Open requests, kept current. If your company holds one of these, whether it is still trading or has closed, the rights holder can license it, with personal data taken out first.

Wanted now

Construction drawings, with their full history

Whole projects from architects, engineers and contractors: every discipline, every revision, and the records that explain each change.

What buyers ask for

Buyers want

Whole projects, not loose drawings: architecture, structure, mechanical and electrical, plumbing, fire, civil, geotechnical and landscape, as delivered
Every revision, and the records that link a problem to its fix: requests for information, clash reports, design reviews, shop-drawing reviews, variations and approvals
The native files beside the PDFs: Revit, AutoCAD, Navisworks, IFC or Tekla, with calculations, specifications and schedules where you have them
An index with one row per file, and a short data sheet for each project
Whole archives across many projects, which buyers value above single jobs

What a buyer sees first

An inventory first: counts by project, discipline and format. Nothing moves.
Then a preview set of one or two complete projects with their history, with personal data taken out first
ABURA holds only a small anonymised preview; the licensed copy goes straight from the owner to the buyer.

You confirm

You made the drawings yourselves: your own staff, or consultants whose work you own
Your contracts and confidentiality clauses allow it, project by project
Security-sensitive parts are left out
Personal details are taken out: names, signatures and contacts in title blocks and correspondence, faces in site photos
AI training is named in writing as a permitted use

Not a fit

Security-sensitive parts of airports, stations and stadiums, such as security systems and secure zones
Drawings you received from others, such as an architect's drawings a contractor was given
Drawings a client owns under contract
Defence or classified projects

Wanted now

Circuit board design files

Complete board designs from electronics teams and product makers, with everything an engineer would need to understand them.

What buyers ask for

Buyers want

The full schematic and board layout, ideally in Altium, KiCad or Eagle (other tools are welcome)
A datasheet for every component on the board
A bill of materials
Boards of medium to high complexity, with 100 or more components
Valued more: a description of each board and the requirements behind it, and notes on the hard parts of the design, such as high-speed traces or analogue noise

What a buyer sees first

An inventory first: how many boards, in which tools, and how complex. Nothing moves.
Then a preview set of the bill of materials and parts list for one or two boards, with personal data taken out first
ABURA holds only a small anonymised preview; the licensed copy goes straight from the owner to the buyer.

You confirm

You designed the boards yourselves, or own the designs outright
No client or partner contract stops you sharing them
AI training is named in writing as a permitted use

Not a fit

Designs a client owns under contract
Reference designs copied from a chip maker
Export-controlled or defence designs

Wanted now

Game code, written by people

Source code from game studios, with the history of how each game was built.

What buyers ask for

Buyers want

Game code written by people, not mostly with AI tools
Its history kept: commits, pull requests and tickets
Studios in the US, the UK, Canada and Western Europe first, though others are welcome

What a buyer sees first

The maker reads each repository as a zip on your computer and keeps only the file tree, the statistics and a few cleaned files
Secrets and author names come out, and no commit history leaves

You confirm

You own the code, not a publisher or a client
Engines and libraries you licensed from others are listed, not sold
AI training is named in writing as a permitted use

Not a fit

Code a publisher owns under contract
Open-source code or forks of it
Code written mostly with AI tools

Wanted now

Whole company workspaces, start to finish

The full record of how a company worked: email, chat, tickets, documents and code, kept together so each piece of work can be followed from start to finish.

What buyers ask for

Buyers want

One company's whole workspace, not a selection: email, chat, support tickets, the CRM and databases, code, meeting notes, documents and designs
The links between them kept, so a piece of work can be followed from start to finish, such as a new lead becoming a signed client
Companies of five or more people, fintech first, other industries welcome
Closed companies too, if the workspace survives whole

What a buyer sees first

A short profile first: industry, country, years running, headcount at its peak, the kinds of workspace data and a rough size. Nothing moves.
Then a preview set, with personal data taken out first
ABURA holds only a small anonymised preview; the licensed copy goes straight from the owner to the buyer.

