AppLovin’s 83% Margin Guide Comes With a Compute Bill

AppLovin guided third-quarter 2026 adjusted EBITDA margin to 83%, and the shares bounced. That print is a step down from 85% in the first quarter and 84% in the second, and management tied it to higher training and inference costs for the Axon ad models.

The stock closed at $323.96 on September 11, 2026, up 6.65% over a month, still down 47.61% in 2026 and sitting against a December 2025 high of $745.61. A popular earnings-bridge model still marks fair value at $625, which is 48% above that close, as if the company were a software compounder whose margins only travel one way.

AppLovin’s 83% Guide Is the First Step Down

AppLovin Corporation, the Palo Alto marketing platform that sells ads through Axon, its AI recommendation system, reported second-quarter results on August 5, 2026. Revenue was $1,924 million, up 53% from $1,259 million a year earlier. Adjusted EBITDA was $1,614 million, up 58% from $1,018 million, and the company printed an 84% adjusted EBITDA margin on that quarter.

Net income was $1,267 million, or $3.76 a share on a diluted basis. Income from operations was $1,494 million, a 78% operating margin, which is the GAAP figure before the add-backs that produce the 84% adjusted line. The third-quarter outlook is the number that put the stock back in motion, and it is not an expansion.

Chief financial officer Matthew Stumpf told investors the company expects third-quarter revenue of $2,055 million to $2,085 million and adjusted EBITDA of $1,710 million to $1,740 million, with adjusted EBITDA margin at 83% at both ends of the range. That is the first guided step down in the disclosed sequence, and it already includes the extra compute for models that went live after the June quarter closed.

THE MARGIN SEQUENCE

An 83% adjusted margin would still be a trophy at almost any other public ad platform. The tell is the direction. Management’s longer-term answer is a low-80% band, not a wider one, and Stumpf said quarterly margin swings may get larger as the models get more complex.

Why the Stock Lost $40 Billion on a Rounding Error

Wall Street had about $1.94 billion in second-quarter revenue. AppLovin came in $16 million light, a gap of 0.8%. Diluted earnings of $3.76 essentially matched the $3.75 consensus. After-hours trading still cut as much as 28.7% off the price from a $417.80 pre-earnings close, wiping out about $40 billion of market value at the session low.

That reaction was a repricing of a premium that had assumed the company would beat every quarter by a wide margin. Co-founder and chief executive Adam Foroughi said the pace of meaningful model improvement was lighter than normal in the June quarter, and that the next step up landed just after the period ended, too late to count. The third-quarter guide uses models already in production and does not bake in further unreleased upgrades.

THE 2026 RESET

1. June 22, 2026: Foroughi opens the self-serve ads platform to any business, ending a 14-year closed system and an eight-month referral phase.
2. June 30, 2026: The second quarter closes with the slowest sequential revenue increase in the company’s restated three-year series, about 4%.
3. August 5, 2026: Results land, the 83% guide is issued, and Stumpf says the SEC has closed its data-practices inquiry with no recommended action.
4. August 12, 2026: Shares touch a 52-week low of $303.17.
5. September 11, 2026: The stock closes at $323.96, a one-month bounce that still leaves the name down 44.34% over a year.

The business did not shrink. Trailing twelve-month revenue through June 30, 2026, was $6,829 million, and trailing net income was $4.41 billion, so the shares trade at about 25 times those earnings on a $108.6 billion market value. What broke was the idea that 50%-plus growth and 85% margins would keep compounding in a straight line.

AppLovin Already Spent Most of a $1.3 Billion Cloud Deal

Stumpf said the main driver of the sequential rise in costs was higher compute for training existing models and building new ones. He also restated the company’s working rule: of each incremental dollar of revenue, the firm spends about $0.10 on compute, and that ratio is what sits inside the third-quarter guide. Cost of revenue, where datacenter and third-party cloud bills land, was $225.801 million in the second quarter.

When additional compute produces substantially more revenue through better model performance, that’s a trade we’ll make every day.

