The atlas · 15 verticals
Every vertical answers the same six questions
Whose budget, can you get the supply, what is the spread, can you hold it, what does AI do to it, and what would kill it. The answers sit in the same order on every page so they can be read side by side. Compare all of them at once.
The three that started this
Expert data for the labs, recruiting, and paid creators. Covered to the bottom.
Expert data for frontier labs
Labs buy throughput of credentialled human labour — annotation, preference data, reasoning traces, RL environments — and pay nine figures for it at a staffing margin.
crowded 27–33% gross margin (Mercor, leaked); 42.8% published take (Prolific)
Contingency recruiting marketplaces
Paraform has raised $65M and paid $50M to recruiters. Back the split out and the whole lifetime business is roughly $71M of placements and $21M of net revenue — against a perm-placement pool an order of magnitude smaller than the TAM slide says.
watch ~25–30% of the fee ≈ 5–7.5% of first-year salary
Paid creators, clipping and UGC ad ops
Pay-per-view clip campaigns and creator-made ad assets. The take is thin, the cheapness claim breaks for exactly the buyer with money, and the evidence base is the weakest in the atlas.
watch 9–18% on marketplaces; undisclosed and materially larger in managed agencies
Other ways to sell people
Same shape, different supply: red teams, teleoperators, voices, experts, engineers.
Adversarial evals and red-team crowds
Labs pay five and six figures per model to be attacked; the attackers are paid in prize money. The widest spread and the scarcest supply in the sweep.
build Effective take on crowd output >90% [UNVERIFIED — inferred, no revenue disclosure]
Robotics teleoperation and physical-world data
Buyers pay $50–200/hour for demonstrations, operators get $25–50. The rig and the floor space are what make this defensible — and what the commodity end lost.
build 40–65% gross on quality teleop; 70–90% on egocentric [UNVERIFIED] — in a segment whose price has collapsed
Voice, speech and low-resource language data
Contributor pay is documented down to the dollar because recruiting is marketing. Nobody publishes a buy-side price at all — so the take rate here is a guess, and the page says so.
watch 60–85% on spec'd low-resource collection [UNVERIFIED — buy-side price is a working assumption, not a source]
Expert networks
$200 to the expert, $800–900 to the client, held for four decades. The oldest version of this model has the most durable rake in the atlas — and may not sell to AI labs at all.
open ~70–80% — $200/hr expert rate against $800–900/hr client billing
Sales talent: AEs, SDRs and the people who close
GTM hiring looks like engineering hiring and behaves nothing like it: shorter tenure, higher churn, outcome-legible candidates, worse CVs, and a buyer who misses a number the month the seat stays empty.
avoid 18–33% of base salary as a fee; $9–14K at SDR, $40–63K at VP Sales
Forward-deployed engineering
Distyl at $1.8B on ~$31M ARR is the purest test of services-as-software. Palantir's 82% gross margin is the only proven escape from the labour multiple — and it works because the humans install a licence rather than being the product.
watch 33–36% on the marketplace route (Turing, disclosed); undisclosed on the build-it-yourself route
Selling a function, not a body
Where the buyer is purchasing an outcome and does not care who produces it.
Outbound and GTM-as-a-service
Agencies selling pipeline to funded startups. Real margin, instant sales cycle, and a hard price ceiling: the buyer's alternative is hiring one SDR.
crowded ~55–75% gross margin [UNVERIFIED — inferred from two published rate cards]
Design, video and content production
The highest inferred take rate in the atlas attached to the weakest defence: abundant supply, a salary-anchored price, and a buyer alternative that gets cheaper every quarter.
avoid 70–88% [WEAK — supply-side figures are self-reported]; 55–65% on generous assumptions
Compliance, finance and back office
Professional-services firms wearing marketplace clothing. The supply is licensed rather than scarce, the take rate is capped by malpractice risk, and the margin has already migrated to the software layer above.
avoid Low, and structurally capped. The margin sits in software attach and contingency fees, not in the spread on licensed hours
Community, campus and events
Developer distribution sold as an outcome. A genuinely new budget, a supply paid in stipends and merch — and almost no published economics anywhere in the category.
watch 30–50% gross for hackathon agencies [UNVERIFIED]. No ambassador-ops intermediary exists to measure
Selling capacity, not labour
The variant where the scarce thing is a machine, a building or a room full of people.
Compute and capacity brokerage
The largest new budget line in the economy, brokered at the thinnest spread in the atlas — because compute is the one supply here that has a public price index and, from October 2026, futures.
watch Low single-digit percentage [UNVERIFIED — SF Compute's fee schedule is not public and its docs page 404s]
Data-centre labour and site brokerage
A 439,000-worker shortage, a 30% wage premium, and $14.2M a month burning on a delayed 60MW build — against a supply that has no price index and cannot be delivered from a laptop.
watch Conventional construction staffing markups run 30–40%. No AI-native intermediary has published anything