You confirm

The company, or whoever holds its rights now, owns the workspace and can license it
Staff and customer details are taken out before anything moves
Client contracts and confidentiality clauses allow it
AI training is named in writing as a permitted use

Not a fit

A handful of loose files from a larger workspace
Data your clients own, or that you hold for them in confidence
Workspaces that cannot be cleared of personal data

Wanted now

Code written by people, before 2023

Private codebases from companies, with the history of how the software was built.

What buyers ask for

Buyers want

Code written by people before 2023, with its commits, pull requests and tickets
Large products, ideally half a million lines or more each, not counting libraries, configuration or generated code; less for rarer languages
COBOL, Rust, Go, Swift, MATLAB, C and C++ first
Valued more: tests that cover the code, and code from finance companies

What a buyer sees first

A description first: what each product does, its languages and size, the years it was written and whether its tickets survive. Nothing moves.
Then a preview set, with secrets and author names taken out first
ABURA holds only a small anonymised preview; the licensed copy goes straight from the owner to the buyer.

You confirm

The code is private: never published, never open-sourced, not a fork
Your company wrote it, or whoever holds its rights now can license it
Open-source libraries it uses are listed and left out
AI training is named in writing as a permitted use

Not a fit

Public or open-source code, or forks of it
Code written mostly with AI tools
Code a client owns under contract

Wanted now

Voice recordings, with every speaker's consent

Natural conversation, interviews and narration, recorded with the speakers' consent and transcribed word for word.

What buyers ask for

Buyers want

Natural conversation, interviews, call-centre calls and narration, in any language, rarer languages especially
One channel per speaker where possible, at phone or studio quality
Word-for-word transcripts, timed, showing who is speaking
A consent record for every speaker

What a buyer sees first

An inventory first: hours by kind, language and quality, and how consent was recorded. Nothing moves.
Then a short preview set, with personal details taken out first
ABURA holds only a small anonymised preview; the licensed copy goes straight from the owner to the buyer.

You confirm

Every speaker agreed to be recorded, and that agreement covers AI use or can be extended to it
Your contracts with clients, hosts or narrators allow it
AI training is named in writing as a permitted use

Not a fit

Calls recorded without the speakers' consent
Recordings a client owns, or keeps confidential under contract
Music or performances whose rights you do not hold

Wanted now

Contracts, with every redline

Agreements as they were negotiated, draft by draft, so a model can learn how terms change and why.

What buyers ask for

Buyers want

Every draft and redline of each agreement, not only the signed copy
The comments and emails that explain each change, where they survive
Many kinds of agreement: supply, services, licences, leases and employment templates
Policies, manuals and reports alongside them

What a buyer sees first

An inventory first: counts by kind of agreement, years and languages. Nothing moves.
Then a preview set of a few agreements with their drafts, with names and identifying details taken out first
ABURA holds only a small anonymised preview; the licensed copy goes straight from the owner to the buyer.

You confirm

The confidentiality terms allow it, or the other party agrees
Names, signatures and details that identify the parties are taken out
AI training is named in writing as a permitted use

Not a fit

Agreements whose terms forbid any sharing
Legal advice protected by privilege
Contracts you hold for clients, such as a law firm's client files

Not on this list?

Requests change week to week

Tell us what your company holds, and we'll match it against every request that comes in from the AI data buyers we work with. Free to list.

← Back to the overview Articles

Notes on data and AI, in plain English.

Short reads for the people who hold company data: what AI teams want, how deals work, and what to watch for. New articles also go out to our mailing list.

2 minute readWhy old code is the hottest data in AI right nowAI teams want code that people wrote before AI started writing it. Many companies have years of it sitting in their repositories.

Models learn to write software by studying software. Since 2023, more and more new code has been written with the help of AI assistants: in Stack Overflow's 2023 survey, 70% of developers said they were already using AI tools in their work or planning to.

That matters because of what happens when models learn from other models. Researchers writing in Nature in 2024 showed that when models are trained on text produced by earlier models, they gradually lose the rare and unusual cases, a problem they called model collapse. A later study found the cure: keep the original, human-made data in the mix.