Adam Foroughi, Co-founder and CEO, on the Q2 2026 earnings call

The cash version of that trade is already on the balance sheet. The June 30, 2026 Form 10-Q shows a multi-year contract with a cloud computing provider. In August 2024 the company amended the deal into a $1.3 billion minimum cloud-computing commitment over three years, and by June 30, 2026 it had paid $953.5 million of that, leaving $346.5 million.

Research and development is a different line. It rose 127% to $99.901 million, and $68.946 million of that was stock-based compensation. Total stock-based compensation from continuing operations was $85.783 million, up 148% from $34.552 million. Adjusted EBITDA adds that compensation back, which is why 84% and 78% can sit in the same quarter without either figure being a misprint.

Free cash flow was $863.3 million, only 53% of adjusted EBITDA, after a year-earlier conversion near 75%. Management blamed the timing of international cash tax and interest payments and still expects conversion of about 75% of adjusted EBITDA for the full year. The $0.10 rule is the open question as models get larger, because Stumpf also said the ratio can move over a longer horizon and that the company will say so if it does.

Four Straight Quarters of Slower Growth

Fifty-three percent year-over-year growth at this scale is still rare in public advertising technology. It is also the slowest rate in AppLovin’s own restated quarterly series, and it is the fourth straight deceleration from a 77% peak in the second quarter of 2025, the same period in which the company finished selling its games studios and became a pure ad platform.

THE YEAR-OVER-YEAR SLOWDOWN

• Q2 2025 peak: Revenue grew 77% year over year, the high point in the disclosed run.
• Q3 2025: Growth cooled to 68% on $1,405 million of revenue.
• Q4 2025: Growth was 66% on $1,658 million, with full-year 2025 revenue at $5,481 million.
• Q1 2026: Growth was 59% on $1,842 million, and adjusted EBITDA margin hit 85%.
• Q2 2026: Growth was 53% on $1,924 million, the first sequential rise under 5% in the twelve-quarter series.
• Q3 2026 guide: The range implies 46% to 48% year-over-year growth and 7% to 8% sequentially, a rebound from a light quarter and a slower annual pace.

Foroughi has talked about roughly 30% annual compounding over the coming years, driven by gaming, the consumer vertical, and further Axon upgrades. The third-quarter guide still sits well above that long-run figure. It also sits well below the rates that built the $745.61 high, which is why a 7% to 8% sequential lift can look like a reacceleration and a 46% to 48% year-over-year print can look like a comedown in the same sentence.

Opening the Ads Platform Does Not Create New Inventory

On June 22, 2026, Foroughi opened AppLovin Ads to all advertisers, writing that for 14 years a relationship with the sales team had been required, then a referral code, and that neither was required any longer. Any business can now sign up to reach the company’s billion-plus daily active users, most of them adults in mobile games. Axon stays the name of the recommendation system; the product the customer buys is AppLovin Ads.

The second-quarter reporting period covers that opening, and the written materials gave no advertiser count, no spend-per-account figure, and no split for the new e-commerce cohort. Sequential revenue still rose only about 4%. That is not a shock for a performance-ad funnel, because new accounts rarely spend at scale in a few weeks, but it does mean the open door has not yet shown up as a step-change in the numbers.

The tighter constraint may be supply, not sign-ups. AppLovin’s growth has been coming from better bids and a richer mix, including consumer and e-commerce advertisers paying more for the same game-pause inventory, rather than from a larger pool of ad slots. Mobile gaming is a mature audience. Lightweight titles keep moving into larger entertainment apps, and higher-end games often lean on in-app purchases instead of ads, so the impressions Axon can buy do not automatically grow with the model’s accuracy.

Foroughi’s blog lists three ways to buy: return on ad spend for games and shops, cost per purchaser for subscriptions and similar, and a newer leads product for categories such as insurance and home services. All three still run on that same mobile-game supply. Connected TV is the obvious extra pool, and it is a crowded one. If eCPM gains and mix shifts have to do all the work, the 46% to 48% year-over-year band is the rate that has to be defended, not the 77% peak.

