Deccan AI, a startup providing post-training data and evaluation services for frontier AI labs, has raised $25 million in a Series A round led by A91 Partners, with participation from Susquehanna International Group and Prosus Ventures.
The company works with customers including Google DeepMind and Snowflake on tasks ranging from expert feedback generation and reinforcement learning environment construction to evaluation and API tool-use training for AI models, according to TechCrunch. Deccan has onboarded about 10 customers and runs roughly two dozen active projects at any given time, founder Rukesh Reddy told TechCrunch.
The India Concentration Bet
Where competitors like Mercor and Turing source contractors from 100+ countries, Deccan has deliberately concentrated most of its contributor base in India. “If you have operations in just one country, it becomes far easier to maintain quality,” Reddy told TechCrunch.
The strategy is built on a network of more than one million contributors, including students, domain experts, and PhDs, with 5,000 to 10,000 active in a typical month. Around 10% of the contributor base holds advanced degrees, though the share runs higher among active contributors depending on project requirements. Entrackr reports the company is opening a new Bengaluru office focused on enterprise growth, alongside existing operations in San Francisco and Hyderabad.
Contributor earnings range from about $10 to $700 per hour, with top contributors earning up to $7,000 per month, according to TechCrunch. The AI training data sector has faced persistent criticism over working conditions and pay for gig workers used to generate training data.
A Crowded and Growing Market
Deccan enters a market already populated by well-funded players. Scale AI, which received a significant investment from Meta, remains a dominant force in data labeling. Mercor was eyeing a $10 billion valuation on a $450 million revenue run rate as of September 2025. Turing raised $111 million at a $2.2 billion valuation last year as a key coding data provider for OpenAI. The Financial Times has reported on the rapid expansion of the broader AI training services market alongside the rise of large language models.
Deccan’s pitch is differentiation through specialization. Founded in October 2024, the company describes itself as “born GenAI,” built for higher-skill post-training work from day one rather than migrating from legacy computer vision labeling. The company grew 10x over the past year and is now at a “double-digit million-dollar revenue run rate,” per Reddy, though he declined to share specifics. About 80% of revenue comes from the top five customers, reflecting how concentrated the frontier AI market remains on the buyer side.
Why It Matters
The post-training layer is where raw model capability turns into reliable real-world performance, and the tolerance for errors is close to zero. “Quality remains an unsolved problem,” Reddy told TechCrunch, noting that mistakes in post-training can directly degrade model performance in production. As frontier labs and enterprises race to deploy AI agents that interact with APIs, tools, and physical environments, the demand for high-quality, domain-specific post-training data is growing faster than the supply of people qualified to produce it.
Deccan’s $25 million is modest compared to the billions flowing into the model layer, but it funds the human infrastructure that makes those models usable. Whether concentrating that infrastructure in a single country creates a quality advantage or a geographic risk is the bet investors are making.