So code written by people before 2023 is scarce, and it only gets scarcer. Epoch AI estimates that AI training could use up the public text written by people between 2026 and 2032, and the 2026 Stanford AI Index finds that AI-made data is still not replacing real data in training.

The AI data buyers we work with ask for human-written code from before 2023, usually hundreds of thousands of lines for each product, not counting third-party libraries, configuration or generated files. Some languages are scarce enough to be wanted in smaller amounts, among them COBOL, Rust, Go, Swift, MATLAB, C and C++. Its history adds to it: the commits, pull requests, review comments and tickets that show how real engineers reasoned and decided.

It does not have to be live code. A retired system, or the archive of a company that has closed, can count, as long as whoever holds the rights agrees. Software Heritage, which preserves public source code, warns that software is already being lost, and private repositories are beyond its reach.

If you lease or license it, the code stays yours. Before anything is shared, secrets and personal data come out, and you check that no customer contract or open-source licence limits what you can offer.

Sources

2 minute readLease, license or sell: what each means for your dataThree ways to work with a buyer. The difference is who keeps a copy, for how long, and who owns it afterwards.

Lease it. A buyer pays for access to a copy for a fixed period. When the period ends, their use stops and the copy is deleted, unless the deal says otherwise. You can lease the same data again.

License it. A buyer receives a copy to use for the purposes and the period you agree. The ownership stays with you, or with whoever holds the rights, and the same data can be licensed to others, unless you agree that one buyer has it exclusively.

Sell it. You transfer the copy, and the rights you choose to sell, for one payment. It is a clean break, if that suits you best.

Why can the same data earn more than once? As the World Bank's 2021 World Development Report puts it, using data for one purpose does not diminish its value. The OECD found that sharing data can create 10 to 20 times more value for the people who use it than for the company holding it.

Most owners lease or license, so the same data can earn again. A combination works too: a licence to one buyer and a lease to another, decided buyer by buyer.

Two things to settle in every deal. First, whether the buyer may train AI models with the data, which is set out deal by deal. Second, exclusivity: an exclusive licence means nobody else can have the data for the agreed period, so weigh it carefully.

You see every offer from the AI data buyers we work with, and nothing moves without your agreement. For a company that has closed, these decisions belong to whoever now holds the rights, such as the liquidator or administrator.

Sources

2 minute readWhat happens to a company's data when it closesThe archive outlives the company. Code, drawings and records often sit untouched on servers long after the last day of trading.

When a company closes, its data does not disappear overnight. Repositories, design files, research results and operational records are part of what the company owns, and they can still be licensed. But left alone, they decay.

A study in Current Biology asked for the data behind 516 research papers and found the odds that it still existed fell by 17% for every year after publication. Closures are common too: US government figures show about two in three new business sites close within ten years.

Who can decide depends on the stage the company is at. While it is in administration or liquidation, an office-holder acts for it. In the UK, an administrator acts as the company's agent and has power to sell or otherwise dispose of its property. Liquidators have similar powers.

Timing matters. In the UK, once a company is dissolved, whatever it still owned passes to the Crown, so nobody else can license it afterwards. If an archive is worth something, the conversation should happen before dissolution.

It can be done. When a farm data company closed in 2024, its data sets and models were sold at auction to another agriculture company. And whoever takes on a closed company's data takes on its promises: the US Federal Trade Commission said any buyer of a bankrupt genetics company's customer data must keep that company's privacy commitments.

Any income goes to whoever holds the rights: the company, or for a closed company its estate, where creditors are paid before shareholders. Former founders cannot sign for a closed company, but they often know exactly what the archive holds and who to speak to, and can introduce it as a scout.

Other countries have their own office-holders and rules. The principle is similar, and the details are for the office-holder and their advisers to confirm.

Sources

2 minute readTaking personal data out before a saleAI buyers want what your data shows about the work, not about people. Here is what taking personal data out really means.

Under Europe's data protection law, the rules do not apply to information that is truly anonymous. But data where people have been swapped for codes, and could be linked back with extra information such as a key, still counts as personal data.

Anonymous means that nobody has a means reasonably likely to be used to identify a person, on its own or combined with other information, taking account of the cost, the time and the technology available. It is a higher bar than removing names.

