The AI-trainer economy turned two years old this quarter, and the structural picture is now sharp enough to publish a cut on. Three platforms — Scale AI, Mercor, and Surge AI — between them anchor more than 70 percent of frontier-lab RLHF and expert-data spend. Scale has crossed the $1B annual revenue line under Alexandr Wang. Mercor, under Brendan Foody, is on a placement-volume trajectory that, by the math the company has disclosed in 2025 capital-raise documents, lands the firm in the $800M to $1.1B annualized run-rate window during the 2026-27 horizon. Surge AI, Edwin Chen's bootstrapped counter-thesis, closed 2025 in the $300-500M revenue range with operating margins above 40 percent and zero venture capital. The labor layer of frontier AI is now a $3B+ annual market, growing at +186% year-over-year placed volume across the top 10 employer cohort, and the cohort itself has split cleanly into three tiers that price, recruit, and graduate talent on different curves.
The question this analysis sets out to answer is a forward-looking one. If the data-for-AI category was created in 2024 and consolidated in 2025, what is the structural shape of the category in late 2026 and 2027? The three findings below — drawn from our Top 10 AI Trainer Employers 2026 ranking, the compensation cascade documented in our Q2 2026 State of AI Hiring report, and the founder cut in our Top 30 AI Founders to Watch in 2026 — are the answers the data forces.
The summary thesis: the trainer economy is bifurcating along the seam between commodity RLHF and expert-IC matching, the in-house migration at frontier labs is the binding constraint, and the platform that wins the next cycle is the one that converts trainer engagement into permanent senior-IC placement faster than the labs can build their own funnels.
Framing
Cohort: the 10 firms in our Top 10 AI Trainer Employers 2026 ranking. Data window: Q4 2024 — Q1 2026, with 2026-27 forecast based on disclosed 2025 capital documents and frontier-lab RLHF spend allocations cross-referenced against our 280-CHRO panel. Senior-IC funnel = the path from RLHF expert engagement to permanent W-2 employment at a frontier lab or AI-native applied firm. AAA = ENTRA Talent Index top-tier rating; Scale, Mercor, and Surge each hold AAA on the 2026 cut. Placed volume = total compensation paid through to experts during the period.
1. The category split into commodity and expert tiers — permanently
The cleanest line through the 2025 data is the bifurcation between commodity RLHF and expert-IC matching. They are no longer the same business.
Commodity RLHF — generalist annotation, preference labeling, generic instruction-tuning data — is now a volume-and-margin race. Outlier and DataAnnotation, both Scale AI sub-brands, route the bulk of this volume; Toloka and Appen handle the residual long tail. The economics are cents-per-task at scale, with margins compressed by the willingness of frontier labs to bring portions of this work in-house through their own data orgs. Anthropic's internal RLHF organization expanded materially through 2025 under the operating leadership of its data-org partners; OpenAI ran a parallel build. The commodity tier still produces revenue, but it is producing it inside a margin compression that the next two years will only sharpen.
Expert-IC matching is the opposite economy. Mercor, the senior-PhD layer of Scale, and Surge AI's premium-only positioning sit on the other side of the bifurcation. PhD rates clear $200/hour at p90 — Foody at Mercor disclosed in Q4 2025 that the top 5 percent of placements earn above $250 — and the work is increasingly indistinguishable from the early stages of a senior-IC interview loop at a frontier lab. The economics here are not volume-and-margin. They are placement-and-conversion: the platform earns its premium by converting trainer engagement into permanent hires at frontier-lab compensation bands.
The structural consequence is that the AI-trainer economy is now two markets sharing a vocabulary. Scale, the only firm that operates meaningfully in both, is increasingly running them as separate P&Ls internally, and the 2025 organizational cuts the company made — separating the Outlier mass-market organization from the senior specialist engagement layer — confirm the separation is now load-bearing.
2. The Mercor IPO question is the wrong question
Press speculation in Q1 2026 has converged on a Mercor IPO, with valuation chatter ranging from $4B to $7B depending on the desk. The question is the wrong question. The right question is whether Mercor's permanent-hire arm — the conversion of expert-marketplace engagement into full-time placement at frontier-lab and AI-native employers — ships before the labs build the equivalent funnel internally.
Foody, Adarsh Hiremath, and Surya Midha built Mercor as an AI-native expert marketplace. The 2025 milestone — $50M+ placed volume, p90 hourly rates above $200, the September 2025 Series B at $250M post — was the validation that the marketplace economics worked. The 2026-27 milestone, the one that determines the company's terminal multiple, is whether the same screening and matching infrastructure that priced expert hours can be redirected to price permanent senior-IC offers. Foody signaled the direction in Q4 2025 disclosures. The execution is the open variable.
The strategic logic is that the most expensive part of a senior-IC frontier-lab hire today is the eighteen-to-thirty-six-month relationship-building that precedes the offer. A platform that has already screened, engaged, and placed an expert on three Anthropic or OpenAI engagements has compressed that relationship-building into a credentialed conversion event. The cycle-time savings — which we modeled in our State of AI Hiring Q2 2026 report at 47 to 64 days against a 14 to 19 day frontier-lab founder-led median — are the value Mercor's permanent arm captures if it ships.
The risk to the thesis is that Anthropic, OpenAI, and Google DeepMind ship their own internal expert-engagement-to-permanent-hire funnels first. The 2025 expansion of internal RLHF orgs at Anthropic and OpenAI is the early signal. The 2026 question is whether those internal orgs can run the conversion conversation at frontier-lab cycle-time without the AI-native screening discipline Mercor has built. We expect the labs to produce a partial internal funnel by Q1 2027 and to retain Mercor for the high-margin specialist domains — biotech-AI, regulated-finance AI, defense-AI — where the expert pool is too small to build in-house economically.
