Market structure
The independent human-data market consolidated twice
EXL completed its acquisition of iMerit while Labelbox acquired Upcraft. AuraOne’s independent, expert-led positioning became scarcer—and more strategically valuable.
Executive Intelligence Brief
The human-data market consolidated twice in one window: EXL completed its acquisition of Tier 1 competitor iMerit, and Labelbox bought Upcraft to automate its expert pipeline. Independent, expert-led positioning is getting scarcer and more valuable.
Amazon is closing Mechanical Turk on September 30. Buyers still on it have five weeks to migrate - the cleanest near-term sales opening of the week.
Rep. McCaul (via NYPost) named SurgeAI, Mercor, Turing, and AfterQuery as supplying Chinese AI labs while serving US and Pentagon clients. A US-clean supply chain just became a sellable attribute.
Evaluation is becoming a product battleground: Mercor is hiring for in-house APEX Benchmarks; Snorkel shipped Terminal-Bench 4.0 plus a grants program. Our evals lane needs a named public offer.
Frontier-lab robotics demand stayed hot (DeepMind partner-fleet pattern, iMerit's Carbon Robotics pretraining deal) - maps to our P0 DeepMind and NVIDIA watchlist entries.
Money followed post-training and compute: Deep Cogito raised $43M; Anthropic reportedly committed $45B to Nscale compute (unverified against a primary source).
Do today: (1) start the MTurk-migration offer, (2) draft the clean-supply-chain positioning note, (3) teardown of Snorkel Terminal-Bench and Mercor APEX vs our evals offer.
News of the Day — Decision Digest
The day’s material announcements, translated into the market shift and decision they create for AuraOne.
Immediate demand
Mechanical Turk’s shutdown created a five-week migration window
Teams still using MTurk need replacement capacity before September 30. This is a time-bound account-discovery motion, not merely industry context.
Procurement
Data provenance is becoming a buying criterion
The McCaul report tied several U.S. data vendors to Chinese labs and Pentagon work. Verifiable consent, sourcing, and U.S.-clean delivery can now affect vendor selection.
Evaluation
Competitors are moving from supplying eval data to owning benchmarks
Mercor’s APEX hiring and Snorkel’s Terminal-Bench expansion show that the benchmark itself is becoming a product and a route into the buyer conversation.
Capital
Funding is flowing into post-training and frontier infrastructure
Deep Cogito’s $43M round and reported hyperscale compute commitments signal downstream demand for RL environments, expert data, and independent evaluation.
Physical AI
Robot builders continue buying external training data
The disclosed iMerit–Carbon Robotics work reinforces a practical pattern: robotics companies procure specialized pretraining and demonstration data rather than building every pipeline internally.
Top 5 Developments for AuraOne
1
EXL COMPLETES ACQUISITION OF iMERIT
Strategy ShiftRelevance9/10ConfidenceHigh
EXL (NYSE-listed, ~68,000 employees) completed its acquisition of iMerit, a Tier 1 human-data competitor strong in expert annotation, RL, and robotics data. Founder Radha Basu joins EXL as EVP. Separately disclosed: iMerit built pretraining data (millions of plant images) for Carbon Robotics' agricultural robot foundation model.
Why AuraOne cares
EXL can bundle data work into enterprise contracts, pressuring our Enterprise Intelligence lane. Flip side: independent expert-led shops are now rarer for frontier labs, and the Carbon deal proves robotics firms buy external demonstration data at scale - validation for our Physical AI lane.
Action
Update the iMerit battlecard to EXL+iMerit this week; use the Carbon Robotics pattern in Physical AI outreach.
2
AMAZON SHUTS DOWN MECHANICAL TURK (SEPT 30)
New SignalRelevance8/10ConfidenceMedium
Amazon confirmed Mechanical Turk closes permanently September 30, 2026 after 21 years; new signups were frozen July 30. Named migration destinations are Prolific, Surge, Mercor, and Scale - we are not on that list, which is fixable.
Why AuraOne cares
Every team still running labeling, surveys, or human review on MTurk must migrate in five weeks. Research-panel work in our Synthetic Research lane is directly affected.
Action
Build an MTurk-migration offer (human review, annotation, research panels) and a target list of MTurk-dependent teams this week.
3
CONGRESSMAN NAMES 4 US DATA FIRMS SERVING CHINESE LABS AND THE PENTAGON
New SignalRelevance8/10ConfidenceMedium
NYPost reports Rep. Michael McCaul (R-TX), citing a new report, says SurgeAI, Mercor, AfterQuery, and Turing supply China's leading AI labs while also working for US labs and the Pentagon - roughly $500M/yr from Chinese labs across the sector.
Why AuraOne cares
If procurement restrictions or disclosure rules follow, a verifiably US-clean, consent-documented supply chain becomes a differentiator we can sell today - especially in the Government lane - and a quiet contrast point against Surge and Mercor.
Action
Draft the clean-supply-chain one-pager this week; obtain the underlying McCaul report.
