AI data market
Turing AI data business model: where talent access ends, data begins
By SourceX Editorial · Reviewed by Noah Loul ·
10 sections
Short answer
Turing's AI data business model, read from its own terms, is mostly talent access: contributors' work for clients is owned by the client as work made for hire, and enterprise clients keep their code and IP. The likeliest places for data assets are pre-defined data packs and Project Lazarus, which acquires startups' codebases and records, but Turing publishes no ownership, exclusivity or reuse terms for either. Value the lines separately.
License terms
Turing AI data business
Key takeaways
- Turing's On-Demand Technical Professional Terms say work created under an order form is owned by the client as work made for hire, so talent-network deliverables are not Turing's data.
- Turing Intelligence's build page says enterprise clients keep control of code and intellectual property, which places that line on the access side too.
- Pre-defined data packs and Project Lazarus are the likeliest lines to hold reusable data, and Turing publishes no ownership, exclusivity or reuse terms for either.
- Turing's 4M+ vetted profiles and its earlier one-million developer community use different units and no dates, so they show registered supply, not growth or work delivered.
- The test for investors is the share of revenue from datasets or archives delivered to more than one customer, which Turing does not disclose.
What is Turing's AI data business model, and what does Turing own?
Turing's own pages organize the company into three areas: AGI Advancement, which supplies data, environments and evaluations to AI labs; Turing Intelligence, which builds enterprise AI systems; and a talent network that places remote engineers. Its published terms put the work that contributors produce for clients in the clients' hands, so most of what Turing sells is access and process rather than data it owns.
Two lines could hold data assets that Turing keeps: pre-defined data packs offered to labs, and Project Lazarus, which acquires codebases and operating records from startups. For neither does Turing publish who owns the output, whether licenses are exclusive or whether a dataset is sold more than once.
That gap is the decision for an investor. Valuing Turing as a data company assumes it owns reusable data; valuing it as a talent platform assumes its margin comes from matching, vetting and managing people. The evidence needed to choose is contract-level: who receives the rights to each deliverable, and whether any deliverable reaches a second buyer. The ledger below traces each line as far as Turing's published material allows, read on 2026-10-28, and leaves the rest blank on purpose.
Turing's lines, traced from contributor to owner
Each row follows one Turing line from the person who makes the work to the party that owns it, using only Turing's own pages and terms. Where Turing publishes nothing, the cell says so.
Two of the five rows end with the client as owner under published terms and three end in a blank. The blanks sit exactly where a data asset would have to be:
| Line | What Turing says | Who makes the work | Owner under published terms | Not disclosed |
|---|---|---|---|---|
| Talent network | Remote engineers matched to companies; 4M+ vetted talent profiles on the homepage | Independent contractors | The client: work under an order form is owned by the client as work made for hire | Active contractors, hours, revenue |
| Turing Intelligence | Builds enterprise AI systems inside the client's environment | Turing teams with the client's staff | The client keeps control of code and IP, per the build page | Pricing, revenue, client count |
| AGI Advancement: custom work | RL environments, data generation systems and benchmarks for frontier labs | Contributors; contributor terms not published | Not published | Contributor terms, lab terms, revenue |
| AGI Advancement: data packs | Pre-defined packs or custom datasets; samples validate scope before full volumes | Contributors; contributor terms not published | Not published | License terms, exclusivity, buyers per pack |
| Project Lazarus | Acquires codebases and operating records; the company decides scope and redaction | Startups, in the normal course of their business | Not published | Ownership after transfer, exclusivity, resale |
What Turing's published terms assign, and to whom
Turing's published contract terms assign contributor work to the client, not to Turing. The On-Demand Technical Professional Terms of Service, in the version indexed when read for this article, say that all work created by technical professionals under the agreement, as incorporated in the applicable order forms, is owned by the client and treated as work made for hire.
The same terms let the client hand its own invention assignment and confidentiality agreement directly to the technical professional, with Turing assisting only in an administrative capacity, and they state that the professionals Turing sources are independent contractors, not employees of either party. The client, in turn, warrants that it holds the rights to whatever materials it gives them.
