India’s push to build an artificial intelligence ecosystem that is both strategically independent and culturally grounded is gathering pace, as policymakers, startups and large enterprises converge around a shared question: how to create AI systems that are secure enough for national priorities and rich enough to reflect the country’s languages, knowledge traditions and institutional memory.
That debate has sharpened after comments by Arun Subramaniyan, founder and chief executive of Articul8, who argued that India’s AI trajectory should not be framed as a choice between sovereign control and cultural depth, but as a need for both. His view comes at a time when the public and private sectors are moving from broad enthusiasm around generative AI to harder questions of ownership, trust, utility and scale.
The timing is notable. Under the IndiaAI Mission, the government has been expanding support for domestic compute, foundational models, datasets, skills and applications, while official messaging at the India AI Impact Summit in February cast sovereign AI as a strategic capability rather than a branding exercise. Government-backed efforts such as BharatGen have also been positioned as multilingual, multimodal systems designed around the country’s linguistic diversity and socio-cultural context. Together, those moves show that the argument for AI sovereignty in India is no longer limited to who owns the model or where the servers sit. It is increasingly about whether AI built for India can understand how people speak, work and make decisions across vastly different regions and sectors.
Subramaniyan’s intervention adds another layer to that discussion. He has suggested that sovereignty on its own can become too narrow if it is treated only as a matter of defence, infrastructure or control. In his framing, heritage matters because AI systems that ignore local language, history and context risk becoming technically impressive but socially shallow. That point has growing resonance as global concern rises over English-heavy training data, cultural blind spots and the danger that smaller languages may become underrepresented in the next generation of digital tools.
For India, the heritage argument carries practical as well as symbolic importance. A country with hundreds of languages, wide variations in literacy and deep differences in economic activity cannot rely on AI products optimised only for Western datasets or metropolitan users. Voice interfaces, translation layers, domain-specific copilots and public-service tools all perform better when they are trained on the speech patterns, terminology and workflows of the people expected to use them. This is why multilingual model development has moved from a niche research issue to a core policy concern.
At the same time, the enterprise case is becoming clearer. Many companies have discovered that generic large language models can help with drafting, summarising and coding, but struggle when deployed in regulated or mission-critical environments without access to internal knowledge. The more valuable opportunity lies in turning proprietary documents, engineering manuals, transaction records, legal archives and operational know-how into structured intelligence that can support decisions. That is the commercial logic behind domain-specific AI, and it helps explain why firms such as Articul8 are pitching secure, enterprise-grade systems built around private data rather than public internet corpora alone.
This shift also reflects a broader global pattern. Corporate surveys and advisory research have shown that many organisations remain stuck between pilot projects and real productivity gains, largely because governance, data quality and workflow integration have lagged behind excitement over model capabilities. In practice, companies tend to get stronger returns when AI is tied to specific functions and high-value use cases rather than treated as a general-purpose novelty. For banks, manufacturers, energy groups and healthcare providers, that often means models that can work inside tightly controlled environments and understand sector language with precision.
India’s opportunity may lie in combining these two strands. One is sovereign capability: domestic compute, public datasets, local model development and policy oversight. The other is heritage depth: linguistic coverage, cultural nuance and knowledge systems that reflect the country’s own realities. The third, emerging alongside them, is institutional depth inside enterprises, where AI can turn decades of accumulated expertise into a working asset instead of leaving it trapped in silos.
There are, however, risks in the current moment. A strong sovereign AI narrative can drift into protectionism or technological duplication if it becomes a race to mimic every frontier model at any cost. Heritage-focused rhetoric can also become vague if it is not matched by usable datasets, benchmark standards and transparent governance. Enterprises face a different danger: spending heavily on AI assistants that are poorly connected to business processes, weak on accuracy and difficult to audit.
