The Twenty Minute VCMercor CEO & Co-Founder, Brendan Foody: How They Grew from $1M to $500M in 17 Months
CHAPTERS
- 0:00 – 1:31
Mercor’s record growth and the RL environments thesis
Brendan opens with headline metrics: Mercor’s rapid climb from $1M to a $500M revenue run rate, premium pay rates, and surging demand. He frames the company’s work as enabling RL environments that could eventually automate vast swaths of repetitive economic activity.
- •Grew from $1M to $500M revenue run rate in 17 months (record pace)
- •Average marketplace pay rate of ~$95/hr vs. ~$30/hr at key rivals
- •Demand is high enough to potentially “double overnight” if capacity allows
- •Belief that RL environments will expand until they touch most of the economy
- •Core motivation: removing monotonous, redundant human work
- 1:31 – 4:34
Childhood hustles: the donut arbitrage that worried his mom
Brendan recounts early entrepreneurship selling donuts at school, competing on price, and skirting school rules by moving just off campus. The story culminates in his mother worrying the hustle mindset could drift into trouble, pushing him toward Catholic school.
- •Bought donuts cheaply and resold individually for a large markup
- •Responded to competition by temporarily underpricing to win back demand
- •Navigated school enforcement by relocating off-campus
- •Mother feared a slippery slope from hustles to illicit selling
- •Catholic high school decision shaped his path and relationships
- 4:34 – 7:22
College skepticism and the sneaker-reseller AWS-credit business
He explains how a high-school consulting agency helping sneaker resellers claim AWS credits earned him hundreds of thousands of dollars. That success made college feel economically irrational, though he ultimately applied late to appease his parents.
- •Built a consulting agency around startup promos (AWS credits) for resellers
- •Earned significant income before college age
- •Viewed college as a poor ROI compared to building full-time
- •Parents pushed back; he applied to colleges at the last minute
- •Advice: education is increasingly accessible online; college is mainly social value
- 7:22 – 9:26
Not a ‘body shop’: why frontier labs now need vetted experts
Harry challenges the data-labeling market as a commodity body shop; Brendan argues the opposite, positioning Mercor as a research partner. He describes the shift from crowdsourced low-skill tasks to sourcing and vetting elite professionals to build and interpret high-complexity training data.
- •Rejects “body shop” framing; emphasizes deep research partnership
- •Scale used Mercor to hire thousands, revealing market transition
- •Move from crowdsourcing to expert sourcing/vetting (bankers, engineers, doctors, lawyers)
- •High-complexity work requires domain experts because researchers can’t interpret everything
- •This shift catalyzed Mercor’s growth
- 9:26 – 12:17
Why ‘running out of smart people’ isn’t the limiter
Harry argues the expert supply narrows as models improve; Brendan counters that the market is bounded by what humans can still do better than models. He uses an RL environment example where as models improved, new complexity layers reopened contribution from a broader set of humans.
- •TAM depends on the frontier gap: what humans do better than models
- •As models get better on one task, new tools/workflows create fresh evaluation needs
- •Example: tool-using RL tasks went from 100 contributors → 20 → back to many as complexity expanded
- •Humans remain essential as verifiers and rubric creators
- •Belief: models aren’t plateauing, but improvement methods are changing
- 12:17 – 14:56
Differentiation vs. Turing/Handshake/etc.: power laws in contributor impact
Brendan explains how many marketplaces copied the “sourcing and vetting” narrative, but outcomes are power-law distributed. Mercor’s edge is finding the top 10–20% of contributors and matching them to the right tasks, plus measuring quality with internal models.
- •Competitors followed the same positioning once the market shifted
- •Model improvements often come disproportionately from top contributors (power law)
- •Moat: proprietary expert access/referrals + matching infrastructure
- •Uses models/algorithms to assess and improve data quality
- •Positioning: intersection of labor marketplace and AI research
- 14:56 – 18:29
Vendor consolidation, customer allocation, and revenue concentration realities
They discuss whether labs intentionally multi-vendor to avoid dependency and how spend evolves over time. Brendan expects early multi-vendor behavior to consolidate toward the highest-outcome partner, and he’s candid that Mercor’s revenue concentration looks similar to NVIDIA’s large-customer exposure.
- •Labs may spread spend early, but ultimately optimize for model performance gains
- •Over time, vendor sets consolidate due to economies of scale and fixed-cost advantages
- •Fragmented markets tend to consolidate as hype cools
- •Mercor’s largest-customer concentration is “similar to NVIDIA” (no exact % shared)
- •Argument: concentration can be acceptable when customers are the best in the world
- 18:29 – 20:21
Post-Scale acquisition shockwave and Mercor’s 1→500 milestone
Brendan describes demand acceleration after Scale’s acquisition and the public confusion around numbers. He confirms Mercor’s “1 to 500” run-rate growth in 17 months and says growth is still accelerating even at the new scale.
- •Scale acquisition triggered a meaningful acceleration in inbound demand
- •Mercor was already at a nine-figure run rate before the surge
- •Public reporting may be incomplete; company will share updated metrics
- •Fastest 1→500 run-rate growth in 17 months (per Brendan)
- •Key driver: being deeply embedded with frontier labs before the inflection
- 20:21 – 22:07
Quality as strategy: paying experts more and treating supply as the product
Harry relays industry critique of Scale quality; Brendan agrees Scale lost product focus and links quality to how workers are treated. He argues Mercor’s higher pay drives better talent, better referrals, and ultimately better model progress.