3. Surge AI is the founder-economics counter-thesis
Edwin Chen's Surge AI is the most underappreciated structural fact in the category. The firm took zero venture capital, runs an internal engineering team of roughly 80, anchored Anthropic's Constitutional AI training pipeline through 2024 and 2025, and closed 2025 in the $300-500M revenue range with operating margins above 40 percent. By per-employee revenue, Surge is the highest in the data-for-AI category and one of the highest in the broader AI-native applied cohort.
The compensation philosophy is the part that should reshape how founders in the category think about subsequent capital raises. Surge senior engineers clear $280K-$520K base, with cash bonus and profit share replacing equity. The pitch — frontier-lab compensation parity without the venture-equity volatility — closes a senior-IC cohort that the venture-funded competitors cannot reach without dilution they cannot tolerate. Chen has, in effect, built the only frontier-lab-adjacent compensation package in the category that does not depend on a 7-to-10-year liquidity event.
The signal-to-the-market is that Foody, Hiremath, and the Outlier and Invisible Technologies founders have all visibly recalibrated their 2025 capital-raise posture toward operating discipline rather than headcount expansion. Foody disclosed in Q4 2025 that the Series B was structured to preserve operating discipline rather than fund growth. The Surge counter-thesis is now embedded in the category's capital-allocation defaults. We expect at least two of the venture-funded firms in the top 10 cohort to convert toward profit-share-and-bonus compensation structures during 2026.
4. The talent funnel from RLHF to senior-IC is now the largest non-academic pipeline
Roughly 3,400 Outlier experts moved into permanent senior-IC roles at frontier labs and AI-native employers through 2025 via structured recruiting funnels operated jointly by Scale and frontier-lab People functions. Mercor's 2025 placement disclosures include a non-trivial fraction of placements that converted to permanent W-2 offers within 90 days. The structural consequence: 6 to 12 months of Mercor or Outlier engagement is now treated by senior PhDs as a credentialed path into Anthropic, OpenAI, Cohere, or Thinking Machines Lab senior-research-engineer bands.
The implication for Q3 and Q4 2026 senior-IC hiring is that the funnel composition shifts. Mira Murati's Thinking Machines Lab, which closed six of seven founding-engineer hires through direct founder outreach in 2025, will run its 2026 hiring against a candidate pool that now includes a meaningful fraction of prior trainer-platform engagement. We expect 15 to 20 percent of the 2027 frontier-lab senior-IC cohort to have prior RLHF engagement experience on resume — up from roughly 4 to 6 percent in 2024. This is not a marginal shift. It is the formation of a new credential category.
The CHRO implication for the broader Fortune 500 enterprise AI tier — the 612 firms in our Q2 State of AI Hiring panel — is that the trainer platforms are now the most efficient sourcing surface for senior AI engineers who already have shipped on a frontier-lab data pipeline. Several Fortune 100 CHROs confirmed in Q1 2026 conversations that they have begun running structured recruiting against the Mercor and Outlier alumni cohorts. The cycle-time savings are real and the conversion rates have already exceeded their internal recruiting baselines.
5. The 2026-27 forecast: three structural shifts
The in-house migration accelerates, but stops short of full vertical integration. Roughly 30 to 40 percent of frontier-lab RLHF spend shifts in-house by end of 2027. Scale, Mercor, and Surge respond with deeper vertical integration into specialist domains — Defense-AI, biotech-AI, regulated-finance AI — that the labs cannot easily build internally. Scale's Defense-AI vertical, anchored on Thunderforge and the DoD primes, is the canonical hedge. Surge's Anthropic Constitutional AI pipeline — bespoke, premium, deeply embedded — is the second model. Mercor's specialist verticals will be the third by Q4 2026.
Mercor's permanent-hire arm ships in Q3 2026 and reshapes the AI recruiting category. The same screening infrastructure that priced expert hours redirects to price permanent senior-IC offers. The category consolidates around three or four AI-native screening platforms by end of 2027, displacing the agency-anchored model that has dominated AI recruiting through 2025. We expect at least one Fortune 100 enterprise to designate Mercor as preferred senior-AI recruiting partner during 2026.
The bootstrapped-and-profitable thesis spreads beyond Surge. At least two of the venture-funded firms in the top 10 cohort restructure compensation toward profit-share and cash bonus during 2026. The signal-to-the-senior-IC-market is that frontier-lab compensation parity is achievable in the category without dependence on venture liquidity events. This will reshape recruiting at the senior-engineer level for the next capital cycle.
The structural takeaway for any CHRO, founder, or capital allocator reading this: the AI-trainer economy is no longer an adjacent market to frontier AI hiring. It is the senior-IC funnel, the compensation-philosophy laboratory, and the candidate-screening infrastructure that the labs and the enterprises will increasingly route their hiring through. The platforms that convert trainer engagement to permanent placement fastest will own the next recruiting cycle. The platforms that stay inside the commodity-RLHF margin compression will be the platforms whose category-defining role we are documenting today and whose structural position will look very different in the 2027 cut.
The next vertical analysis publishes Thursday, May 14 — IT vertical week — and will cover the cybersecurity hiring shape post-Microsoft Copilot for Security GA. For the underlying ranking, see Top 10 AI Trainer Employers 2026. For the founder cohort closing these moves, see Top 30 AI Founders to Watch in 2026. For the broader market context, see The State of AI Hiring — Q2 2026. For the founder-economics piece in dialogue with the Surge counter-thesis, see We Charged $1 Per Job Post.