4
EVALUATION BECOMES A PRODUCT BATTLEGROUND
Strategy Shift / AccelerationRelevance7/10ConfidenceMedium
Mercor posted a Research Scientist role for in-house "APEX Benchmarks" plus Frontier Data Products engineers. Snorkel AI shipped Terminal-Bench 4.0 (rolling agentic SWE benchmark), Terminal-Bench-Science, and an Open Benchmarks Grants program.
Why AuraOne cares
Whoever owns the benchmark owns the buyer conversation. Two data-market competitors are moving from supplying data for evals to owning the evaluation artifacts. Our evals lane needs a named, public offer with a visible artifact.
Action
Teardown of Terminal-Bench and Mercor APEX vs our capabilities by Friday; decide benchmark, benchmark-operations service, or both.
5
ANTHROPIC REPORTEDLY COMMITS $45B TO NSCALE COMPUTE
New SignalRelevance7/10ConfidenceMedium
Reports say Anthropic signed a $45B six-year compute rental with neocloud Nscale (~460MW, Monarch campus); Nscale separately closed ~$3B in deployment financing. Anthropic is simultaneously hiring Staff+ engineers for an internal RL Data Platform.
Why AuraOne cares
Compute commitments at this scale precede training and post-training scale-ups, which precede data, RL environment, and evaluation procurement. The window to sell data operations to Anthropic is now, before more is internalized.
Action
Verify the $45B figure before external use; refresh the Anthropic account plan around RL-environment operations.
Frontier Lab Intelligence
Google DeepMind: Robotics partner ecosystem kept expanding (Boston Dynamics Spot on Gemini; an Agile Robots partnership claim, unconfirmed). Strongly indicated need: external holdout sets of robot failures, human corrections, and adjudicated regressions across partner fleets. Buyers per SSOT: Sally Jesmonth, Caden Lu. Pitch: partner robots generate failures daily; we turn them into training and release-eval assets.
Anthropic: $45B compute buildout + internal RL Data Platform hiring. Strongly indicated need: RL environments, evals, vendor-run production data pipelines while the internal platform is built. Buyer per SSOT: Brett Wittmershaus (existing thread). Pitch: your platform team builds the platform; we run expert operations on top this quarter.
Meta Superintelligence Labs: Aggressive post-training and agentic-safety hiring continues; Muse Spark 1.1 already on our SSOT. Strongly indicated need: agentic tool-use failure holdouts and expert-authored intervention benchmarks. Note: a circulating "Meta bought Assured Robot Intelligence this week" claim is stale (May 2026) and was excluded.
DeepSeek: Published its first V4 vision model as open MIT-licensed weights (single low-authority source, low confidence). Possible need: independent multimodal evaluation. Confirm before acting.
Per the customer exclusion rule, OpenAI (canceled pilot) and xAI (excluded) are not covered as customer news.
Competitor War Room
- iMerit (Tier 1, now EXL): acquired; Carbon Robotics pretraining-data partnership disclosed. Threat: Medium. Battlecard update needed.
- Surge AI (Tier 1): named in the McCaul report; secondary claims of >$1B 2024 revenue. The "$1B raise" item circulating is recycled July 2025 news - excluded. Threat: High (unchanged).
- Mercor (Tier 1): APEX Benchmarks hiring + $614M H1 2026 revenue claim + McCaul naming. Threat raised Medium to High on the evals move.
- Innodata (Tier 1): record quarter (+58% YoY) and CEO transition to Rahul Singhal; hiring in health-AI evaluation. Threat: Medium.
- Appen (Tier 1): H1 revenue US$119.9M (+17%), China segment +80% to $76.2M (now majority). Turnaround real but China-weighted. Threat: Medium.
- Snorkel AI (Tier 2): Terminal-Bench 4.0/Science + grants - positioning to own agentic eval standards.
- Labelbox (Tier 2): acquired Upcraft to automate its expert data pipeline.
- Checked, no verified material change: Scale AI, Invisible Technologies, Turing, Sama, Toloka, Handshake, Prolific, DataAnnotation, TELUS Digital AI, Encord, Centific.
Money Movement
- EXL acquires iMerit - amount not disclosed. Enterprise services buying expert data and robotics capability.
- Anthropic-Nscale: $45B over six years (unverified secondary figure); Nscale ~$3B financing (press release). Frontier-lab compute buildout via neoclouds.
- Deep Cogito: $43M Series A (TQ Ventures; Benchmark, Nexus) for a post-training engine. A new funded buyer of post-training data and evaluation - qualify as P2 GTM target.
- Ringg (India voice AI): backed by Peak XV. Voice-agent builders are future buyers of multilingual speech data.
- Capital-flow conclusion: capital is concentrating at both ends of the stack - compute at the bottom, post-training and evaluation at the top - while the independent human-data middle consolidates. AuraOne's margin lives in that middle; defend it with expert depth and clean provenance.
Signals Worth Watching
Datoric (YC W26, formerly Arzule): private custom training-data pipelines for frontier labs, 300k+ vetted contributors claimed. Verify; check Sept 14. Confidence: Low.
