered on the earnings call, before any suit existed. "There's no guarantee that we're always going to have lifts in every single period of three months," CEO Adam Foroughi said. "There are going to be periods where we don't get material lifts" ⁵.

Demand held: publisher earnings grew double digits quarter-over-quarter, consumer ad spend set a record 28% above the Q4 2025 seasonal peak, and the SEC closed its inquiry with no recommended action ⁵. Growth has slowed four straight quarters, yet the Q3 guide of $2.055–$2.085 billion still means 46–48% growth, on model improvements already live ⁵⁹.

A plateau isn't fraud; pitching a plateau-proof engine may be

AppLovin still trades near 15 times trailing revenue↗ ⁸, a price that assumes the lifts keep coming. The framing a court will weigh is Hagens Berman's: software may legally hit a wall, but a public pitch of perpetual compounding may not ⁷. A model plateau is not a crime; selling a plateau-proof engine is the allegation.

The SEC's exit makes the class bar the active enforcer. If the framing holds, class-period buyers are the receiving side, AppLovin and its directors-and-officers insurers the paying side, and the risk reprices every AI premium sold on perpetual model improvement ⁵⁷. The market rules first, on Q3 results due early November against that guide ⁶. The courtroom's ledger opens 16 November ³.