Who Is Actually Hiring AI Trainers in the US

Three companies account for most US AI training spend. Mercor reached roughly $2.0bn annualised gross revenue by June 2026, Surge AI a $1.4bn run rate, and Scale AI guided to about $1bn. Their end clients are the frontier labs, and the work is almost entirely contract.

Key takeaways

• Six platforms carried 2,282 active AI training listings in September 2026, with Mercor holding 1,277 of them, or 56.0% (aitrainer.work census, 8 September 2026).
• Mercor paid out more than $2m a day to contractors during 2026, up from more than $1.5m a day in October 2025, and raised $350m at a $10bn valuation in October 2025.
• OpenAI, Anthropic and Meta are named as end clients across the three largest vendors, so the work traces back to the frontier labs rather than to the platforms themselves.
• Only 10.0% of one platform’s 300,000-strong registered pool is active in a given week, which is why registering is not the same as getting work.
• Between 97.8% and 100% of listings across the major platforms require a screening interview, so any offer of immediate paid work without assessment is worth a second look.

The three companies behind most AI training work

Revenue, not listing counts, shows where the money is. Mercor went from about $500m annualised revenue in October 2025 to roughly $2.0bn annualised gross by June 2026, booking $614m in the first half of 2026 alone.

Surge AI reported a $1.4bn run rate in late 2025 on a staff of about 110, running roughly 50,000 expert contractors. Scale AI recorded $870m in 2024 and has guided to about $1bn for 2026, down from roughly $2bn in 2025 after Meta took a 49% non-voting stake for $14.3bn in June 2025 and several labs moved their contracts elsewhere.

Which platforms carry the most live listings?

Mercor, by a wide margin. Across 2,282 listings active on 8 September 2026, Mercor held 1,277 or 56.0%, Micro1 461 or 20.2%, SME Careers 253 or 11.1%, AfterQuery 204 or 8.9%, Turing 82 or 3.6% and Alignerr 5 or 0.2%.

Listing share and contractor scale do not match. Mercor reports over 100,000 contractors hired to date against 400 or more employees, Scale AI over 100,000 globally, and Surge AI roughly 50,000 experts on about 110 staff. That last ratio works out at 455 contractors per employee, which tells you how much of the process is self-service.

Who are the end clients paying for this work?

The frontier labs. OpenAI and Anthropic are named as Mercor customers, with Meta reported as a major client, and Surge AI names Anthropic as its reference reinforcement learning partner for Claude with OpenAI and Meta also clients.

That matters for anyone weighing the work, because the demand is not platform marketing. It is a procurement line at companies spending billions on model development, and the roles reflect what those programmes need rather than what a marketplace wants to sell. The same pattern runs through AI contract hiring generally, where the split between engineers who build models and those who deploy them decides which skills get contracted.

Why the demand appeared so suddenly

Because reasoning models changed what the data has to contain. Labelled examples trained earlier systems; reasoning models need worked professional judgement, and the only source of that is the professional.

Mercor chief executive Brendan Foody explained the commercial blockage at TechCrunch Disrupt in October 2025, telling TechCrunch that “there is an argument that Goldman Sachs doesn’t love the idea of having models that are able to automate their value chain” and that “their customers don’t want to give them data to automate large portions of their value chains”. The same article reported Mercor paying industry experts up to $200 an hour and planning expansion into law, finance and medicine.

Why are labs hiring people instead of buying data?

Because the firms holding the data will not sell it, and no price fixes that. A bank or a law firm whose value sits in professional judgement has no incentive to hand over the material that would let a model reproduce it.

Hiring the former senior employee routes around a contract that will never be signed. That is why the fastest-growing part of this market is credentialed rather than general: physicians, lawyers, engineers and researchers producing output in their own field. The adjacent demand for the machine learning roles employers are competing hardest for comes from the same programmes.

What the work actually pays

Two tiers, and the gap between them is the single most important thing to understand before applying. Generalist rating averages $40.50 an hour in the US. Credentialed specialist work averages $108, reaching $180 in healthcare and $500 for STEM professors.

