Back to Guides
Enterprise Software 15 min read

Buying AI training through a marketplace: what you trade away

Buying AI training through a marketplace: what you trade away

A marketplace is built to make choosing fast. Choosing was never the part that failed. The sessions that go nowhere usually came from a reasonable shortlist, a fair price, and a provider with good reviews. What was missing sat outside the listing: nobody had read the team’s work before the room filled up, and nobody had agreed to be judged on whether that work changed.

One note first, because the phrase misleads. Search AI training marketplace and most of what comes back concerns marketplaces for model training data: labeling vendors, dataset catalogs, licensing deals. Different industry. This page is about the platforms through which a company buys AI courses, workshops, and trainers for its own people. Whether to buy that capability at all or build it is worked through in our guide to buying versus building AI enablement capability, and why a stack of purchased seats stopped satisfying finance committees is argued in the case against buying licenses and calling it capability.

Four Different Things Are Called a Marketplace

They have different economics and fail in different ways, so knowing which one you face answers most of the rest.

Type Example What the platform does What it never does
Course marketplace Udemy, Maven Lists courses from independent instructors, handles payment, ranks results Write curricula, employ the instructor, or guarantee an outcome
Content aggregator Go1, OpenSesame Bundles third-party course libraries into one subscription with one contract Teach live, or tailor anything to your documents
Provider broker TrainUp and similar directories Matches a request to firms in its network and takes a fee or margin on the booking Deliver the session, which a network firm does
Talent marketplace Freelance and expert-network platforms Surfaces individual trainers, holds escrow, charges both sides Scope the work or stand behind the result

The broker model is the one buyers misread most often. A directory that answers within a day feels like a vendor, but it is a routing layer, and the firm that walks into your meeting never saw your request in the form you wrote it. TrainUp, as one public example, describes an open network of more than 300 training companies where niche providers compete alongside larger partners. That is a real service, and a statement that the platform is not the provider.

The Take Rate Is the Number Nobody Shows the Buyer

Every marketplace keeps a cut, and its size shapes who lists, how much effort a listing earns, and whether an instructor would rather route you through a platform or sell to you directly. Two published sets of terms sit at opposite ends.

Udemy’s published instructor terms give the instructor 37 percent of a sale the platform sources. For subscription revenue, which covers the Udemy Business catalog, the instructor share was set on a declining schedule: 20 percent in January 2024, 17.5 percent in January 2025, and 15 percent in January 2026.

Maven keeps 10 percent and passes 90 percent to the instructor, after card processing. Its worked example on a $1,000 enrollment: $100 to the platform, $26.40 to Stripe, $873.60 to the instructor.

Read those as predictions, not trivia. Where the platform keeps most of the money, the catalog fills with content that is cheap to make and reusable across every buyer, and strong instructors treat the listing as marketing for work they would rather sell direct. Where the platform keeps a tenth, the instructor’s name carries the sale and what you get depends on which person you booked. They are different products, and the second is why Maven cohorts and a private company session rarely compete for the same budget.

What the Intermediary Is Accountable For

Less than buyers assume, and the platforms are usually honest about it in writing. A listing for a seat-based AI course on a cloud marketplace states a per-user annual price, a 12-month term, and a vendor refund policy of “N/A”, and notes that entitlements expire if the contract is not renewed. Nothing there is deceptive. It is a clear description of a transaction with no result obligation attached. Three gaps follow, and they are the gaps that decide whether a purchase lands.

Nobody scopes. A marketplace matches a request to a catalog, and the request is usually a category and a headcount. No one reads your renewal notes, your statement of work template, or the spreadsheet three people rebuild every Monday. A session built without that material teaches a tool rather than your work.

Nobody owns the result. If capability does not move, the platform points to the provider, the provider points to attendance and satisfaction scores, and both are correct inside their own contract. The terms that put a result obligation on paper are the subject of our guide to the contract terms that decide whether training lands, and almost none survive a click-through purchase.

The confidentiality chain gets longer. Your nondisclosure terms, if any, now run to a network firm you did not pick and possibly to a subcontracted facilitator. For a finance, legal, or HR team, settle that before the price.

The Catalog Count Is Not Your Entitlement

Aggregator headlines advertise the library. Contracts sell a tier.

Go1 publishes an entry tier whose defining feature is a limit: choose 10, 25, or 50 courses from the library, swappable at any time. Its higher tiers advertise unlimited access to more than 80,000 courses from over 300 providers in 120-plus languages. A buyer who signed up on the headline figure can be entitled to 25 courses.

