Selling Game Code and Game Data to AI Companies

Yes. AI companies license game studios' archives, and two parts of an archive matter most to them: the source code with its full history, and gameplay recorded together with the player's inputs. Mercor, which licenses companies' records to AI labs, publishes a listing for game studios that puts what a studio earns at $100,000 to $1 million per company. Worldmodeldata, a firm that supplies gameplay to developers of world models (AI that learns to predict how a scene moves and responds), says it has licensed almost 1 million hours of data from studios behind popular games, as WIRED reported. Code on its own is priced like any private codebase: the middle half of public prices per repository runs $1,352 to $5,869 per repository. What a studio cannot license is what it did not make. Unreal Engine's licence bars using the engine as a training input to any generative AI program, Unity's asset store bars training on bought assets without the express consent of the provider, and Unity's own terms require Unity's prior authorization before anything made with its software trains a model.

This guide is for founders, studio heads, finance leads and counsel of game studios: independent and mid-sized developers with years of shipped and unshipped work, studios that are closing or have closed, and the publishers and holding companies that own studio archives. It covers who buys game data and why, what each part of a studio's archive is worth to which buyer, the prices on record, what world-model developers want from gameplay, the parts of an archive a studio does not own (the engine, store assets, middleware, music, performances, publishers' and contractors' rights, and players' data), how to prepare an archive, what changes when a studio is closing, and a worked example. It sits beside our guides to selling a codebase and selling video, which go deeper on code and footage in general.

AIData.Broker also acts as the agent of holders who license or sell their data, so we are a party in this market. Check any figure you plan to rely on against its source.

This guide is research from public sources, not legal, tax or investment advice.

Contents

1. Who buys game data, and why
2. What a studio's archive holds, and who wants each part
3. What studios are offered: the prices on record
4. Gameplay with inputs: what world-model developers want
5. Game code: what code buyers check
6. What a studio does not own
7. How to prepare a studio archive
8. When the studio is closing
9. Worked example: a mid-sized studio
10. How a sale through us works
11. Questions people ask
12. What this means for you

1. Who buys game data, and why

A game is one of the few places where an AI developer can get pictures and the actions that caused them, recorded together. Every frame of a game is the result of a player pressing a button, moving a stick or clicking a mouse, and the game knows exactly which input produced which change on screen. Video taken from the open web has no such record: a model watching it can see that a car turned, but not what the driver did. That pairing of screen and input is what the newest kinds of model need, and it is why studios, whose archives hold years of it, now have something to license in a market that until recently bought mostly text.

The buyers fall into five groups, and they want different parts of the same archive. Between them and the studios, a few firms gather gameplay from many studios and license it on (section 3).

World-model developers. A world model learns to predict the next moment of a scene from the current one and an action, so that it can generate interactive worlds or train robots and agents in simulation. Its developers consume video at a scale that dwarfs other buyers: NVIDIA pretrained its Cosmos world models on data that included 20 million hours of driving, robotics, synthetic and related data. Gameplay recorded with inputs is the cheapest way to get such footage with the actions attached. There is no ready supply to buy instead: WIRED reported in September 2026 that, unlike language models trained on oceans of text, world model labs have no similar corpus to draw on.

The developers of world models have gone to games for it. Microsoft trained its Muse model on more than 1 billion images and controller actions, over 7 years of continuous human gameplay from one Xbox game, gathered over years of collection with the studio that made it. Google DeepMind collaborated with eight game studios to train and test its game-playing agent. General Intuition, spun out of Medal, a platform where players upload billions of gameplay clips every year, raised $133.7 million in its first round and $220 million more in September 2026 (both as reported), after OpenAI had reportedly tried to buy Medal for $500 million. Public footage is plentiful: Tencent trained a game video model on over one million gameplay recordings across over 100 AAA games. What public footage lacks is what a studio holds: the inputs, the engine's own state and footage of games no player could record.