In practice, names, email addresses, phone numbers and ID numbers are the easy part. Personal details hide in free text, such as support tickets and chat messages, in images and recordings, and in rare combinations: a job title in a small town can point to one person.

Some data has its own test. In the US, health data counts as de-identified either when a fixed list of identifiers has been removed, or when an expert has documented that the risk of identifying anyone is very small.

Done well, it works at scale. During the pandemic, Google published aggregated, anonymised movement trends to help public health decisions, with no personal information made available.

On ABURA, a buyer first sees a small anonymised preview, made on the owner's own computer. ABURA holds only that small anonymised preview; the licensed copy goes straight from the owner to the buyer, with personal data taken out first.

If taking personal data out is more than your team can do, say so when you list. Help is part of the process.

Sources

2 minute readGiving data for good: where tax relief applies, and where it doesn'tSome data does the most good given away. In a few countries the gift can also bring a tax deduction, and in many it brings none.

Giving data means passing it outright to a university, hospital or charity that will use it. Europe now has a name for it: the EU's Data Governance Act recognises data altruism, making data available for the general interest without seeking a reward.

Where relief is possible. In the United States a gift of data can be deductible, but usually at what the data cost you, not what it might sell for. Giving a copy or only a licence generally brings no deduction. In Portugal, since August 2026, a company's gifts in kind can count, under a new rule with no rulings on data yet, due for review at the end of 2026. The Netherlands allows gifts in kind to a registered public-benefit charity, and Canada counts a gift at its value, but taxes it as if you had sold it.

Where it is smaller. France gives relief on part of what the data cost to make. Germany allows only a whole asset to be given, never a licence.

Where there is none. The UK and Singapore give no tax relief for a gift of data.

A donor-advised fund in the US adds no extra tax benefit. It is a way of giving now and choosing the charities later, through a fund that can accept an unusual gift.

A gift is most useful when it is ready to use. The FAIR principles, which research funders now ask for, set out how: data that can be found, accessed, combined and reused.

Whether a gift suits you, and what relief applies, is for your own tax adviser to confirm before you give. Nothing here is tax advice.

Sources

2 minute readLost in time: how company and research data quietly disappearsNobody deletes it on purpose. It simply stops being looked after, and one day it is gone.

Digital records feel permanent, but they are not. Pew Research Center found that 38% of web pages that existed in 2013 could no longer be reached ten years later.

Research data fares no better. A study in Current Biology asked for the data behind 516 papers and found the odds it still existed fell by 17% for every year after publication. Another study found that one scientific article in five points to web material that has since vanished or changed.

The reason is rarely a decision. People move on, laptops are wiped, servers are switched off and nobody is clearly responsible. UNESCO warned in 2003 that digital heritage is at risk of being lost to posterity for exactly these reasons.

Company data is most at risk when a company closes or a project ends. The people who knew what the files were, and why they mattered, leave first.

It does not have to end that way. When a farm data company closed in 2024, its data sets survived because another company bought them at auction.

What helps: know what you hold and where it sits, keep it in one place with a note of what it is, check who holds the rights, and decide early whether to license it, sell it or give it, while the people who understand it are still around.

Sources

2 minute readWhy AI still needs data made by peopleAI teams are running short of good human-made data, and the open web is closing its doors. Licensed company data is one of the few new sources left.

Today's AI models were trained largely on public text written by people. Epoch AI estimates the stock of good public text at about 300 trillion tokens, and that AI training could use it all up between 2026 and 2032.

At the same time, that public data is closing off. Researchers tracking AI training sources found that within one year, 45% of a widely used training set had become restricted by websites' terms of use.

Could AI simply make its own data? Researchers writing in Nature showed that models trained on other models' output lose the rare cases over time, and the 2026 Stanford AI Index finds AI-made data is still not replacing real data in training. A further study shows the answer is to keep real, human-made data in the mix.

That leaves data that was never public: the code, drawings, records and research inside companies, made by people doing real work. It is the kind of data AI teams cannot scrape and cannot fake.