Full rate tables by percentile, vertical and platform sit in our dedicated pay-rate breakdown. The short version for anyone deciding whether to pursue this: the credential sets the band, and platform tenure does not move it.

How does contract AI training compare with an in-house role?

Contract pays more per hour and far less reliably. An in-house enterprise AI trainer earns $47,000 to $87,000 a year with a $63,000 median, against $84,240 annualised for the contract average at $40.50 an hour over 2,080 hours.

The catch is that nobody works 2,080 hours on these contracts. Only 3.3% of the 460 Mercor listings that disclose hours state a floor of 35 hours a week or more, and only 36.0% disclose hours at all. Engagements average 3.0 weeks for business work and 23.5 for language work, so continuous income needs several overlapping contracts rather than one. That calculation mirrors when interim work beats a permanent seat, and the honest read is that this is contract work with uneven volume rather than a job.

How to check an AI training listing is worth pursuing

Step 1. Confirm who pays whom. No legitimate platform charges a worker to apply, to be assessed or to access work. Any request for payment from your side ends the conversation.

Step 2. Expect an assessment. Between 97.8% and 100% of listings on the major platforms require a screening interview. Immediate paid work with no assessment does not match how this category operates.

Step 3. Check the rate band is published before onboarding. Live specialist listings disclose bands from $22 to $500 an hour. A rate withheld until after you have signed up is a signal, not an administrative delay.

Step 4. Read the hours disclosure, or note its absence. Only 36.0% of Mercor listings and 4.6% of Micro1 listings state weekly hours. Where hours are not stated, treat the income as unforecastable rather than assuming full-time.

Step 5. Register on more than one platform. Mercor holds 56.0% of listings but only 10.0% of its registered pool is active weekly, so a single registration is a thin pipeline.

Step 6. Lead with the credential, not the job history. Rate bands track licences, doctorates and senior industry records. A licence gets you into the $75 to $500 tier on day one; platform hours do not.

AI trainer hiring questions

Who is hiring AI trainers in the US?

Mercor, Micro1, SME Careers, AfterQuery, Turing and Alignerr carried 2,282 active listings in September 2026, with Mercor holding 56.0%. The end clients are frontier labs: OpenAI, Anthropic and Meta are named across the three largest vendors.

Is AI training work legitimate?

The category is real and traceable. One platform paid out more than $2m a day to contractors in 2026 and raised $350m at a $10bn valuation in October 2025. Individual listings vary, so check for a published rate band, a screening assessment and no worker-side fee.

How much do AI trainers earn?

Generalist rating averages $40.50 an hour in the US and credentialed specialist work $108, rising to $180 in healthcare and $500 for STEM professors. Full tables by percentile, vertical and platform sit in our pay-rate breakdown.

Can you make a full-time income from AI training?

Rarely from one contract. Only 3.3% of Mercor listings that disclose hours state a 35-hour floor, and engagements average 3.0 to 23.5 weeks by domain. Continuous income needs several overlapping contracts and registration across more than one platform.

What qualifications do AI training platforms want?

None for generalist rating at the $40.50 average. The $75 to $500 band requires a licence, doctorate or senior industry record in a named field such as medicine, law, engineering or research. The credential, not platform tenure, sets the rate.

Acceler8 Talent contracts credentialed US domain experts into AI training and model evaluation programmes, and places the engineers building those models. If your field is medicine, law, engineering, finance or research and you want contract work assessed against a real brief rather than a marketplace queue, our US AI recruitment desk will tell you straight whether there is demand for your credential. On the employer side, the distinction between task and title shows up the same way it does for the day-to-day work behind a contract ML engineering role.

About the Author

Dale Swords is Chief Commercial Officer at Acceler8 Talent, a US-focused recruitment firm specialising in AI, machine learning, semiconductor and high-performance computing hiring. He leads commercial terms and client delivery across the firm’s US accounts, including contract engagements with AI labs and model developers. His profile is at acceler8talent.com/about-us/meet-the-team/profile/dale-swords.