Coding-model developers and the vendors that supply them. Game code is large, long-lived and unusual: engines and gameplay systems in C++ and C#, tools, build scripts and years of fixes, much of it never published. Buyers of private code turn it into training tasks for coding models (our guide to RL environments explains how). Mercor asks companies for GitHub repositories, commits, PRs, and issues, and its game studio listing is one of the few that names a whole industry.

Builders of environments from games. Some firms turn games themselves into places where models practise: Good Start Labs, for one, builds interactive learning environments from curated games with unique, verifiable rewards, and it raised $3.6 million to do so. A studio's game, with its rules and win conditions, is the raw material.

Buyers of 3D models, animation and sound. Models that generate 3D assets, motion and audio need original source files. The largest stock libraries already license such material for training: Shutterstock's data licensing offer includes 1 million or more 3D models beside its images and video, and Wirestock, which sources images, video, design assets and 3D content from its artists, reports $40 million in annual run-rate revenue. A studio's own models, scenes, animation and audio projects are the same kind of material, with a history behind them.

Buyers of whole companies' records. The largest buyers of company data license a company's working records as one package, design documents, tickets, chat and email beside the code (the guide to selling company data to AI labs compares them). For a studio, those records hold the reasoning behind every design decision, which is exactly what buyers say they value.

The last row is a floor, not a comparison. A player paid a few dollars an hour for background recordings of a popular game is selling something abundant. A studio holds what no player can: the source, the unreleased builds, the design history and footage of games that were never shipped.

2. What a studio's archive holds, and who wants each part

A studio's archive is not one asset. It is many kinds of data, each with its own buyers and its own rights questions, and the first job is to separate them. The table is a map; the sections after it go through the parts that decide a sale.

The distinction that matters most is between what the studio made itself and what it used under someone else's licence. Code its staff wrote, art its artists drew, footage of its own builds and its own design documents are the studio's to license. Engine code, store assets, middleware, licensed music and stock sound effects are not, even when they sit in the same repository, and an AI buyer's contract will ask the studio to warrant the difference.

Two parts deserve a word here because studios underrate them. Unreleased work is worth more than its commercial history suggests: buyers pay for what they cannot get elsewhere, and a prototype that never shipped exists nowhere else. And the records around the code, design documents, tickets and post-mortems, turn a repository into a record of decisions: why a system was rewritten, which fix closed which bug. Buyers of code ask for exactly that context, and the standard export tools lose much of it: a GitLab project export, for instance, leaves out the links between issues and to related merge requests.

3. What studios are offered: the prices on record

There is not yet a public price for a game studio's archive as a whole in a closed sale. What is on record is what buyers publish they pay, what firms supplying gameplay say they have licensed, and the prices for the parts (code, video, assets) set in sales of other kinds of data. Read together, they bound what a studio can expect.

How studios are paid through the firms that license gameplay

Much of the studio gameplay that reaches developers goes through a few firms that gather titles from many studios, and their terms are revenue shares and negotiated fees rather than published prices. Origin Lab, which raised $8M in May 2026, lists 25+ publisher partners and 80+ licensed titles. It evaluates a studio's games at no cost, negotiates fees whose terms vary by modality, exclusivity and scope, and sends rights holders a quarterly royalty statement by title, buyer, hours and revenue. Protege formalises each licence through a revenue share agreement. VARSAV Game Studios, a game developer listed in Warsaw, signed a framework agreement under which its games and digital assets may enter Protege's catalogue, each licence negotiated separately, revenue share included, and reported on 15 June 2026 that no licensing agreement had yet been executed. Worldmodeldata says it acquires data through formal licensing agreements and publishes no price; in July 2026 The Next Web reported it had no finalised customer contracts and no revenue. Troveo, a video licensing firm that names gaming as one of its six modalities, says it gathers more than 7,000 licensors, around 95 percent exclusive.