Catalog numbers also wobble. OpenSesame’s own properties variously claim more than 30,000, 40,000, 50,000, and 60,000 courses, the last paired with 140-plus publishers. Counts change and pages lag, so treat any catalog figure as a marketing number rather than a contract term.

Published prices are rarer still. Go1’s own pricing page names its tiers and lists no price against any of them, which is normal in this part of the market. Third-party purchase data from the procurement platform Vendr puts median annual Go1 spend near $16,000 and median OpenSesame spend near $26,000, with wide ranges either side. Those are benchmarks collected from other buyers, not list prices, and your own quote will be set by headcount, term length, and how hard you push. Two numbers matter more than the catalog size: how many courses your tier opens, and what the price does when headcount grows mid-term.

Buying Training on a Cloud Procurement Channel

This path is barely written about and is quietly useful. Seat-based training sits on cloud software marketplaces alongside databases and monitoring tools. One live listing for an enterprise course library shows $250.00 per user per year for a single license and $242.00 per user per year for two to 100 licenses, on a 12-month contract, with a private offer for a custom quote. Purchases run through the cloud bill.

For a company with committed cloud spend and slow vendor onboarding, that is a real shortcut: no new supplier record, no fresh security review, budget drawn from a commitment you already made. Two cautions. The published tiers are rack rates, and the private-offer path exists because larger buyers do not pay them. And the channel changes who signs, not what is delivered. The same library arrives, with the same absence of anyone reading your documents.

What You Gain, Stated Honestly

A marketplace is the right instrument more often than a practitioner likes to admit.

Speed. A broker network returns options in days. A scoped session needs a conversation about your work before anyone can quote it.

Breadth when needs are scattered. Four functions, four unrelated gaps, nobody needing the same thing: that is what a catalog is for, and one session would serve none of them well.

One contract and one invoice. Aggregators exist because managing 30 content vendors is worse than paying a margin for one.

A path with no statement of work. Procurement cycles kill more good training than bad providers do, and a manager with discretionary budget can buy seats this week.

Substitution. If an instructor disappoints, the catalog has another. A sole provider has to be replaced, which is harder than swapping a catalog course.

What You Trade Away

The matching ledger, in rough order of how much damage each one does.

  1. Scoping. Nobody arrives having read your material, and that one fact tells you more about whether a session will change a weekly task than price, format, or brand.
  2. A named accountable party. Three parties, each accountable for its own slice, add up to no one accountable for the result.
  3. Curriculum control. You buy the version that sells across many buyers. A catalog course with your logo on the first slide is still a catalog course, and that trade is set out in our look at a catalog provider’s team offering.
  4. Confidentiality posture. Shared rooms run on sample data by necessity, and your real documents are where the value is.
  5. Remedies. Platform terms replace the clauses you would negotiate directly: a redo, a withheld final payment, a named facilitator nobody can swap without consent.
  6. A measurement frame. Marketplaces report completions, hours, and satisfaction, because that is what they can see. None of those is a task that now takes less time.
  7. Bargaining room. You accept posted terms. The provider never learns your constraints, so it never adapts to them.

The Hours Column Is Missing From Every Listing

Seats are priced in dollars and spent in hours, and the hours are the scarcer budget.

ATD’s 2026 State of the Industry report puts average formal learning at 16.7 hours per employee in 2025, up from 13.7 the year before, with direct learning spend at $846 per employee, down from $1,254. That is two days of formal learning per person per year, and less cash per head than one marketplace course costs.

Put both numbers in the comparison. A $250 annual seat that consumes 20 hours is a larger claim on a person than a session costing $600 a head that consumes four. At a loaded $75 an hour, those 20 hours are worth $1,500, so the cheap seat is the expensive purchase. That arithmetic is worked through in full in our cost-per-outcome comparison of live training and online courses.

Your Shortlist Can Shrink Without You

Marketplaces consolidate. Coursera completed its acquisition of Udemy on May 11, 2026, in an all-stock deal announced at roughly $2.5 billion in equity value. Two of the largest course marketplaces became one company, and every buyer who kept both on a shortlist for vendor diversity lost that diversity without doing anything.

More is coming, because the field is crowded: Training Industry’s 2026 landscape map counts more than 1,400 corporate training providers across 21 segments. A multi-year platform commitment in a market reshaping itself that fast deserves a shorter term than the vendor proposes.

