Mercor’s meteoric rise exposes AI data supply risks

Mercor's revenue trajectory has been nothing short of meteoric, doubling from $1 billion to $2 billion in just four months during 2026, a pace that eclipses typical startup growth and underscores the explosive expansion of the AI training market. This rapid scaling is reflected in its first-half 2026 revenue of $614 million, already surpassing total 2025 revenue, and projects an annualized run rate expected to reach $2.8 billion by year-end, positioning Mercor as a dominant force in the AI data services ecosystem.

The company’s soaring valuation—from $10 billion in late 2025 to whispers of $20 billion in mid-2026—mirrors investor confidence in Mercor’s unique business model that leverages domain experts like scientists, lawyers, and doctors to provide high-quality training and evaluation services for AI giants such as OpenAI, Anthropic, and Meta. This human-in-the-loop approach not only validates the vast and fast-growing centralized market for AI training infrastructure but also sets a financial benchmark for decentralized AI labor protocols, with contractor rates averaging $105 per hour.

Mercor’s rapid revenue growth is emblematic of a broader structural surge in the AI startup sector, driven less by isolated company strategies and more by the massive influx of capital and labor dedicated to AI model training. As analysts note, Mercor’s trajectory signals that the market for human-in-the-loop AI services is not speculative but a real, fast-scaling industry where leading AI labs are willing to spend aggressively to secure quality training data, highlighting the critical bottleneck in AI development that Mercor is uniquely positioned to address.

Mercor's revenue is overwhelmingly dependent on a narrow client base, with over 90%—approximately $553 million of its $614 million H1 2026 revenue—sourced from just three AI foundation model giants: OpenAI, Anthropic, and Google DeepMind. This extreme concentration underscores a significant vulnerability, as any shift in these clients' procurement strategies could materially impact Mercor's financial health.

The risks tied to this client concentration are not hypothetical. For instance, Meta's suspension of cooperation following a March supply chain attack vividly illustrates how security incidents can abruptly disrupt Mercor's revenue streams, emphasizing the fragility of relying heavily on a handful of major clients.

Recognizing these risks, Mercor is proactively pursuing diversification, aiming to reduce foundation model companies’ share of its revenue to less than one-third by 2030. This strategic pivot involves expanding its client base to include AI startups and Fortune 2000 companies, signaling a deliberate effort to mitigate concentration risk and stabilize long-term growth.

Mercor’s client concentration also highlights a broader industry vulnerability that decentralized platforms seek to address by distributing human labor coordination without centralized intermediaries. This emerging competition underscores the strategic imperative for Mercor to broaden its client portfolio to maintain its dominant position amid evolving market dynamics.

By early 2026, Mercor had scaled its global contractor workforce to roughly 30,000 domain experts, logging over 2.5 million expert hours in a single quarter—equivalent to about 4,800 full-time workers—who perform critical AI model training tasks such as data labeling, model evaluation, and reinforcement learning from human feedback. This massive scale underscores the operational complexity of coordinating a diverse labor pool spanning specialties from Chemistry and Biology to CUDA programming and law, all contributing to the nuanced demands of AI training.

Central to managing this sprawling network is Mercor’s sophisticated use of Slack, which serves as the backbone for rapid project staffing, seamless communication, and secure access control. Automation through Slack’s APIs and Mercor’s proprietary Team Platform has transformed what once took days of manual coordination into near real-time operations, enabling client requests to become fully staffed workspaces within hours while instantly revoking access upon contract completion to mitigate data security risks.

The nature of AI model training gig work represents a paradigm shift in knowledge work, focusing on enabling AI systems rather than traditional human-to-human deliverables. As founder Brendan Foody remarked, 'Instead of the investment bankers doing the analysis, they’ll build RL environments and train agents... within five years, a majority of high-end knowledge workers are training models.' However, despite the intellectual rigor involved, there is skepticism about sustaining long-term worker motivation given the abstract and indirect nature of these tasks.

Mercor’s operational model highlights the persistent challenge of coordinating distributed human labor at scale, a bottleneck that decentralized protocols aim to address by reducing fees and increasing transparency. With contractors earning an average of $105 per hour and daily payouts exceeding $2 million, the platform sets a high pricing benchmark in the AI training gig economy, inviting competition from decentralized labor platforms that promise comparable quality with lower margins, potentially reshaping how AI labor is sourced and managed.

Mercor's labor model embodies a paradox where highly credentialed contractors—ranging from Hebrew-speaking voice actors to physicists and physicians—train AI systems to replicate their expert judgment, effectively accelerating their own obsolescence. As Carolina Perez Sands observed, her corrections became redundant within a week as the AI absorbed them, underscoring a built-in obsolescence that challenges traditional economic security based on skill scarcity, a concern highlighted by MIT Sloan's Danielle Li.

To transcend the limitations of simple data labeling and address the accelerating displacement of expert contractors, Mercor has strategically acquired startups like Deeptune to simulate entire workplaces, aiming to automate complex human workflows. This move, described by Mercor’s founder as solving 'the bottleneck to a frontier lab automating everything that people do,' signals a push toward deeper AI training sophistication and scalability amid the paradoxical labor dynamics.

Mercor’s massive contractor network—(https://www.google.com/goto?url=CAESeQHuR6pNYCj7AoCTuYbMj1v6lUTZTveYyQnLefL5Cchh6sCQvUFvqlwaF6TZ2FZlCwuc4yayTbB4ks47EjOkNcJ2Yq6xQctGKW3D6RvBM5jDAyP-S28mHPnAADfJfp_YNQS_Y0cQ5XpSdEMhP0MG0pOSRB9S5Hv_-dA%3D) strong and paid an average of $105 per hour with daily payouts exceeding $2 million—presents formidable operational and security challenges, exemplified by a supply chain attack that led Meta to suspend cooperation in March 2026. Managing insider threats and securing this sprawling workforce is critical as Mercor coordinates labor that directly contributes to AI models potentially displacing those very contractors, complicating labor relations and sustainability.

Facing over 90% of its $614 million revenue concentrated among a handful of AI foundation model companies like OpenAI, Mercor is actively diversifying its client base to mitigate concentration risk and ensure sustainable growth. By courting financial institutions such as Ramp, Blue Owl, and Citigroup, Mercor anticipates that by 2028, revenue from AI startups and Fortune 2000 firms will surpass that from foundation model clients, a strategic pivot essential to scaling its contractor network and improving gross margins beyond the current 33%.