In January 2016, Machine Zone was priced at a $5 billion valuation. It had raised close to $600 million. Its flagship game was doing more revenue per player than almost anything on mobile.
Four years later, AppLovin bought the entire company. The SEC filing puts the price at $328.6 million, of which $287.1 million was cash paid to Machine Zone’s lenders. Preferred stockholders and lenders split warrants worth $38.2 million. The rest was a settled receivable.
Read that allocation again. The largest amount of money in the transaction went to the people the company owed, not the people who owned it.
Nothing about Machine Zone’s design was careless. The games worked. The marketing worked. The math worked, for about three years. Then the inputs moved a few dollars in the wrong direction and there was nothing underneath.
Deconstructor of Fun took Machine Zone apart in 2017, across three posts on the history, the design, and the monetization, while the company was still at the top. This piece picks up where that ends: what the money did after the model stopped working, and why the same shape of exposure sits in most studios’ ad stacks today.
The company that learned to buy players
Machine Zone started as Addmired, a Y Combinator company from 2008 that made a social network plugin, then mafia-themed mobile games. It renamed itself in 2012 and put 80 people on a new title for 19 months. That title was Game of War: Fire Age, and it shipped in 2013.
The game did fine. What changed everything was the marketing.
By mid-2014 the Wall Street Journal reported the company expected more than $600 million in revenue that year, against a valuation being negotiated at $3 billion or more. Then Machine Zone hired Kate Upton and put $40 million behind a four-month campaign, including a Super Bowl spot. In the hours after that ad ran, Game of War passed Clash of Clans in the App Store rankings.
It was not a one-off. In 2015 the company spent $92.7 million on US television alone, a 272% increase over the prior year. Mobile Strike launched in July 2015 and got its own celebrity treatment with Arnold Schwarzenegger, including Super Bowl spots. Two titles from one publisher sat in the top handful of grossing games worldwide.
Most coverage at the time treated the ad spend as excess. It was the product.
Why paying $60 for an install made sense
Here is the part that gets skipped, and the part that matters if you run monetization.
Machine Zone was not buying installs at a normal price. Deconstructor of Fun reported bids of $60 per user were not unusual, and that the company was rumoured to have bought effectively all available YouTube inventory when Mobile Strike launched, to force it into the top ten grossing within days.
That bid only looks insane if you do not know what a Game of War player was worth.
In 2015, the research firm Slice Intelligence measured the average Game of War player at $550 spent, roughly six times the mobile game average that year. The 4X design was built for that outcome: alliances that create social obligation, permanent troop losses that create catch-up spending, build timers that create impatience, and an offer engine that keeps moving a player up to the next spending tier the way a casino floor does. The genre has the deepest spending ceiling in mobile because there is no level cap on how much advantage money can buy.
So the trade was clean. Buy an install for something between $10 and $60. Extract a few hundred dollars over a lifetime measured in years, concentrated in a small share of players. Pocket the spread. Recycle it into more installs.
Two things made the spread defensible rather than lucky. Machine Zone ran far more ad creatives than anyone else in the category and optimized them relentlessly, which held CPI down while volume went up. And it built serious real-time infrastructure to identify and price high-value users, the same technology it would later try to spin out as a separate business.
This is an arbitrage, and it is a legitimate one. Plenty of good businesses are arbitrages. The problem with an arbitrage is structural: your margin is a difference between two numbers you do not control, and you have no product moat underneath it. You are solvent as long as the gap stays open.
What closed the spread
Both sides moved at once.
On the buy side, everyone copied the game. Dozens of 4X clones shipped, many with better visuals, and they were all bidding for the same narrow audience of people who will spend hundreds of dollars on a strategy game. Lilith’s Rise of Kingdoms arrived in 2018 with smoother onboarding and RPG hooks and went on to more than $3.5 billion lifetime. Chinese publishers with lower cost bases and better creative pipelines took the genre. PocketGamer.biz traces the whole evolution from Game of War forward, and Machine Zone is the first generation in that story, not the current one.
On the sell side, the whale pool was finite and aging. A player who has spent $2,000 over three years does not double it in year four. Retention in the existing titles eroded while the cost of replacing that player kept climbing.
The revenue line shows exactly when the two curves crossed. Game of War peaked in November 2015 at an estimated $69 million in a single month net of store fees, per AppMagic. Mobile Strike peaked around $60 million a month in 2016. Both dropped off sharply in 2017 and never recovered.
Then the company took its eye off the thing that was still working. In early 2017 it launched Satori, repositioning the in-game real-time messaging stack as a data platform for smart cities and IoT. In March 2018 it announced a blockchain partnership. In June 2018 the board replaced founder Gabe Leydon with COO Kristen Dumont, who told Fast Company the pivot had cost the company its focus and that on television, “TV’s kind of a rip-off, to be honest.”
That same month, Machine Zone laid off roughly 125 people from marketing and user acquisition, including the entire internal UA team and everyone at Cognant, its UA-for-hire arm.
The company whose growth engine was a buying machine dismantled the buying machine to reach profitability. Which was the right call by then. It was also the confirmation that the spread was gone.
The sale
AppLovin acquired Machine Zone on May 19, 2020. Terms were not announced publicly at the time, and trade coverage reported an undisclosed sum. The number surfaced later in AppLovin’s pre-IPO filings: $328.6 million aggregate, structured as $287.1 million cash to lenders, $38.2 million in warrants to lenders and preferred stockholders, and $3.3 million of settled receivable. AppLovin also moved its offices into Machine Zone’s Palo Alto headquarters.