A Decision Rule by What You Need

Your situation Buy through Why
Scattered needs across several functions Course marketplace or aggregator Breadth is the point; alignment is not the goal
One department, one recurring task to change Direct, scoped session The output is a shared method, which no catalog ships
Compliance or safe-use coverage at headcount Aggregator subscription Audit trails and completion records are the deliverable
Sensitive documents at the center of the work Direct, with your own confidentiality terms Shared rooms run on sample data
A named expert on a specialist skill Course marketplace The instructor’s reputation is the product
Urgent, no procurement runway Broker or cloud marketplace Speed is worth the margin this once
Board asking what the AI spend produced Direct, with a measurement agreement Completion data will not answer the question

Most companies of 50 to 500 people need both, in order: one scoped session for the team whose work matters most, then catalog seats for whoever kept going. Buying both at once, company-wide, is how a budget disappears with nothing to show. For who sits where in this market, see our survey of AI training companies serving corporate teams.

Diligence Questions for a Marketplace Purchase

Six questions, each of which has cost someone money.

  1. Who delivers, and is that named in the contract? If the answer is a network, ask who assigns and whether you can refuse.
  2. How many courses does my tier open? Not the catalog headline. The tier.
  3. Does the instructor also sell directly? If so, direct is often cheaper and better scoped.
  4. What is the refund and non-renewal position, in writing?
  5. What happens to the price if headcount grows 20 percent mid-term? Overage terms can run at or above list.
  6. What will you have read of ours before the first session? A platform cannot answer this. That is the point of asking.

Frequently Asked Questions

What is an AI training marketplace?

A platform that connects companies to AI courses, workshops, or trainers it does not employ, taking a cut of each transaction. Four kinds exist: course marketplaces of independent instructors, content aggregators that bundle third-party libraries into one subscription, broker directories that route requests to a network, and talent platforms that surface individual trainers. It handles discovery and payment. It does not scope the work.

Why do searches for “AI training marketplace” return data-labeling companies?

Because the same phrase names a much larger industry: marketplaces where AI developers buy training data and human labeling work. If you are buying courses for your team, add words like corporate, for employees, or vendor to the search.

Is buying AI training through a marketplace cheaper?

Per seat, usually yes. Per outcome, often no. The seat price excludes the employee hours the course consumes, and those hours are the larger cost at any realistic loaded rate. A marketplace is cheapest when the need is scattered and no shared method is wanted.

What do marketplaces charge providers?

It varies more than any other number in this market. Published terms range from a platform keeping 10 percent to a platform keeping roughly 63 percent of a sale it sources, with subscription catalog pools paying listed instructors as little as 15 percent of attributed revenue in 2026. A high take rate predicts reusable content and absent instructors.

Can I get a custom workshop through a marketplace?

Sometimes, and the word “custom” is doing a lot of work. Broker networks and some course platforms sell private sessions, which usually means an existing curriculum on your own calendar. Ask what changes besides the schedule, and which of your documents the session will use. If neither answer is specific, you have bought a catalog course with a private label.

Who is liable if marketplace-bought training does not work?

In practice, no one. The platform’s obligation is access, the provider’s is the session, and neither contracts on capability. Refund policies on seat-based listings are sometimes stated as not applicable. A result obligation has to be written directly with whoever stands in the room.

Can we buy corporate training through a cloud marketplace?

Yes. Seat-based course libraries are listed on cloud software marketplaces with published per-user annual prices, 12-month terms, and a private-offer path for custom quotes, billed through your cloud account. It is a procurement shortcut for companies with committed spend. It changes who signs, not what gets delivered.

Does marketplace consolidation affect buyers?

It does. Coursera completed its purchase of Udemy in May 2026, so two of the largest course catalogs now sit inside one company. Keep platform commitments to 12 months where you can, and check that catalog access survives a change of ownership.

Where to Start

Price both paths on one sheet with four columns: fee, hours per person, loaded cost of those hours, and the one task you expect to look different in six weeks. Then ask each option the last diligence question. A platform sends a catalog link. A provider who should get your money asks what your team does on Mondays.

If you are holding a quote for 40 catalog seats and a quiet doubt that anything will change, that doubt is the useful signal. Book a free AI-readiness call and we will go through the work your team repeats every week, where a model would carry real weight, and whether you need a catalog, a session, or both in sequence.

Last Updated: Nov 9, 2026

AW

Arthur Wandzel

SFAI Labs helps companies build AI-powered products that work. We focus on practical solutions, not hype.

Make your team fluent in AI — then automate what proves out

  • Hands-on training applied to your own documents and workflows
  • Often reimbursable through state small-business training programs
  • Built for 5–50 person firms with no IT department

Related articles