India's AI Moment: What's Working, What Isn't, and What's Coming Next

I have been building AI products in Ahmedabad for five years. The honest assessments of Indian AI that I find most useful are the ones that avoid both the boosterism ("India is becoming an AI superpower") and the dismissal ("India is just an outsourcing market"). The reality is more specific and more interesting than either narrative. Here is mine, as of mid-2026.

What Is Actually Working

Vernacular AI is the category where India has a structural advantage that is not easily replicated elsewhere. The pressure to support Hindi, Tamil, Telugu, Bengali, Gujarati, and other languages at production quality has forced Indian AI builders to solve multilingual problems that Western builders ignore. The models and infrastructure built for this — including tools we built for SmartON and MIRA — are now genuinely good and represent a capability gap that Indian AI companies can exploit in markets with similar linguistic diversity globally.

Fintech AI is working, driven by the GST compliance requirement that created a massive forced-automation market. XwFin is one product in a category that has seen genuine adoption at scale. AI for GST, invoicing, and reconciliation is a category that did not exist in meaningful form three years ago and now has substantial revenue and user bases.

Assistive technology AI has progressed significantly. The combination of better multilingual models, cheaper edge inference hardware, and increased awareness of accessibility needs has produced real products — SmartON being one — that serve populations that global AI companies have not prioritised. This is a pattern that will extend: India's diversity of constraints is producing AI solutions for populations and problems the global market ignores.

What Is Not Working

Enterprise AI adoption is lagging significantly behind the startup narrative. Large Indian corporations — outside of IT services companies that sell AI to global customers — are adopting AI slowly. The risk aversion is understandable but the lag is real. Companies that have been "piloting" AI for two years without production deployments are not going to catch up quickly, and the gap between AI-native competitors (many of them smaller and faster-moving) is widening.

Talent cost inflation is a genuine problem. AI engineer salaries in India have risen dramatically over the past two years, driven partly by global remote-first hiring and partly by domestic demand. For product companies building AI in India, the cost structure that made Indian AI development economical five years ago has changed. This is not a crisis — it is an adjustment — but it changes the calculus for early-stage companies planning hiring.

What to Watch in H2 2026

The DPDP Act implementation is the regulatory event to watch. How India's data protection framework is enforced in practice will shape what AI products can be built with Indian user data. The uncertainty has already slowed some investment in AI products that depend on data collection. Clarity — even if restrictive — is better than the current ambiguity.

The AI for agriculture opportunity is underexplored relative to its scale. India has 120 million farming households. AI applications for crop disease detection, irrigation optimisation, and market price forecasting have the scale potential that healthcare and fintech AI have but have not yet found the product-market fit that those sectors have. The teams working seriously on this problem are ones I watch closely.