Against $593 million raised and a $5 billion mark four years earlier, that is not a sale. It is a debt resolution with a logo attached.
For AppLovin the logic was different and consistent with everything else it bought in that period. As CEO Adam Foroughi put it when the studios were later sold on, “we began acquiring gaming studios to help train our earliest machine learning models.” Machine Zone brought a large, sophisticated UA operation and years of spending data on exactly the kind of high-value user an ad model wants to predict.
What is left
Final Fantasy XV: A New Empire, a title that itself passed a billion dollars, went offline at the end of 2024. Its successor, War for Eos, shut down in October 2024. In that same month AppLovin cut over 120 staff across itself and Machine Zone. In early 2025 a California WARN filing covered 97 more, including Machine Zone’s CEO, CTO and design director. A soft-launched new title, Reign of Vampires, was cancelled.
AppMagic estimates Game of War made $1.4 million across the first four months of 2025, and Mobile Strike $2.8 million. Compare that to a single month in late 2015.
On June 30, 2025, AppLovin closed the sale of all ten of its game studios to Tripledot for $400 million in cash plus roughly 20% of Tripledot, a deal valued around $800 million. Machine Zone went with them. Tripledot’s CEO confirmed the studio still exists and has no capacity to build anything new, and that the plan is to keep the legacy titles alive for the players who remain.
The version of this running in your stack
The easy reading is that Machine Zone got greedy with TV money. That reading is comfortable and it will not protect anyone.
Here is the harder one. Machine Zone’s margin was a spread between what it paid for attention and what it earned from attention. Your studio runs the same spread. You buy attention through UA and you sell attention through your ad stack, and the difference is your business.
Now look at how each side of that spread is managed.
The buy side is instrumented to the minute. Bids adjust automatically. ROAS is checked daily at minimum, often hourly. Nobody at any studio would accept a UA system that repriced once a day.
The sell side gets reviewed every few hours by a person who also has four other jobs. Floors were set at some point and mostly left there. An eCPM decay in one geo overnight gets noticed the next morning, if the dashboard is opened before the standup.
Machine Zone’s spread closed on the buy side, loudly, in public, over about three years, and it still killed the company. A spread that closes on the sell side does not announce itself. There is no headline when your floors go stale, no Super Bowl ad to look back on. Revenue just runs a few percent under where it should, every day, forever.
The exposure is the same shape. Only the speed of detection is different.
What to actually check this quarter
Ordered by how fast each one pays back.
- Write down both review cadences on one line. How often UA bids get adjusted, and how often ad yield gets reviewed. If the first is measured in minutes and the second in hours or days, you have found the asymmetry. You do not need a tool to see it, and the number will surprise the people who approve budgets.
- Plot revenue by hour of day against your team’s working hours. Every studio with a global player base has a window where the users are awake and the operators are not. Find yours, then look at what happens to eCPM inside it. The overnight gap is where the loss concentrates.
- Date-stamp every floor price in your setup. Not what they are, when they were last changed. Anything untouched for a quarter is priced for a market that no longer exists.
- Run a sensitivity, not a point estimate, on your unit economics. You know your CPI and your blended LTV including ad revenue. Calculate how much CPI increase, or how much ARPDAU decline, closes the gap entirely. That single number tells you how much cushion you actually have. CPI is still rising across mobile, so treat it as a moving input.
- Measure your concentration on both sides. What share of revenue comes from your top 1% of payers, and what share of ad revenue comes from your top network and top geo. Machine Zone’s revenue concentration was its strength until the concentrated group stopped growing. Concentration is leverage in both directions.
FAQ
How much did AppLovin actually pay for Machine Zone? $328.6 million, disclosed in AppLovin’s SEC filing rather than at announcement. The structure matters more than the headline: $287.1 million went to Machine Zone’s lenders as cash, with $38.2 million in warrants to lenders and preferred stockholders.
Did the Kate Upton campaign work? By its own measure, yes. Game of War overtook Clash of Clans in App Store rankings after the Super Bowl spot, and the game was collecting over $1 million a day by early 2015. The campaign was not the mistake. Building a company whose only growth engine was that kind of spend was the exposure.
Why did Game of War revenue collapse so fast? Both sides of the trade moved. Competing 4X titles bid up the same high-value audience while the existing paying base aged and churned. AppMagic estimates show the peak at $69 million in November 2015 followed by a sharp fall through 2017.
Are Game of War and Mobile Strike still running? Yes, under Tripledot, operated by a small team with no mandate to build anything new. Tripledot’s CEO said the focus is maintaining the legacy titles for the players still there.
Is a UA-driven growth model always an arbitrage? No. The distinction is whether anything survives the spread closing. A studio with organic acquisition, a live-ops calendar that grows spend per existing player, or a portfolio that shares audience has other engines. A studio whose entire growth is bought installs has one number, and that number is set by an auction it does not control.
What is the ad ops equivalent of the spread closing? Slow eCPM decay against a fixed floor. It produces no incident, no alert and no meeting. Revenue that drops and appears to recover a few hours later is the visible version of the same problem, and most stacks only catch the visible version.
Find out how fast your sell side actually reacts
Machine Zone knew the value of a user to the dollar and repriced its bids constantly. Almost no studio applies that standard to the other half of the spread.
If you want to know what your ad stack is leaving in the gap between a drop and a reaction, run it on one app for ten days. Your data, your numbers, no commitment. Talk to the UndrAds team and we will set it up against your existing Google Ad Manager setup, with every action visible in real time.