Two lessons follow for a studio. A revenue share pays only when the intermediary licenses the data on, so ask how often it has done so, for which titles and on what terms, before signing. And many of these agreements are exclusive, which shuts every other buyer out of the same footage for the length of the deal. The offer check reads such terms against the market, and the guide to exclusive and non-exclusive licences explains how exclusivity is priced.

Reading the numbers

Three things about these numbers. First, Mercor's range is a per-company figure for code, footage and records together, and its listing adds that larger studios with full company data have earned more; it is a buyer's own range, not an average of closed deals. Second, most of the other figures are ceilings or offers: SimpleClosure's test market recovered about $10K a deal against a quoted range far above it, and the same gap between headline and outcome shows up at other intermediaries. Third, the per-hour video prices are for footage in general. Gameplay with inputs is scarcer than ordinary footage and is bought for a different purpose, but no buyer has published a per-hour price for it yet; until one does, the per-hour medians are a floor, not a ceiling.

The valuation puts a range on a whole archive from these records, and the rates page shows how each median moves as sales are added.

4. Gameplay with inputs: what world-model developers want

World-model developers want footage that teaches cause and effect, and the input log is what turns a recording into that. A recording of a game session with the controller or keyboard state for every frame lets a model learn that this button, at this moment, made the character jump. Without the inputs, the same footage is ordinary video, priced like any other.

The firms that license it say the same in their own words. Troveo says the most useful gameplay data pairs the footage with what the player did and what changed, and that whole sessions, not highlights teach the ordinary cause and effect of moving through a world. Worldmodeldata records raw controller and keystroke inputs aligned to the exact frame and state they produced and extracts 200+ structured properties per frame from the engine. Origin Lab's list of what every hour it sells contains is the most complete specification on record: video before and after the HUD, audio, keyboard and mouse inputs, camera telemetry, world state and physics, and player position and pose, frame-synced on one clock, with 100,000+ hours ready to ship. VARSAV describes its own datasets as motion trajectories, telemetry, player input data, gameplay events and synchronized video with depth maps.

A studio can supply this without handing over its code. VARSAV says its data packages come without sharing source code or dedicated game builds, and DeepMind's game agent doesn't need access to a game's source code. Microsoft's Muse shows the consent a large studio relied on: matches were recorded only if the player agreed to the End User License Agreement, and the research team's use was governed by a data-sharing agreement with the game studio.

Game data has limits, and buyers know them. NVIDIA's head of world-model development told WIRED that models trained on game inputs are unlikely to fare well with fine-grained motor control such as handling objects, so game footage teaches movement, navigation and cause and effect rather than robot hands. Worldmodeldata's chief executive believes video game data will make up the majority of world models' training material; that is an intermediary's view, not a measurement.

What makes a studio's gameplay worth more, in the terms buyers use:

• Inputs recorded with the frames. Controller, keyboard and mouse state, with timestamps that line up with the video, captured in the build rather than reconstructed afterwards.
• Hours, and variety. Many hours across levels, modes, genres and games. A studio with several titles, prototypes and unreleased builds holds variety no single game offers.
• Unedited sessions. Raw playtests, QA runs and internal capture, not cut trailers. Mistakes, retries and idle moments are part of what a model learns from.
• Context. Which build, which level, which settings, and the game's own state where it was logged. Telemetry tied to the session adds it.
• Footage of games no one else has. Public footage of a released game can be recorded by anyone who buys it; footage of a prototype cannot.

What lowers it is the same as for any video, with one addition. People in the frame or on the microphone (playtesters' faces, voice chat, streamers' commentary) bring their own consent questions, and some buyers refuse identifiable people outright: Protege, for one, does not permit Name-Image-Likeness (NIL) duplication / copying / use from media companies, including voice. Recordings made on players' machines are the players' footage, not the studio's, unless the studio's terms say otherwise. And music and licensed sound in the mix stay with their rights holders; our video guide explains how footage is cleared of them.

