# What Does Mercor Actually Do That Makes It Worth $10 Billion?

Mercor is not an AI model company. It does not build GPT-6 or Claude 4. According to Bloomberg reporter Tom Foster, Mercor "hires high-skilled workers — lawyers, accountants, software engineers — and pays them to train AI systems to do their jobs." The company acts as a specialized labor marketplace, connecting AI labs with professionals who can generate the high-quality training data that frontier models require. According to Bloomberg's earlier reporting from October 2025, Mercor had already reached a $2 billion valuation before this latest round. The jump to $10 billion reflects the AI industry's insatiable hunger for expert-labeled data, particularly as models approach the limits of internet-scale training. Foster noted on the podcast that "the biggest AI companies are running out of data to train on, and Mercor has figured out how to supply it at scale." My take: This valuation makes sense only if you believe that human expert data will remain the binding constraint on AI progress for the next several years. If Mercor can maintain its position as the primary broker, it becomes a toll booth on the highway to AGI. But the moment AI models can generate their own training data reliably — a capability that OpenAI and Anthropic are actively pursuing — Mercor's moat evaporates.

# Who Are Mercor's Founders and Why Should We Care?

The founders — Brendan Foody, Adarsh Hiremath, and Surya Midha — are college dropouts who started Mercor while still in their teens. Foster described them on the podcast as "part of a new generation of founders who see AI not as a product to build but as a labor problem to solve." This is a critical distinction. The founders' lack of a traditional tech pedigree is framed as an advantage: they don't carry assumptions about how work should be organized. According to Foster, the founders "built Mercor out of a college dorm room, initially running it as a side project to help students find freelance work." The pivot to AI training data came when they realized that the same platform could serve the exploding demand from AI labs. The company now claims to have "tens of thousands of workers" on its platform, though exact numbers were not disclosed. My take: The college-dropout narrative is convenient but misleading. What matters is that these founders recognized a structural arbitrage: AI companies need expert labor but don't want to manage it. Mercor inserts itself as the middleman, capturing the spread between what AI labs pay and what workers receive. This is a classic platform play, not a technology breakthrough.

# Is Mercor a Labor Platform or a Replacement Factory?

The central tension in Mercor's business model is that it profits from both sides of a transaction that ultimately eliminates one side. Foster captured this neatly: "The workers are training AI to do their jobs. That's the whole point." But this creates a paradox: the better the training, the less demand for the workers providing it. According to Foster, Mercor's workers are "paid well by gig economy standards — often $30 to $50 an hour for specialized work" — but they are explicitly told that their output will be used to automate their own roles. This transparency, if it exists, is unusual. Most AI training data is collected without workers fully understanding its end use. The comparison to earlier platforms is instructive:

Dimension	Mercor	Amazon Mechanical Turk	Scale AI	Upwork
Worker skill level	High (professionals)	Low (crowd workers)	Medium (labelers)	Variable (freelancers)
Primary customer	AI labs	Researchers	Autonomous vehicle companies	Enterprises
Valuation	$10B (2026)	Acquired by Amazon	$14B (2024 peak)	~$2B (public)
Worker awareness of AI replacement	Explicit (per Bloomberg)	Implicit	Implicit	Not relevant
Business model risk	AI self-training eliminates need for humans	Low (tasks are AI-resistant)	Moderate (automation of labeling)	Low (project-based)
Verdict	Highest upside, highest existential risk	Low margin, stable	High margin, volatile	Low growth, stable

My take: Mercor's valuation premium over Scale AI ($10B vs $14B at peak) is striking given that Scale has a longer track record and more diversified revenue. The market is betting that Mercor's focus on high-skill labor will be more valuable than Scale's lower-skill labeling. I think this bet is correct in the short term but fragile: the moment a frontier model can generate its own expert-quality training data, Mercor's entire value proposition collapses.

# Who Loses If Mercor Succeeds?

The most obvious losers are the professionals who train the AI. Foster noted that "a lawyer earning $50 an hour on Mercor today may find that the AI they trained can do their work in seconds by next year." This is not a hypothetical; it is the explicit business model. The workers are being paid to accelerate their own obsolescence. But there are institutional losers too. Law firms, accounting practices, and consulting firms that rely on junior talent to do the work that Mercor's AI training targets will find their talent pipeline disrupted. If AI can handle the work of a first-year associate or junior analyst, the traditional apprenticeship model breaks down. According to Foster, "the biggest AI labs are already Mercor's customers," though he did not name them specifically. This suggests that OpenAI, Anthropic, and Google DeepMind are all leveraging Mercor's workforce. The concentration of power is concerning: a handful of AI labs control the frontier models, and Mercor controls the human data supply that feeds them. My take: The real loser may be the broader labor market. Mercor creates a race to the bottom where professionals compete to train their replacements, driving down the value of expertise while enriching the platform middleman. This is a classic "tragedy of the commons" scenario, and regulators are not paying attention.

# Predictions

1. Mercor will reach a $25 billion valuation by Q4 2027 before declining sharply as synthetic data methods mature, wiping out at least 40% of its value by Q2 2028. 2. OpenAI will announce a synthetic data breakthrough by mid-2027 that reduces its need for human expert training data by 60%, triggering a revaluation of all data-brokerage startups. 3. Regulators in the EU will investigate Mercor's labor practices by 2028, specifically the transparency of informing workers they are training their replacements, potentially forcing business model changes.

# Timeline

• 2023 — Mercor founded by Brendan Foody, Adarsh Hiremath, and Surya Midha as a college freelance platform.
• 2024 — Pivot to AI training data; early contracts with unnamed AI labs.
• October 2025 — Bloomberg reports Mercor reaches $2 billion valuation.
• April 2026 — Mercor reaches $10 billion valuation; Bloomberg Big Take podcast features the story.
• Expected 2027-2028 — Synthetic data breakthroughs reduce human data dependence; Mercor valuation corrects.

# Article Summary

• Mercor's $10 billion valuation is not about technology but about controlling the human labor supply for AI training data — a structurally fragile position.
• The company's business model creates a direct conflict of interest where professionals are paid to accelerate their own replacement, a dynamic regulators have not addressed.
• The comparison to Scale AI and Mechanical Turk reveals that Mercor's high-skill focus commands a premium but carries higher existential risk from self-training AI.
• Bloomberg's reporting confirms that AI labs are desperate for expert data, but the same labs are also investing heavily in synthetic data alternatives that could make Mercor obsolete.
• The real winners are the founders and early investors who are cashing out before the synthetic data reckoning arrives, likely within 24 months.