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

Deccan AI thinks the post-training industry has been doing it wrong. Most competitors cast their nets across 100 or more countries to find AI evaluators and trainers. Deccan is betting that concentrating operations in a single country — India, with a large in-house team in Hyderabad — produces more consistent quality. The company, founded in October 2024 and now at a $25 million Series A, is making a narrow, testable claim: geographic focus solves a problem that breadth has not.

The pitch deserves attention because the problem is real. Post-training quality control remains genuinely unsolved, and tolerance for errors in the work is close to zero, as founder Rukesh Reddy told TechCrunch. "Getting an agent to work in a demo is one thing; getting it to handle high-stakes business logic is another," he said. The companies doing this work — Mercor, Scale AI, Surge AI — have raised hundreds of millions and are still working through it.

Deccan, led by A91 Partners with participation from Susquehanna International Group and Prosus Ventures, is 14 months old and relatively small: about 125 employees and a network of more than 1 million contributors, per TechCrunch. The company has onboarded roughly 10 customers and runs a couple dozen active projects at any given time, with Google DeepMind and Snowflake as named clients. Roughly 80 percent of revenue comes from the top five customers. Deccan grew 10x over the past year to a double-digit million-dollar revenue run rate, the report said.

That concentration is worth noting. A handful of anchor customers carrying most of the revenue is a dependency, not a signal of diversification — something Deccan will need to address as it scales. DeepMind as a customer is meaningful credibility; the search giant does not lack for options in this market. But the number to watch is the revenue concentration, not the growth rate.

The geographic bet is where Deccan departs most sharply from industry norms. Hyderabad has become a significant hub for AI annotation and evaluation work as global tech companies offshored more of that labor. Deccan is not competing on cost alone — earnings on its platform range from roughly $10 to $700 per hour, with top contributors earning up to $7,000 a month, suggesting the work involves domain expertise that commands real rates. The range itself is informative: it reflects a spectrum from basic annotation to complex evaluation tasks, and the upper end signals Deccan is competing for experts who have options elsewhere.

Kaushik Anand, a partner at A91 Partners, which led the round, framed the bet broadly. "Deccan is building the essential infrastructure for the next decade of software," he told Economic Times. A91 Partners was founded in 2018 by former Sequoia Capital India (now Peak XV Partners) partners Abhay Pandey, VT Bharadwaj, and Gautam Mago — investors who know the Indian talent market well.

Reddy brings his own credentials: an MBA from the Indian Institute of Management Ahmedabad and an undergraduate degree from IIT Bombay, with prior stints at Citi and Monitor Deloitte. Deccan offers two products — Helix, an evaluation suite, and EnterpriseOS, an operations automation platform — that position it as a tools provider as much as a labor marketplace.

Mercor, the most prominent competitor, has raised $492 million over four rounds from 17 investors, per Tracxn. It is the obvious comparison point, and the one Deccan is most directly arguing against. Whether geographic concentration produces meaningfully better results than breadth is the question Deccan is asking, and the data to answer it is not public.

The $25 million round is small relative to what Mercor and others have raised, which reflects the stage. Deccan is 14 months old and still proving its model works outside a demo environment. The geographic focus is a coherent theory, backed by a coherent explanation. Whether it holds when the contributor network expands to tens of thousands of workers across different domains — and whether DeepMind and Snowflake remain references or become a foundation — is what the next 12 months should answer.

Sources: TechCrunch | Economic Times | Tracxn | The Company Check