A studio that kept its own playtest and QA recordings with the inputs already holds what these firms build capture pipelines to produce. One that did not can start recording its own builds this way now, with testers who have agreed in writing to the use, and every hour it captures is footage of its own games that no one else can record.

5. Game code: what code buyers check

Code buyers judge a game repository the way they judge any other, and the codebase guide covers their screens in full. Three of their rules matter most for a studio. Turing's codebase screen asks for at least 50,000 lines of code, English comments and issues, a Dockerfile that builds and runs in a container and more, which rules out small or tool-only repositories. It takes only pull requests by human authors, so code written with an AI assistant counts against a repository. And buyers want the history and the tickets with it, not a snapshot: an export that loses the links between commits and issues loses much of what they pay for.

Game code raises three questions of its own.

Where the engine ends and the game begins. A studio on Unreal Engine often keeps a modified copy of the engine in its repository. Epic's licence says the licensed technology includes any modified version of any of the foregoing that you make under the License, and engine code may go only to a third party who is separately licensed by us to use the same version. The game code built on top is different: as between the studio and Epic, the studio owns all rights, other than rights in the Licensed Technology, in the Products it develops. The engine directory comes out; the game code stays.

Middleware and platform code. Physics, audio, networking and analytics SDKs, and the code a studio writes against a console maker's development kit, are third-party code. The SDKs themselves are their makers' to license, and code written against a console maker's development kit can carry material covered by that maker's own agreements, which a studio reads before any such code leaves the building. Strip vendor code and measure what is net and the studio's own; buyers do the same.

Version control that is not Git. Many studios keep their history in Perforce or another system built for large binary files. The history is just as valuable there, but buyers' tools mostly expect Git, so plan the export, and keep the changelists' descriptions and the links to tickets when you convert.

6. What a studio does not own

Buyers' contracts ask the studio to warrant that it holds the rights to what it licenses, and buyers have reasons to check: one data vendor's head of environments described being offered a large company's codebase by a group that had illicitly obtained a copy of the codebase. A game archive is unusually full of other people's property. These are the parts to find and separate before anything is offered, and they are why a rights file comes before a price.

The engine

Unreal Engine. Epic's end user licence lists, among the things a licensee's activities must not result in, using the Licensed Technology as a training input to any Generative AI Program, and the licensed technology includes Epic's engine code, its examples and starter content, and the studio's own modifications of them. The same licence leaves the studio owning its own Products, so the game's own code, art and design are the studio's to license. The engine and Epic's content inside the project are not.

Unity. Unity's terms say plainly that your User Content is yours. But the same terms, as updated on 30 June 2026, also say a customer will not use Unity's software, or data, assets or outputs derived from it or generated with it, to train AI models without Unity's prior authorization, nor authorize or cooperate with anyone who does. Unity does not say whether it reads that clause to reach a studio's own game builds, exported assets or gameplay footage. Until it does, a studio that built in Unity asks Unity in writing before it licenses any of them for training, and keeps the answer in its rights file. The studio's own source code, written in its own editor, is the clearest case of its own content; footage and builds produced by Unity's runtime are the ones to ask about.

Assets bought from stores

Assets bought from an asset store are licensed to the studio for use in its games, not owned by it. Unity's Asset Store licence says the buyer may not use Assets for purposes such as training an artificial intelligence or machine learning model without the express consent from the Provider and/or Unity, and its store terms extend that to using Assets for data sets or as inputs to such models. Other stores and libraries have their own licences: read each one before any of its content goes into an archive. In practice the studio lists every third-party asset in the project, removes them from what it licenses, and says in the datasheet that it has done so.

The same applies to footage. Gameplay that shows store-bought models, textures or effects on screen is the studio's recording of its own game, but the assets in the frame are someone else's work. Whether a licence covers footage that merely shows an asset is a question for the studio's counsel and the buyer, asset by asset; buyers that want clean footage prefer games and scenes built from the studio's own work.