Turing Intelligence follows the same pattern. Its build page says clients keep control of code, intellectual property and project direction, and that the work happens inside the client's own environment. What Turing claims for itself is narrower: its website terms say the intellectual property in its Services, their original content, features and functionality, remains Turing's exclusive property. That is the platform and the method, not the deliverables.
Read together, these terms describe a business whose retained asset is the ability to find, vet and organize contributors quickly. Nothing in them gives Turing a right to reuse a client's deliverable, which is why an investor should not count client work toward any data inventory Turing might be said to hold.
Talent access: what 4 million profiles do and do not measure
Turing's talent figures measure registered supply, not work delivered. The homepage reports 4M+ vetted AI and engineering talent profiles from more than 100 countries; the earlier Series D post, which described Turing as an AI-backed Intelligent Talent Cloud matching remote software engineers with companies, said its developer community had passed one million.
The two figures use different units, profiles against developers, and neither carries a date in the excerpts read, so the change between them is not a growth rate. Neither counts active contractors, billed hours or utilization, which are the measures that drive a talent platform's revenue and margin.
For an investor, talent access is valuable when three things hold: the platform can supply scarce skills faster than a lab or enterprise could hire them, the vetting holds up in delivered quality, and the platform keeps the relationship when the project ends. None of those depends on owning data. They depend on matching speed, contributor retention and customer concentration, and Turing publishes figures for none of them.
Where a Turing data asset could sit: packs and acquired archives
A Turing data asset, if one exists, would sit in the pre-defined data packs or in the archives Project Lazarus acquires. Both are described in Turing's own words, and both stop short of the terms that would settle ownership.
Turing's AGI Advancement page says its post-training offering includes curated Data Packs, structured RL Environments and research-grade Benchmarks. Its datasets page presents them as pre-defined packs or custom datasets tailored to a lab's experiments, says each pack is available as a sample, and says samples are designed to validate scope and quality before engagement on full volumes. No license terms, prices or usage restrictions appear. A pre-defined pack sold to more than one lab would require Turing to hold resale rights from the contributors who made it; contributor terms for that work are not published.
Project Lazarus is clearer about intent. Its landing page says Turing acquires codebases and operating records from startups, that the company decides what is in scope, what is redacted and what never leaves its environment, and that payment is a one-time amount confirmed after a sample review. Turing's December 2025 newsletter, titled Can an AI Learn From a Company That No Longer Exists?, describes acquiring full private repositories, design documents, testing logs, infrastructure manifests and internal wikis from defunct startups in fintech, pharmaceuticals, manufacturing and SaaS, on the argument that existing datasets miss how work actually gets done.
Lazarus is the one Turing line that acquires a stock of records instead of commissioning new work. Whether it becomes a reusable asset depends on terms Turing has not published: who owns the records after transfer, whether the license is exclusive, and whether Turing may package them for several labs.
Project Lazarus and SourceX, compared on published terms
Project Lazarus and SourceX both pay companies for records of work already done, so they can be compared on the terms each publishes. The comparison uses Turing's Lazarus pages and SourceX's live pages, read on 2026-10-28; it measures what is public, not which program is better, and ranks neither.
The rows where Lazarus reads not published are the same rows that decide whether an acquired archive is an asset: ownership after transfer, exclusivity and prior licensing. SourceX's column is a labeled application of the same test, in the words of its live pages:
| Published term | Turing Project Lazarus | SourceX, from its live pages |
|---|---|---|
| What is acquired | Codebases and operating records from startups, including defunct ones | Business records a company licenses from its own systems |
| Who sets scope | The company decides what is in scope and what never leaves its environment | Set by the company in the scope and terms it signs, ahead of any delivery |
| Removal of personal details | The company decides what is redacted | SourceX removes identifying details itself or through its processing partner |
| Ownership after transfer | Not published | The company keeps its original data; SourceX owns the de-identified dataset under the signed agreement, subject to the company's IP |
| Exclusivity | Not published | Exclusive for AI training; the term is set in each Dataset Deal Confirmation |
| Prior licensing | Not published | The company confirms the data has not been licensed for AI training before |
| Payment structure | One-time amount confirmed after a sample review | Usually a one-time fee, fixed in the signed Dataset Deal Confirmation |
The case that Turing already runs a data business
The best case against this page is that Turing already runs a data business and the talent network is its production cost. Part of the evidence points that way.