Buybacks Continue After the SEC Closed Its File

The same call that walked through the miss also buried the item shorts had been selling for a year. Stumpf said the SEC had advised the company it concluded its inquiry with no recommended action. The probe had examined claims, pushed by several short reports, that Axon’s targeting leaned on device fingerprinting and identifier bridging in ways that would clash with App Tracking Transparency privacy rules. No enforcement action followed. The company said it had no material loss contingencies for legal proceedings as of June 30, 2026.

The SEC has recently advised us that it concluded its inquiry with no recommended action. We’re pleased to have it resolved.

Matthew Stumpf, Chief Financial Officer, on the Q2 2026 earnings call

That clearance did not re-rate the stock, because it arrived in the same hour as the revenue miss. Short interest as of August 14, 2026, was 11.37 million shares, or 3.90% of the public float, down 4.53% from the prior report, with a 1.4-day cover. The float is not heavily shorted. The multiple still is.

THE JUNE 30 BALANCE SHEET

• Cash: $3,053 million, up from $2,487 million at year-end 2025.
• Long-term debt: $3,515 million, essentially unchanged, or 111% of stockholders’ equity.
• Equity: $3,163 million, lifted by retained earnings after a year-end base of $2,135 million.
• Net debt: $462 million, a modest residual against trailing free cash flow of $4.53 billion.

During the second quarter the company repurchased and withheld 1.1 million Class A shares for $551.3 million, a figure that mixes open-market buys with tax withholdings on vested awards. For the first half it retired 3,275,252 shares for $1.5 billion, and $1.8 billion remained on the authorization at June 30. Combined Class A and Class B shares outstanding were 335.291 million, down from 338.313 million at December 31, 2025.

A $625 Fair Value Still Treats This as Software

The bull math that still circulates takes trailing earnings, compounds them for years, applies a 25 times multiple reserved for high-margin software, and discounts the result back to about $625 a share. At $323.96, that mark is 48% above the last close. It is also the same 25 times the market already pays on today’s $4.41 billion of trailing net income, only with a much larger earnings base assumed in the out years.

That model has little room for a quarter in which a late model drop and a $0.10 compute ratio can move the print by $16 million and the equity value by $40 billion. It also has little room for a cloud contract that has already consumed $953.5 million of a $1.3 billion minimum. The stock can still work from here if the live third-quarter models deliver the 7% to 8% sequential rise and hold the 83% margin. It works a lot less like the software compounder that ran to $745.61.

WHAT WE KNOW

• The guide: Third-quarter revenue of $2,055 million to $2,085 million and adjusted EBITDA of $1,710 million to $1,740 million at an 83% margin, using models already in production.
• The ratio: Management still budgets about $0.10 of compute per incremental revenue dollar and a low-80% adjusted EBITDA band over time.
• The file: The SEC inquiry into advertising data practices ended with no recommended action.

WHAT IS UNCONFIRMED

• The pairing: Whether the extra compute actually buys the sequential sales in the guide, which is the test management set for itself.
• The inventory: Whether e-commerce mix and higher eCPM can offset a mature, possibly shrinking pool of mobile-game ad slots.
• The $0.10: Whether that compute ratio holds as Axon models add parameters, which Stumpf already flagged as a longer-term variable.

Management’s own test is the pairing of that revenue range with $1,710 million to $1,740 million of adjusted EBITDA. If both land, the extra compute bought the sales. If they do not, the 83% margin will look like a ceiling rather than a floor.

Disclaimer: This article is news reporting and analysis for information only and is not a recommendation to buy, sell, or hold AppLovin stock or any other security. It does not take account of any reader’s objectives, risk tolerance, tax position, or time horizon, and it is not investment, tax, or legal advice. Readers should consult a licensed financial adviser or other qualified professional who can review their own situation before making any investment decision. Share prices, margins, guidance, and legal statuses are drawn from company filings, management comments, and market data available for this piece and can change with the next print or the next session.