Music, sound and voice

Licensed music, stock sound effects and sound libraries are licensed for use in the game, and almost never for AI training. Exclude them from audio projects and footage, or deliver footage with the music removed. The studio's own recordings, composed and recorded under contracts that give the studio the rights, are the studio's.

Voice actors and motion-capture performers are a separate question, because their performance is their likeness. A performer's contract may grant the studio the recording for the game and nothing else, and California makes a clause that lets a company replace a performer with a digital replica unenforceable unless it describes the intended uses with reasonable specificity or the performer had a lawyer or a union behind them. A union agreement the studio signed can add terms of its own. Read every performer's contract before licensing any recording of them, and expect buyers to ask: some will not take identifiable voices or likenesses at all.

Publishers and funders

A studio that made a game under a publishing agreement may not own all of it. Publishing deals variously assign the game's intellectual property to the publisher, license it exclusively, or leave it with the studio and license distribution only; funding agreements for unreleased projects can do the same. The agreement decides, title by title. A studio licenses only the titles whose rights it kept, or licenses the others with the publisher's written consent, and lists which is which in the rights file.

Contractors and freelancers

Under US copyright law, work an employee does within the job belongs to the employer, and work ordered from a contractor counts as made for hire only in listed categories, among them a contribution made as a part of a motion picture or other audiovisual work, and only with a signed agreement saying so. Otherwise copyright stays with the contractor, and a transfer is not valid unless it is in writing and signed. Game studios rely heavily on freelance artists, composers, writers and porting houses. Check that each one signed an assignment, or a work-for-hire agreement that fits, and exclude their work until the gap is closed. One intermediary names this as the usual blocker: missing ownership causes most problems.

Players

Player accounts, emails, chat logs, voice chat and telemetry tied to a player are personal data, and the rules that govern them are those of any customer data: what the game's privacy policy promised when the data was collected, and the privacy laws of the players' states and countries ( is it legal to sell customer data? walks through them). Games played by children carry a stricter rule. The US Federal Trade Commission said in its 2025 amendments to the children's privacy rule that disclosing a child's personal information to train or otherwise develop artificial intelligence technologies is not integral to the service and needs separate parental consent. Most studios leave player data out of an archive altogether, or license only aggregate figures.

7. How to prepare a studio archive

Preparation is most of the work of a sale, and for a studio it is mostly sorting. The guide to preparing data for sale covers the general steps; these are the ones particular to games.

1. Make an inventory, not a copy. List every repository, title, prototype, tool and archive by system, with dates, sizes and what each holds. Buyers decide from the inventory and a datasheet whether to go further, and no data needs to move to get there.
2. Mark what is yours. For each item, note the engine, the store assets, the middleware, the licensed audio, the performers and the contractors involved, and which titles were made under a publishing agreement. That list is the rights file, and it decides what can be offered.
3. Keep the history whole. Export repositories with their full history and their links to tickets and reviews, from Git or Perforce. A snapshot without history is worth a fraction of the same code with it.
4. Record gameplay the way buyers want it. If you hold builds that still run, capture sessions with the inputs logged frame by frame, from testers who have agreed to the use in writing.
5. Take out personal data. Players' data, testers' faces and voices, staff names in tickets and authors' emails in commits are replaced or removed before anything is shown. The de-identification standard sets out how, and the de-identification tool does it on your own computer.
6. Write the datasheet and offer a sample under an agreement. A datasheet describes the archive in the terms buyers read first. A small, cleaned sample under a confidentiality agreement shows quality without handing over the archive; the guide to proving data quality without sharing it explains the levels of proof.
7. Decide whether to sell it as one package. A whole-company buyer prices the archive as one; code buyers price per repository and video buyers per hour. A non-exclusive licence lets a studio license the same archive to more than one buyer, or different parts to different buyers, and the price of exclusivity is a separate negotiation.

