India leads the world in enterprise AI deployment. That may be exactly the problem.

There is a number that India’s technology sector has been celebrating this month, and it deserves a closer look before we decide whether it is good news.

Deloitte’s State of AI in the Enterprise report for 2026 finds that 40% of Indian respondents report significant or full AI usage, compared with a global average of approximately 28%. India is, by this measure, the world’s leading enterprise AI adopter. The announcements have been swift and self-congratulatory. We are ahead of the United States. Ahead of Europe. Ahead of China. The numbers are in, and we are winning.

Except the same report contains another number, quieter and less quoted. Only 0 to 4% of Indian companies possess a high level of AI expertise, lagging behind the global average of 2 to 8%. Read that again. India leads the world in deploying AI, and sits near the bottom of the world in understanding it. We are running the machine without reading the manual. And 94% of Indian firms plan to increase their AI budgets over the next year. We are accelerating into the gap.

This is not a story about ambition failing. It is a story about ambition succeeding so fast that comprehension cannot keep pace. And in technology, that particular kind of success has a well-documented tendency to produce a specific kind of dependency.

The pattern is not new to India. We have been here before, in different registers. The IT outsourcing economy built a $280 billion industry on someone else’s intellectual property, someone else’s enterprise software, someone else’s cloud. We wrote the code. We ran the helpdesks. We managed the implementations. But the platforms, the architectures, the foundational decisions about what the technology would do and who it would serve, those remained elsewhere. The value we captured was real but it was also capped, structurally, by the fact that we were operating on top of systems we did not own.

Agentic AI is setting up the same structure at a faster speed and a larger scale. Experts have warned that startups built merely on front-end layers using APIs of large foreign models may struggle to sustain themselves, as companies like Google, Anthropic, and OpenAI will ultimately prioritise monetising their own ecosystems. One researcher at the India AI Impact Summit was more blunt, describing the dynamic as digital neo-colonialism: “If we are sending all our data, we are missing an opportunity to build our model, and they will now build products which are trained on our data to sell it back to us.” 

I have called this the wrapper economy before. India owns the interface. Someone else owns the intelligence. The 40% adoption figure, celebrated without context, is a near-perfect description of that economy in action. We are the world’s most enthusiastic users of AI systems we did not build, cannot fully audit, and do not control. That is not a technology strategy. It is a procurement strategy dressed in the language of transformation.

Without domestic capabilities in compute infrastructure, foundational models, and data governance, India risks becoming a net importer of intelligence, capturing limited downstream gains while exporting economic rents. The rents are not abstract. Every Indian enterprise running a foreign large language model at scale is paying, in compute costs and data flows and licensing fees, for intelligence that was not built here and will not remain here if the geopolitical weather changes. Overdependence on foreign platforms creates risks including data leakage, covert surveillance, and technology restrictions. These are not hypothetical risks in 2026. They are observed ones.

To be fair, India has not been standing still. The IndiaAI Mission has produced real compute infrastructure. BHARATGen, India’s first fully sovereign multimodal large language model, launched in June 2025, and Sarvam AI has developed models running on 105 billion parameters, prioritising efficiency and localisation over scale. The New Delhi Declaration at the AI Impact Summit brought nearly 90 countries together around shared AI infrastructure commitments. These are not nothing. But they are also not yet sufficient to close the gap between what we are deploying and what we understand about what we are deploying.

The Deloitte report, perhaps unintentionally, makes this point with its own language. The most common responses Indian organisations cite for building AI capability include upskilling and reskilling programs at 61%, incentives to drive adoption at 59%, and broader workforce education at 53%. Notice what is not on that list. Governance. Auditing. The ability to interrogate a model’s outputs, trace its decisions, or identify where it is wrong. We are teaching people to use AI. We are not yet, at scale, teaching people to question it.

When organisations scale AI faster than they build skills, they create systems that are difficult to audit, correct, or improve over time.  In an enterprise, a system that cannot be corrected produces inefficiencies. In a government deployment, or a healthcare system, or a credit-scoring infrastructure, it produces harm that is invisible until it is too late, and opaque even then.

The 40% number is not a verdict. It is a moment. India is genuinely at an inflection point in its AI trajectory, and the decisions made in the next two to three years about where to build depth rather than just breadth will determine whether this moment becomes a foundation or another layer of dependency.

What makes this particularly urgent is the speed. Agentic AI, which refers to systems that can take actions and complete tasks independently, is currently used at least moderately by 23% of global companies and is expected to grow to 74% within two years. We are not moving from comprehension toward deployment. We are moving from deployment toward even more autonomous deployment, with the comprehension gap still open underneath.

A country that leads the world in using AI it does not understand is not an AI power. It is an AI market. The difference matters enormously, for who captures the value, who governs the risk, and who gets to shape what this technology does next.

If India is serious about the sovereignty it has been announcing, it needs one specific and measurable commitment: every public-sector AI deployment must require a domestic capability audit before signing any foreign platform contract, not as a barrier to adoption, but as the minimum condition for understanding what is being adopted.

The manual does not have to be written before the machine is switched on. But someone in the room should be able to read it.

END OF ARTICLE