Mercor Closes $200 Million Round at $2 Billion Valuation to Build Data-Labeling Pipelines for Nvidia's Open Models

Mercor, the AI-data-labeling startup founded in 2023 by former OpenAI and Scale AI employees, has closed a $200 million Series C at a $2 billion post-money valuation, the company confirmed on August 22. The round was led by Victor Liang at Felicis, with participation from NVIDIA's venture arm, General Catalyst and Bain Capital Ventures. Cumulative funding now stands at $520 million, and the company said it would use the proceeds to triple its contractor headcount to 1,500 by the end of 2026 and to expand beyond its current customer base of OpenAI, Anthropic, and the AI-coding startup Cursor.

The Strategic Angle

Mercor's new customer is NVIDIA, which is building open-weight foundation models under the Nemotron brand and needs labeled training data at a scale that Scale AI's previous dominant position cannot deliver. Scale AI's market share has dropped from 67% in 2024 to 39% in Q2 2026, according to the Ramp AI Index, following the July departure of key customers including Microsoft and Google. NVIDIA's investment is a direct bet that Mercor can fill that gap — and that the data-labeling category itself is large enough to support two $2 billion-plus players.

The Headcount Race

Data labeling has traditionally been a low-margin, labor-intensive business, but Mercor's pitch is that its platform routes work through a curated network of 30,000 vetted contractors and uses AI to pre-label 70% of the work before human review. The result is a cost per labeled example of roughly $0.08, versus $0.32 for Scale AI's full-service tier. The new round will fund 1,500 full-time employees focused on three specialty areas: code-execution traces for AI coding models, multi-modal video annotation, and synthetic-data generation for safety fine-tuning.

Why the Valuation Held

Mercor's revenue run-rate grew from $60 million at the start of 2026 to $290 million by August, according to a company statement, implying a roughly 33x multiple on current annualized revenue — aggressive even by AI-infrastructure standards. Investors accepted the multiple because the business is recurring-revenue dominant: 78% of Q2 revenue came from multi-year contracts with frontier model labs, and the average contract length is 22 months. Felicis's Liang said the firm sized the round "to ensure Mercor has the runway to get to default profitability without a fourth round."

Competitive Landscape

Three players now dominate the enterprise data-labeling market: Scale AI (last valued at $13.8 billion in 2024), Surge AI (private, no public valuation), and Mercor. Surge has held its enterprise position with research-focused labs including Anthropic and DeepMind, but has not pursued the NVIDIA-style infrastructure customer. The fourth credible challenger is Amazon's Mechanical Turk, which launched a "MTurk Pro" tier in March 2026 targeting the same volume tier as Mercor at a roughly 25% lower price point. Mercor's investors are betting that the company's full-stack platform — pre-labeling, contractor management, quality scoring and dataset versioning — creates enough switching cost to defend against Amazon's price war.

What to Watch Through Year-End

Three checkpoints follow. Mercor's Q4 launch of "Mercor Code," a packaged dataset of 12 million verified code-execution traces curated for training AI coding models, will be the first test of whether the company can build a product-grade data asset rather than just a labeling service. The Scale AI Series F round, reportedly in market at a $20 billion valuation, will reveal whether the leader is recovering or conceding the high end of the market. And the open-weight model release cadence from NVIDIA — Nemotron-4 is expected in October — will determine whether the data-labeling category grows with the model count or plateaus as labs shift toward synthetic-only training pipelines.