8. When the studio is closing

A closing studio's archive is often its most valuable remaining asset, and it is perishable. Hosted tools delete a cancelled account's data on their own schedules, so the first step in any shutdown is a full export of every system (repositories, issue tracker, wiki, chat, build servers and cloud storage) before subscriptions lapse. Atlassian's agreement, for one, says the customer owns all intellectual property and other rights in Customer Data, but the data has to be retrieved before the account ends. The deletion countdown shows each system's window.

Who may sell depends on how the studio is closing. A company winding down on its own can sell through its directors; one in an insolvency proceeding sells through the trustee or administrator; and a dissolved company's assets follow its jurisdiction's rules. The page on closed and insolvent companies, the guide to selling customer data in bankruptcy and the guide to who owns a dissolved company's data cover each case. A studio sold to or absorbed by a publisher usually carries its archive with it, and the archive is then the publisher's to license.

Supply is not the constraint. SteamDB counted 20,008 games released on Steam in 2025 alone, and every one of them can be recorded by anyone who buys it. What a buyer cannot record is a studio's source, its unreleased builds and its design history, which is why a closing studio's archive is worth selling before it is deleted.

9. Worked example: a mid-sized studio

An example, not a real company. A studio of about 40 people in the United States has made games for 12 years. It shipped four titles in Unity, one of them under a publishing agreement that assigned that game's intellectual property to the publisher, and it holds two prototypes that never shipped. Its archive: Git repositories with full history and an issue tracker linked to them, design documents, internal QA recordings of all its games (some with input logs, most without), voice-over recorded by contracted actors, music composed by a freelancer and a few store-bought asset packs.

What can be licensed. The source of three shipped titles and both prototypes, with history and tickets; the design documents and post-mortems for the same; the QA recordings with input logs, after testers' faces and voices are removed; and the studio's own 3D source files. Because everything was built in Unity, the studio first writes to Unity under its terms on AI training and keeps the answer in its rights file.

What stays out. The published title's code and footage, unless the publisher consents in writing; every store asset, and footage scenes built around them; the freelancer's music, unless the freelancer's contract assigned it; and the voice-over, unless the actors' contracts cover this use. Player data stays out entirely.

Where it goes. The package as a whole fits a whole-company buyer such as Mercor's game studio listing, whose published range for a studio is $100,000 to $1 million. The code alone would be priced per repository by code buyers, within the public range of $1,352 to $5,869 per repository for typical repositories and above it for large or rare ones. The input-logged recordings are the part a world-model developer would bid for separately. A competitive round puts the package and its parts in front of each kind of buyer, and a non-exclusive licence keeps the studio free to license the parts the first buyer does not take.

10. How a sale through us works

AIData.Broker represents holders selling or licensing their data to AI companies, a studio's archive among them. We work for the holder only: we value the archive against public prices, settle its rights with the holder, and put it before every buyer that fits at once, so the holder sees competing offers rather than the first one. The holder itself licenses or sells to the buyer, and we do not take ownership of the archive.

1. A valuation first. The valuation asks a few questions about the archive (what it holds, how many years, who holds the rights) and returns a range from the public prices on the board, with no upload.
2. The rights file and the datasheet. We go through the rights questions with the holder, part by part (engine, assets, performers, publishers, contractors), record the answers, and write the datasheet that describes the archive; the holder approves it before any buyer sees it.
3. The round. A blind profile of the archive goes to every buyer that fits, without the holder's name, less any buyer the holder leaves out. Buyers who want to go further sign an undertaking and see the datasheet and, where the holder agrees, a sample.
4. Offers and the licence. Buyers make offers, which we set side by side; the offer check reads any written offer against market terms. The holder chooses, and signs the licence with the buyer directly.