Turing's own Series E announcement, read on 2026-10-28 and also covered by its AGI Advance newsletter of March 11, 2025, said the company raised $111 million at a valuation of $2.2 billion, bringing total funding to $225 million since its founding in 2018, and named AGI advancement first among the uses of the round, describing Turing as a research accelerator for frontier labs through its data and post-training programs. Its current AGI Advancement page goes further, saying Turing develops large-scale RL environments and data generation systems that train multimodal agents for coding, real-world tasks and advanced STEM reasoning. Pre-defined packs and an archive acquisition program are the building blocks a data company would use.
Against it, the only ownership terms Turing publishes send deliverables to clients, and the figures it emphasizes, vetted profiles and matching, are talent metrics. Turing's press page also lists a revenue headline whose underlying release could not be opened for this article, so no revenue figure is used here. Until Turing reports revenue by line and the terms behind its packs, the defensible reading is a talent platform with two possible asset lines of unknown size. One number would test it: the share of revenue from datasets or archives delivered to more than one customer.
Where the ownership ledger could mislead an investor
Six risks could make this reading wrong, or make Turing's model less durable than its funding suggests.
At a glance
Where the ownership ledger could mislead an investor
- Dated terms: the On-Demand Technical Professional Terms were read in the version indexed by search, and later order forms or statements of work may assign rights differently.
- Unpublished contributor terms: if contributors to data packs assign their work to Turing rather than to a lab, Turing may hold resale rights that this page treats as unknown.
- Undated figures: talent counts appear without dates in the excerpts, so neither growth nor scale can be computed from them.
- Buyer concentration: frontier labs are few, and Turing names no lab customers or revenue shares on the pages read.
- Rights in acquired archives: records from defunct startups can contain customer, employee and third-party content, and a closed company may be unable to warrant every right a licensee needs.
- Automation of contributor work: if labs generate more post-training data synthetically or grade it automatically, demand for vetted talent hours could fall faster than demand for recorded work.
Diligence questions on Turing's contract chain, and on SourceX's
Ask for the contract chain behind each line, not just its revenue. The same questions apply to SourceX.
SourceX sits on the archive side of this comparison. As its investor page tells it, SourceX obtains licenses to records from the businesses that created them and then acts as licensor of record to AI developers, covering every step from sourcing and rights review to contracts, preparation, delivery and payment; the supplying company signs off on scope and terms before anything is handed over, and only that signed agreement lets SourceX use, sublicense and distribute the de-identified dataset. Revenue and deal metrics are not posted on its site, and investors receive company and commercial figures directly. Accredited investors can request materials through the SourceX investor page, as the company is raising under Rule 506(c).
- Which party owns a deliverable on each line, and which contract says so?
- Do contributors to pre-defined data packs assign their work to Turing or to a lab?
- How many times has any pack or acquired archive been licensed, and to how many buyers?
- Who owns Project Lazarus records after transfer, and is the license exclusive?
- What share of revenue came from talent hours, enterprise builds, packs and archives last fiscal year?
- What share of revenue comes from the largest customer and the five largest?
Methods note: sources, dates and the cells left blank
The ownership ledger and the Lazarus comparison are editorial frameworks built for this article on 2026-10-28. The evidence set is Turing's own domains: the homepage, the Series E announcement, the AGI Advancement and datasets pages, the On-Demand Technical Professional Terms of Service, the website Terms of Service, the Turing Intelligence build page, the Project Lazarus landing page, the December 2025 newsletter on Lazarus and the Series D post. SourceX's column uses its live how-it-works and investor pages.
