Mercor competitor Deccan AI raises $25M, sources experts from India

In a significant boost for the artificial intelligence sector, Deccan AI has successfully raised $25 million in its initial funding round. This startup, which specializes in providing post-training data and evaluation services, primarily utilizes a skilled workforce based in India to meet the growing demand for refining AI models. The funding round, which was an all-equity Series A, saw A91 Partners leading the investment, with notable participation from Susquehanna International Group and Prosus Ventures. As leading AI laboratories like OpenAI and Anthropic focus on developing core models internally, the need for outsourcing post-training work—ranging from data generation to evaluation and reinforcement learning—becomes increasingly apparent. Deccan AI is emerging as a key player in this evolving market. Founded in October 2024, Deccan AI offers a variety of services aimed at enhancing coding and agent capabilities of AI models, as well as facilitating interaction with application programming interfaces (APIs). This helps integrate AI systems with existing software infrastructures. The startup collaborates with advanced AI labs to provide expert feedback, conduct evaluations, and create reinforcement learning environments. Additionally, Deccan AI serves businesses with its evaluation suite, Helix, and an operations automation platform. With the advent of advanced AI models that extend beyond text to include physical environments—such as robotics and vision systems—Deccan AI is well-positioned to address this shift. Current clients include prominent organizations like Google DeepMind and Snowflake, with founder Rukesh Reddy indicating that the company is managing around 10 clients and dozens of active projects at any time. Headquartered in the San Francisco Bay Area and with a substantial operations team in Hyderabad, Deccan AI employs approximately 125 individuals. The company has built a network of over 1 million contributors, which includes students, domain experts, and PhDs, with typically 5,000 to 10,000 contributors active each month. Notably, around 10% of these contributors hold advanced degrees, with the percentage climbing higher for those engaged in specific projects. As the market for AI training services rapidly expands, competition intensifies among companies like Meta-owned Scale AI and Surge AI, as well as startups such as Turing and Mercor. These firms are all vying to deliver data labeling and evaluation services. Reddy emphasized the critical nature of quality in post-training processes, underscoring that tolerance for errors is minimal, as mistakes can have dire consequences for model performance in real-world applications. The urgency of the sector is marked by the need for large volumes of high-quality data, often required on tight deadlines, complicating the balance between speed and accuracy. Despite concerns regarding working conditions and pay for gig workers in this field, Reddy noted that contributors on Deccan’s platform can earn between $10 to $700 per hour, with top earners reaching up to $7,000 monthly. While Deccan’s primary customer base is comprised of U.S.-based AI labs, a significant portion of its contributors are sourced from India. Reddy explained that the decision to centralize operations in India aids in maintaining high quality. "Many of our competitors source from over 100 countries, but focusing on one location simplifies quality control," he stated. This strategy emphasizes India’s role in the global AI landscape as a provider of skilled talent and training data, even as the development of frontier models remains dominated by a few companies in the U.S. and China. However, Deccan is also beginning to tap into talent from other regions, including the U.S., to bring niche expertise into its projects. Positioning itself as a 'born GenAI' company, Deccan AI has concentrated on higher-skill tasks since its inception, distinguishing itself from traditional data-labeling firms that initially focused on simpler computer vision projects. With a remarkable tenfold growth over the past year, the company is now on a trajectory to achieve a double-digit million-dollar revenue run rate, although Reddy opted not to disclose specific figures. Approximately 80% of the revenue is generated from its top five clients, highlighting the concentrated nature of the frontier AI market.