India’s AI infrastructure opportunity won’t be won by models – It will be won with context
By Bianca Lewis, Executive Director, OpenSearch Software Foundation
Across India, the AI infrastructure race is accelerating: new foundation models are being funded, compute clusters are being commissioned, and government initiatives are fueling growth. But beneath the momentum lies a strategic blind spot that could cost Indian enterprises dearly: the assumption that model access equals AI advantage. It does not. What separates organisations that extract genuine value from AI and those that are merely running expensive pilots is something more fundamental and far harder to buy: context.
Models are fast becoming a commodity. The same base architectures, fine-tuned on similar data, are available to everyone. Deploying a specific LLM doesn’t differentiate an enterprise. Context is what transforms a model into a system that businesses can actually trust. Without it, AI is a parlour trick running on expensive infrastructure.
In enterprise terms, context means several things simultaneously. It is the structured and unstructured data that an organisation has accumulated over the years: customer records, regulatory filings, internal policy documents, and support histories. It is the memory of past interactions, so that an AI agent at step ten of a complex workflow hasn’t forgotten steps one through nine. It is domain specificity — a healthcare AI that understands clinical terminology and treatment protocols differently from how it understands general language. It is real-time signals: live inventory, pricing feeds, and breaking regulatory updates. And it is deep integration with the systems, APIs, and workflows that constitute how an enterprise actually functions.
Without this, AI outputs remain superficially impressive and operationally useless. A model without context gives generic answers to specific problems. It confidently misses the point. This is precisely why so many enterprise AI pilots fail to graduate to production: the model was fine, but the context layer was never built.
For India specifically, this challenge is not just technical, it is structural. Indian enterprises operate across multiple regional languages and fragmented data ecosystems built on decades of legacy architecture. They span public-sector banking, private hospital networks, and logistics companies that run on WhatsApp-based coordination. No model, however capable, can navigate this complexity without deep contextual grounding. A credit decisioning system that works in Mumbai needs different contextual signals than one deployed in a rural cooperative bank in Odisha. An AI assistant for a state government department needs to understand administrative processes, regulatory language, and citizen interaction patterns that exist nowhere in any global training corpus.
That complexity is precisely what makes India’s opportunity so significant. The nation that solves for contextual AI at scale, across languages, sectors, and institutional complexity, will not just build a better AI ecosystem for itself. It will build an exportable model for every emerging economy grappling with the same problem.
Managing context at scale requires open, highly scalable infrastructure—which is why technologies like OpenSearch have become foundational to the enterprise AI stack. Systems cannot rely on basic databases to feed an LLM; they must index, retrieve, and score relevant information dynamically across vast, multi-lingual, and fragmented data stores. Context overflow, where an AI agent loses track of earlier information and degrades in quality, is not an edge case. It is the default outcome when the retrieval layer is an afterthought.
The solution lies in treating context management as a first-class engineering discipline. However, context is not something that happens organically. It is engineered.
When enterprises get this right, the impact compounds. Decision-making improves because AI systems are working from accurate, relevant information rather than generalised approximations. Outputs become trustworthy enough to act on, not just interesting enough to read. Deployment moves from controlled pilots to production workflows. And the organisation builds a contextual asset – a layer of structured, relevant, integrated knowledge – that competitors cannot simply replicate by switching models.
India’s AI race will not be determined by which country can access the most powerful model. Models are available to everyone with a credit card and an API key. The race will be determined by who builds the deepest, most relevant, most operationally integrated context layer on top of those models. That is where the moat lies. That is where the real infrastructure investment needs to go.
Models are the starting point. Context is the endgame. India has the data, the domain diversity, and the institutional complexity to make contextual AI a genuine area of global leadership. But only if it chooses to invest there, rather than simply celebrating the access it already has.