The AI training gold rush just got geographically specific, and $25M says India is the best place to mine it.

The Summary

  • Deccan AI raised $25M to compete with Mercor by building a concentrated expert network in India for AI model training
  • The geographic focus is the strategy: tighter quality control in a market where most competitors spread thin across global freelance pools
  • India gets positioned as infrastructure for the agent economy, not just cheap labor arbitrage

The Signal

Deccan AI is betting that the next constraint in AI development is not compute or algorithms, but human judgment at scale. The company is building a managed network of domain experts in India to feed the insatiable appetite for training data that actually teaches models something useful. This is the unsexy middle layer of the agent economy: the humans teaching machines to think before those machines can work autonomously.

The India concentration is tactical, not just cost-driven. While competitors like Mercor cast wide nets for global talent, Deccan is building vertical integration in one geography. This matters because AI training quality has become a bottleneck. When you're paying for expert feedback on medical reasoning or legal analysis, you need consistency. Random Upwork contractors reviewing edge cases at 3am in different time zones creates noise, not signal. A concentrated workforce means standardized processes, coordinated quality checks, and the ability to iterate fast when training protocols change.

The $25M bet also signals where the AI infrastructure stack is really heading. Foundation models are commoditizing. The value is moving to whoever can create the highest-quality training loops. That requires domain expertise, cultural context, and enough scale to matter to companies training frontier models. India has the talent density in technical fields, the English proficiency for nuanced feedback, and the timezone alignment with both US and Asian AI labs.

This is not about replacing knowledge workers. It's about creating a new category: the professional AI trainer. Someone who understands both the domain (law, medicine, engineering) and how to translate that into machine-readable patterns. Deccan is industrializing that role in one place rather than hoping the gig economy produces it organically.

The Implication

Watch for more geographic clustering in AI training infrastructure. The global talent marketplace was the Web2 story. Web4 needs concentrated expertise networks that can deliver consistent quality at speed. If you're building AI products, your training data pipeline matters as much as your model architecture. And if you're a domain expert in a knowledge-dense region, there's a new premium on your ability to teach machines, not just do the work yourself.

Source: TechCrunch AI