Every Turing source was verified through search excerpts on Turing's domains, not by opening the page, and a reviewer should open each before release. Excluded: press estimates of revenue or valuation, a revenue headline on Turing's press page whose release could not be opened, job postings, pay rates, the payment amount on the Lazarus page and conflicting Series D round sizes on Turing's careers pages.
Not measured: revenue by line, margins, active contractors, billed hours, contributor terms for data packs, Lazarus contract terms, reuse of any dataset and customer concentration. No company is ranked, and SourceX metrics are not compared because SourceX publishes none.
Frequently asked questions
What does Turing do in AI?
Turing's homepage organizes its work into AGI Advancement, which supplies data, environments and evaluations to AI labs, Turing Intelligence, which builds enterprise AI systems, and a talent network that places remote engineers. It also runs Project Lazarus, which acquires codebases and operating records from startups.
Who owns the work Turing's contractors produce for clients?
Turing's On-Demand Technical Professional Terms say work created under an order form is owned by the client and treated as work made for hire, and the client can have contractors sign its own invention assignment. Ownership terms for data packs and Project Lazarus are not published.
Is Turing a data company or a talent platform?
On its published terms, most lines are talent access: clients own the deliverables. Pre-defined data packs and Project Lazarus could hold reusable data, but without published ownership and reuse terms, the share of Turing that behaves like a data company cannot be measured from outside.
What is Turing's Project Lazarus?
Turing describes Project Lazarus as a program that acquires codebases and operating records from startups, with the company deciding what is in scope and what is redacted, for a one-time payment confirmed after a sample review. Its December 2025 newsletter says it targets defunct startups in fintech, pharmaceuticals, manufacturing and SaaS.
What should an investor ask Turing before calling it a data company?
Ask which party owns each deliverable and under which contract, whether contributors to data packs assign their work to Turing, and how many times any pack or acquired archive has been licensed. Revenue by line answers the rest.
Sources
- Turing's homepage organizes its work into AGI Advancement, Turing Intelligence and a talent network, and reports 4M+ vetted AI and engineering talent profiles from more than 100 countries. Source
- Turing's own Series E announcement says it raised $111 million at a $2.2 billion valuation, bringing total funding to $225 million since its 2018 founding, with AGI advancement first among the uses and Turing described as a research accelerator for frontier labs. Source
- Turing's AGI Advancement page says its post-training offering includes curated Data Packs, structured RL Environments and research-grade Benchmarks, and that it develops RL environments and data generation systems for coding, real-world tasks and STEM reasoning. Source
- Turing's datasets page presents pre-defined packs or custom datasets, with each pack available as a sample to validate scope and quality before engagement on full volumes. Source
- Turing's On-Demand Technical Professional Terms say work created by technical professionals under order forms is owned by the client as work made for hire, that professionals are independent contractors, and that clients may supply their own invention assignment. Source
- Turing's website Terms of Service say intellectual property in its Services and their original content, features and functionality remains Turing's exclusive property. Source
- Turing's enterprise build page says clients keep control of code, IP and project direction and that work happens inside the client's environment. Source
- Turing's Project Lazarus page says it acquires codebases and operating records from startups, that the company decides what is in scope, redacted or kept in its environment, and that payment is one-time, confirmed after a sample review. Source
- Turing's December 2025 newsletter describes acquiring repositories, design docs, testing logs, infrastructure manifests and internal wikis from defunct startups in fintech, pharma, manufacturing and SaaS. Source
- Turing's AGI Advance newsletter dated March 11, 2025 covers the Series E announcement. Source
- Turing's Series D post described it as an AI-backed Intelligent Talent Cloud matching remote software engineers with companies, with a developer community of over one million. Source
Request SourceX investor materials
SourceX shares company and commercial metrics with investors directly. Any offering is made only to verified accredited investors.