AfterQuery: untracked data firm named in the McCaul report alongside Surge, Mercor, Turing. Verify profile; check Sept 7. Confidence: Low.
Zappi Synthetic Idea Pre-Screener: incumbent insights platform shipping synthetic respondents - encroaches on our Synthetic Research lane at the low end. Monitor monthly.
MimesisIQ: persistent synthetic consumer panels for CPG decisions. Check Sept 30. Confidence: Low.
Tech Mahindra x Uniphore "Agentic AI Factory": large SIs packaging agent-plus-human delivery - the Enterprise Intelligence pattern at scale. Check Sept 30.
Government: UK announced a 100M GBP competition for British AI companies fixing public services (Aug 31) - check eligibility. MiGA dataset preprint (103k demonstrations, five gripper types) shows scaled demo data is a live robotics bottleneck - Physical AI opening, time horizon: now.
Net New Leads & Account Triggers
Evidence-backed revenue candidates discovered in today’s intelligence. Every candidate must be deduplicated against the Revenue & GTM workbook before it is created or routed.
Net new lab
P1
Deep Cogito
- Verified trigger
- $43M Series A for a post-training engine.
- AuraOne fit
- Expert reasoning data, RL environments, and independent model evaluation.
- Likely buyer
- Post-training, research, data, or evaluation leadership.
- Route
- Deduplicate in the GTM workbook; qualify funding, team, current vendors, and buyer before outreach.
Verified data buyer
P1
Carbon Robotics
- Verified trigger
- A disclosed external-data relationship used millions of plant images for an agricultural robot foundation model.
- AuraOne fit
- Physical AI collection, long-tail field data, expert adjudication, and release evaluation.
- Likely buyer
- AI/ML, autonomy, data operations, or foundation-model leadership.
- Route
- Check existing-account status, map the current data program, and build a non-duplicative account hypothesis.
Emerging buyer
P2
Ringg
- Verified trigger
- $5.5M Series A and disclosed production scale of more than 1.5M customer conversations per month across 18+ languages.
- AuraOne fit
- Multilingual conversational speech, accented audio, agent evaluation, and human review.
- Likely buyer
- Voice AI, data, model quality, or product leadership.
- Route
- Monitor and qualify its product scope, languages, funding, and internal data capacity before promotion to P1.
Government opportunity
P1
UK public-services AI competition
- Verified trigger
- A £100M competition was announced for British AI companies addressing public-service problems.
- AuraOne fit
- Evaluation, managed data operations, red teaming, or delivery through an eligible partner/prime.
- Likely buyer
- Program authority, eligible prime, or consortium lead.
- Route
- Verify eligibility, deadline, procurement vehicle, and subcontracting path before creating an opportunity record.
Reviewed, not routed as net-new leads
- Competitors named in the McCaul report are competitive-intelligence records, not prospects without a verified buyer or partner rationale.
- The unnamed population of MTurk-dependent teams is a list-building thesis, not a set of lead records until specific companies are verified.
Today's Action List
P0 - Do now
- Build and ship the MTurk-migration offer with target list. Owner: GTM/Growth. Live within one week; MTurk closes Sept 30.
- Draft the clean-supply-chain positioning one-pager (sourcing, consent records, US handling). Owner: CEO/GTM. This week.
- Verify the Anthropic-Nscale $45B figure against a primary source before external use. Owner: Research. Today/tomorrow.
P1 - This week
- Teardown: Snorkel Terminal-Bench 4.0/Science and Mercor APEX vs our evals capabilities; decide benchmark vs benchmark-operations product. Owner: Product.
- Update the iMerit battlecard to EXL+iMerit; pick two EXL enterprise accounts for expert-quality pilot pitches. Owner: GTM.
- Advance Anthropic (existing thread) with an RL-environment operations offer. Owner: GTM.
- Refresh DeepMind and Meta robotics/safety outreach with the partner-fleet failure-eval angle. Owner: CEO.
P2 - Monitor
- Verify AfterQuery and Datoric profiles; decide tier upgrades. By Sept 14.
- Check UK 100M GBP competition eligibility. By Sept 14.
- Track Anthropic RL Data Platform hiring as an insourcing signal. Ongoing.
Strategic Insights
1. Consolidation is hollowing out the independent middle of the human-data market (EXL-iMerit, Labelbox-Upcraft). Make "independent and expert-led" an explicit positioning pillar.
2. The end of Mechanical Turk marks the final migration from crowd labor to expert labor. The market's center of gravity moved to what we sell; the task is capture, not education.
3. Data provenance and geopolitics are becoming procurement criteria (McCaul report; Appen's China-majority revenue). A documented US-clean supply chain wins deals size alone would lose.
4. Evaluation is the next contested layer and it is being productized without us (Mercor APEX, Snorkel Terminal-Bench, Innodata health evals). Decide this week: benchmark, benchmark-operations service, or both.
5. Physical AI demand keeps compounding (DeepMind partner fleets, iMerit-Carbon Robotics, MiGA's 103k demos). Package the Physical AI offer as a named product.