The steps and documents are on how it works. Our fee is a success fee on what the holder is actually paid, band by band: 12% of the first $250,000, 10% up to $1 million, 8% up to $3 million and 6% above $3 million, and nothing if nothing sells; on a $300,000 sale it comes to $35,000. The mandate is not exclusive: the fee is due only on a sale to a buyer we introduced, during the mandate and, if that buyer is on the list we send in writing when it ends, within 12 months after it. We run sales only for holders based in the United States, the United Kingdom and Canada, and the smallest sale we take on is $25,000.

Questions people ask

Can I sell my game's source code to an AI company? Yes, the code your studio wrote, with its history, is what code buyers want, and they price it per repository: the middle half of public prices is $1,352 to $5,869 per repository, with legacy and unusual codebases priced higher, up to $10K to $100K+ at Turing. Engine code is not yours to include: Unreal Engine's licence lets engine code go only to someone Epic separately licenses.

How much do AI companies pay for a game studio's data? Mercor's listing for game studios puts it at $100,000 to $1 million per company for records, code and footage together. Code alone and footage alone are priced by the repository and the hour; no buyer has yet published a per-hour price for gameplay recorded with inputs.

Can I sell gameplay footage for AI training? Yes, footage of your own builds, recorded by your studio, is yours to license, and world-model developers want it most when the inputs are recorded with it. Leave out music and sound you licensed, people who appear or speak without having agreed to this use, and any store assets you have not cleared. If your game was built in Unity, ask Unity first: its terms require prior authorization for training on outputs made with its software.

Can I license assets I bought from the Unity Asset Store or another store? No, not without the provider's consent. Unity's Asset Store licence forbids using assets for training an artificial intelligence or machine learning model without the express consent from the Provider and/or Unity, and other stores' licences need reading one by one. Remove them from what you license.

Do I need my publisher's permission to license my game's data? It depends on the publishing agreement. If it assigned the game's intellectual property to the publisher, or licensed it exclusively, the publisher holds the rights and must agree in writing; if the studio kept the rights and licensed only distribution, it usually does not. Read the agreement title by title.

Can I sell my players' data? Rarely, and it is the part buyers trust least. Player data is governed by your privacy policy at the time it was collected and by privacy law; for children, the FTC requires separate parental consent before their data is used to train AI. Most studios leave it out.

What do world-model companies want from game studios? Hours of gameplay with the controller or keyboard inputs recorded frame by frame, across many games and situations, with the game's state where it was logged. Worldmodeldata says it has licensed almost 1 million hours of such data from studios, and footage of unreleased games, which no one else can record, is the scarcest kind.

What this means for you

A game studio's archive is one of the few holdings in this market that several kinds of buyer want at once, for different reasons: code buyers for the source, world-model developers for the gameplay, whole-company buyers for the records around both. The studios that earn the most will be those that know which parts are theirs, keep the history and the inputs, and put the archive in front of all of those buyers rather than the first one that writes.

1. Know the numbers. The price board, the rates page, the source code page and the video page hold every public price behind this guide, and the valuation puts a range on your archive from them.
2. Sort the archive before anyone sees it. Separate the engine, store assets, middleware, licensed audio, performers' recordings, publishers' titles and contractors' work from what your studio made. The rights check asks the questions a buyer will.
3. Keep and capture what is scarce. Export repositories with their full history and tickets; record your own builds with inputs, from testers who have agreed to it.
4. Find the buyers that fit. The buyer board lists who buys code, footage and company records, with their published terms; buyer match finds the ones whose rules fit what you hold, and what labs buy shows where demand is.
5. Read every offer against the market. The offer check compares a written offer with market terms, the guide to data licensing agreements covers the clauses, and the page on going direct or through a broker sets out the trade-off.

AIData.Broker acts as the holder's agent: the holder itself licenses or sells its archive to the buyer, and we do not take ownership of it. The holder pays us a fee fixed in the mandate and paid only if a sale closes, and we run sales only for holders based in the United States, the United Kingdom and Canada.