Licensing it, with the owner's agreement and with personal data taken out first, is the honest way for that data to reach AI. It gives AI teams what they need, and gives the people who made it a say and a share.

Sources

2 minute readWhen company data saved livesThree times ordinary company data, made for something else, ended up doing public good.

After the Haiti earthquake. In 2010, the mobile operator Digicel shared anonymised location records from 1.9 million SIM cards. Researchers used them to show that about 630,000 people left Port-au-Prince within 19 days, a far better match for a later UN survey than the official estimates, and they mapped movements within 12 hours during the cholera outbreak that followed.

Against cancer. Drug companies and research groups pooled de-identified patient data from completed cancer trials on one free platform. By October 2014 it held data from 14 trials covering about 9,000 patients, data that would otherwise have sat unused.

In the weather forecast. Airlines' aircraft send weather readings as they fly. Europe's main forecasting centre found they gave 13% of the value of all its observations in 2019, and when flights collapsed in 2020, removing aircraft data made some forecasts up to 15% worse.

Shared research data has even trained Nobel-winning AI: AlphaFold2, which predicts the shapes of proteins, learned from the public databases of known protein structures.

In each case, the safeguard came first: data anonymised, de-identified or aggregated so that no one could be singled out. That is what made sharing possible.

The UN calls this data philanthropy, and EU law now recognises data altruism. If your data could do good like this, giving it is one option, alongside leasing or licensing it.

Sources

Use cases

What buyers train with it.

Examples of what AI teams build with each kind of company data. No buyer is named, and real stories, anonymised, will join these as deals close.

Impact and SDG data

Models that forecast harvests, disease and where help is needed, trained on years of field records.

Source code

Coding assistants that write, review and fix software, including the old languages that still run banks and governments.

Drawings and CAD

Design tools that draft, check and revise engineering and construction drawings.

Operational data

AI agents that follow procedures, handle support tickets and run back-office work.

Financial data

Models that spot fraud, judge credit risk and forecast payments.

Medical and clinical

Imaging and diagnostic models, trained only on de-identified, consented records.

Sensors and IoT

Models that spot faults in machines, vehicles and sites before they fail.

Research and science

Discovery models that read lab images and predict results.

Media and film

Tools that generate and edit video, images and sound, where the rights are clear.

Sport

Performance, tactics and injury models built on tracking and match data.

Contracts and documents

Legal and compliance tools that draft, review and compare terms, learning from every redline.

Closed companies

Any of the above, from archives that outlive the company, with the rights holder's agreement.

AI's next breakthroughs need data the internet doesn't have. Your company may hold it.

People and planet

Medical researchers

Infrastructure builders

Software teams

Sport

Media and film

Scientists and robotics teams

Financial services

A copy, not a handover.

List your data, or be introduced

We check the rights

Buyers compete

A copy goes to the buyer

Impact and SDG companies

Source code

Drawings and CAD

Operational data

Financial data

Medical and clinical

Sensors and IoT

Research and science

Media and film

Sport

Contracts and documents

Closed companies

What the AI data buyers we work with are asking for

Construction drawings, with their full history

Buyers want

What a buyer sees first

You confirm

Not a fit

Circuit board design files

Buyers want

What a buyer sees first

You confirm

Not a fit

Game code, written by people

Buyers want

What a buyer sees first

You confirm

Not a fit

Whole company workspaces, start to finish

Buyers want

What a buyer sees first

You confirm

Not a fit

Code written by people, before 2023

Buyers want

What a buyer sees first

You confirm

Not a fit

Voice recordings, with every speaker's consent

Buyers want

What a buyer sees first

You confirm

Not a fit

Contracts, with every redline

Buyers want

What a buyer sees first

You confirm

Not a fit

Requests change week to week

Don't let the work be lost in time.

From the tools you run to the offers you choose

Your preview set

Offers 3

Lease it

License it

Sell it

Interested in using your data as collateral to raise debt?

Twelve kinds of data

Rights checked first

Previews under NDA

Lease, license or buy

Join as a scout

Spot and check

Make the introduction

Help train what AI discovers next, and earn from it.

Talk it through with the team.