The Activation Gap: Why India Is Announcing AI Agents Faster Than It Can Deploy Them

India's IT services industry is announcing agentic AI at scale. The execution data is not keeping up.

That gap — between what Indian IT firms are selling as AI-native transformation and what their own numbers show about actual deployment — is the real story of this moment in the market. Wipro launched a dedicated AI-Native Business and Platforms unit on April 1st. Infosys tied up with Anthropic in February to embed Claude models across its enterprise platform. Tech Mahindra and NVIDIA shipped a telco network reasoning agent at MWC Barcelona in March. Persistent Systems built an agentic drug discovery platform on NVIDIA's BioNeMo stack. Microsoft pledged 200,000 Copilot licenses to four Indian IT giants and committed $17.5 billion to India's AI infrastructure through 2029.

The announcements are real. The gap between announcement and execution is also real. According to the Ernst & Young AIdea of India: Outlook 2026 report, 47% of Indian organizations claim to have GenAI use cases running in production. Only 10% are scaling those use cases across the business. That difference — between claiming a deployment and running one at enterprise scale — is where the ad paradox lives.

The distribution of that gap is not random. The 47% clusters heavily at the top of the Indian IT hierarchy — the Wipros, Infosyses, and Tech Mahindra layer that can afford dedicated AI units, formal Center of Excellence arrangements with model providers, and the internal engineering depth to take a demo and make it work on a Tuesday afternoon. Below that tier, the channel partners, regional system integrators and managed service providers are being asked to sell and deploy agentic AI systems whose complexity exceeds anything they were trained to manage, for customers whose expectations were set by the same announcements coming out of Bangalore and Hyderabad.

The shift in what enterprises are actually buying is measurable. According to EY, 91% of Indian organizations now cite speed of deployment as the single biggest factor driving their AI purchasing decisions — ahead of cost, ahead of model accuracy, ahead of vendor relationships. This is a rational response to years of pilot paralysis, but it creates a structural pressure: the firms best positioned to win on speed are the ones already at scale, not the ones trying to build their first agentic deployment.

The DPDP Act is adding a second dimension. India's data protection legislation is forcing enterprise buyers to think harder about where AI agents put data, who can audit them, and what happens when they make a decision that violates compliance requirements. For channel partners, this is supposed to be an opportunity — governance, compliance and long-term operational management are exactly the services that become more valuable as agentic systems become more embedded. But building those capabilities requires investment that many regional partners have not yet made.

The counterargument is that the activation gap will close on its own. The EY data shows nearly half of surveyed organizations have already moved more than a fifth of their GenAI proofs of concept into production — not a trivial base of real deployment. The Indian IT industry's institutional capacity to close execution gaps is, historically, one of its genuine competitive advantages. If the 47% figure is even half right, India still has more AI in production than almost any other large market.

What the numbers cannot show is the composition of that production. A GenAI use case running in a controlled pilot on a subset of a company's data is categorically different from an agentic system making decisions autonomously across enterprise workflows. The Indian IT firms that understand this distinction — and can explain it honestly to customers rather than overselling the autonomous future while underdelivering the assisted present — are the ones that will build the durable channel businesses. The ones that packaged the announcement as the product will find their customers asking harder questions when the pilot ends and the production bill arrives.

The Wipro AI-Native unit's press release uses the phrase "services as software" to describe what it is building. It is a useful frame — the outcome a customer buys is not person-hours but a working system. Whether that outcome arrives on the timeline the announcements imply, and whether the channel that delivers it survives the transition with its business model intact, is the question the next six months will answer.