- •View that Scale’s quality issues were broadly understood in the ecosystem
- •Quality is tied to internal obsession and how contributors are treated
- •Mercor pays materially higher (~$95/hr) to attract and retain top talent
- •Better treatment improves referrals and raises the caliber of data produced
- •Capacity constraint: high-quality human supply is a strategic bottleneck
- 22:07 – 24:08
Synthetic data vs. humans: why experts still matter for frontier progress
Brendan argues synthetic augmentation will help, but cannot replace humans when pushing new capabilities. Until models surpass humans at everything, humans provide the “stasis point” needed to define, verify, and measure progress in real workflows.
- •Synthetic data can augment efficiency, but frontier tasks still need human grounding
- •Training requires human-verifiable targets for what models can’t yet do
- •Superintelligence would change this, but he believes it’s far off
- •Models excel at narrow benchmarks yet fail at practical tool-using workflows
- •Automating the economy requires humans to build evals for workflows and tools
- 24:08 – 26:41
Benchmarks are ‘BS’: building evals that map to real enterprise outcomes
Brendan criticizes academic-style leaderboards as disconnected from what customers want. He advocates evals based on real job workflows, tool use, and rubric-based grading to close the real-to-sim gap and make AI actually useful in organizations.
- •Humanity’s Last Exam/Olympiad-style benchmarks don’t measure practical value
- •Enterprises care about outputs like banking models, consulting docs, and software delivery
- •Right approach: evals that reflect real distributions of tasks and tool usage
- •Use rubrics (like grading) for workflow quality, not just “right answers”
- •“Eval is the PRD”: avoid ‘vibe spending’ by defining measurable success criteria
- 26:41 – 31:56
Raising at ‘punchy’ valuations: thinking in possibilities, not comps
They unpack Mercor’s fundraising logic and why high revenue multiples made sense given momentum and customer pull. Brendan shares revenue run-rate snapshots at key rounds and notes the business is now profitable enough to be funding-optional.
- •Valuation should reflect what’s possible, not just comps and multiples
- •Seed: ~$2M run rate at ~$250M valuation (~100x+)
- •Series B: ~$20M run rate at term sheet, also ~100x multiple
- •Now ~25x larger than Series B run rate and strongly profitable
- •Financing is optional, but inbound interest is high
- 31:56 – 35:34
Stay private or go public? Capital abundance, signaling, and froth
Brendan explains why he hasn’t prioritized IPO planning and cites advice from Jack Dorsey to stay private to preserve long-term focus. They debate whether private markets have too much cash, the risk of funding weak competitors, and how to think about ‘froth’ on different time horizons.
- •IPO discussed as plausible at scale, but Brendan prefers long-term private flexibility
- •Jack Dorsey’s advice: staying private reduces quarterly short-termism
- •Private capital is abundant; may be excessive from an investor perspective
- •Overfunding can support low-quality competitors in hot markets
- •Heuristic: overestimated short-term, underestimated long-term (dot-com analogy)
- 35:34 – 38:57
AI business durability: retention, margins, and switching costs
Brendan gives an investor lens on separating real AI revenue from hype: retention and successful deployments matter more than flashy top-line growth. On margins, he argues context matters—subsidies can be rational if they buy durable LTV, but dangerous in low-switching-cost markets.
- •Assess sustainability via retention and evidence pilots are converting to real value
- •If most pilots fail, revenue is likely fragile
- •Margins matter, but can be strategically sacrificed if efficiency will improve quickly
- •Risk zone: heavy subsidies in commoditized markets with low switching costs
- •Mercor emphasizes capital efficiency with positive gross and net margins
- 38:57 – 47:10
Model landscape bets, talent economics, and building a ‘missionary’ culture
They cover which segments feel overhyped, why coding tools are valuable yet competitive, and why Brendan expects more engineers—not fewer—due to elasticity of software demand. He also discusses foundation-model concentration due to capex, AI talent costs, and the importance of purpose plus equity to retain committed teams.
- •Coding assistants deliver real utility but have low switching costs today
- •Expect more engineers over time as productivity increases expand scope of software built
- •Belief: biggest model builders largely already exist due to massive compute/data/talent costs
- •AI talent economics are extreme; purpose + upside helps retain long-term builders
- •Missionaries vs. mercenaries framing; Zuck’s spending may still work given team quality
- 47:10 – 1:00:35
996 debate, fear vs. aggression, and closing with quick-fire predictions
Brendan clarifies Mercor never mandated hours; early intensity was organic and output-driven. They discuss whether to burn capital aggressively to outcompete, supply constraints limiting growth, and then end with quick-fire views on superintelligence timelines, OpenAI strategy, investor wish-list, and Mercor’s growing share in RL environments.
- •996 was descriptive of early intensity, not a mandate; focus has shifted to impact/output
- •Founder tension: capital efficiency vs. aggressive subsidy strategies to win markets
- •Biggest current bottleneck is scaling capacity; they turn down projects daily
- •Quick fire: superintelligence in 3 years is ‘wrong’; OpenAI should focus on customization
- •RL environments share estimate ~50–60% and belief they can ‘subsume the economy’
