India shifts towards sovereign, agentic AI to foster autonomous systemic control and localised trust

India is progressing from copilot AI to autonomous, agentic systems, prioritising control, localisation, and trustworthy data frameworks amid significant public and private sector investments aimed at developing indigenous AI capabilities and reducing reliance on external providers.

India is moving from copilot-style AI to agentic systems that can act autonomously across industry and government, driven by a deliberate push to build sovereign infrastructure, trusted data...

frameworks, human-plus-AI accountability and “responsible AI by design”. The shift, outlined in a recent analysis by NDTV Profit, is already being matched by major public and private initiatives that aim to anchor compute, models and governance inside the country while allowing governed interfaces to global capabilities.

Sovereign infrastructure: control at scale
The core argument for sovereign infrastructure is control: where data lives, how compute is provisioned, how models are developed and how agents are operated at scale. According to the Principal Scientific Adviser’s description of the IndiaAI Mission, the government plans to build a robust AI infrastructure including more than 10,000 GPUs through public–private partnerships, and to support indigenous large multimodal and domain-specific foundational models. Industry observers describe emerging hybrid sovereign setups that combine domestic data centres, sector-specific clouds for banking, government and public sector units, and governed access to hyperscalers and enterprise software.

Private investment and consulting firms are aligning with that trajectory. Microsoft announced a record $17.5 billion investment in India over four years to expand cloud and AI infrastructure, the largest such pledge in Asia, underlining hyperscalers’ competitive interest in India’s market, according to the Associated Press. At the same time Deloitte has launched a global AI Infrastructure Centre of Excellence to help clients design and operate specialised AI data centres and is positioning that work as part of a factory-as-a-service approach to move from silicon to business-ready AI solutions.

The national programme also reports rapid capacity growth. Organisers at an IndiaAI event said national compute capacity has crossed 34,000 GPUs and highlighted three startup projects selected to develop indigenous foundation models and voice AI capabilities. Those capacity gains and startup awards are intended to reduce reliance on external stacks while enabling domestic innovation.

Trusted data and context-aware models
Autonomous agents depend on continuous access to accurate, relevant enterprise and citizen data, which raises questions of provenance, lineage and governance. Deloitte’s survey, “The State of Generative AI in the Enterprise,” finds that data quality and integration remain primary obstacles to scaling generative AI, reinforcing the need for strong, enforceable data governance. The IndiaAI Mission emphasises streamlining access to high-quality non-personal datasets and building data labs in Tier 2 and 3 cities to broaden the talent and dataset base, according to the Principal Scientific Adviser’s documentation.

Context-aware models, tuned to local languages, regulatory regimes and operating conditions, are central to improving explainability and reliability as agents interact with employees, customers and citizens. That localisation drive is visible in the government’s selection of startups: programmes announced at the IndiaAI event include teams building multilingual and sector-focused large models aimed at healthcare, defence and education. However, not all sovereign projects are being framed as open public goods: reporting by Medianama notes that the government-selected company Sarvam AI will develop a sovereign large language model that will not be open-sourced and that a government body will take equity in the company in return for funding and access to compute resources.

Human-plus-AI accountability
The proposed operating model for agentic AI is “human plus AI”: humans set goals, guardrails and strategic intent while agents carry out high-volume operational tasks. NDTV Profit’s analysis stresses that this combination is essential for high-stakes domains such as payments and supply-chain management. India’s deep pool of skilled technology workers is cited as an enabling factor for designing, overseeing and improving such systems at scale. The IndiaAI Mission’s emphasis on AI education across academic levels and the establishment of Data and AI Labs aims to expand that workforce, according to the Principal Scientific Adviser.

Responsible AI by design
As agents assume operational roles, the article argues, trust cannot rely solely on after-the-fact audits. Responsible AI must be embedded in real time: explainability, observability and auditability should be live within workflows so that decisions, data inputs and applied rules are visible as they occur. Deloitte’s analysis supports this position, noting that organisations with continuous monitoring report fewer compliance incidents and that real-time controls reduce the risk of systemic failures when agents act autonomously.

Policy and commercial tensions
The convergence of public funding, private capital and consulting-driven infrastructure raises practical and policy tensions. Large private investments, such as Microsoft’s $17.5 billion pledge, promise faster scale and skills transfer but also intensify questions about dependency on foreign cloud providers and the commercial terms under which hyperscaler technologies integrate with sovereign stacks. The IndiaAI Mission’s mix of public compute, startup funding and sector clouds is designed to mitigate those dependencies, yet the government’s decision to take equity in select commercial model vendors, and to permit non-open-source sovereign models, highlights trade-offs between strategic control and public access.

Challenges ahead
Even with rapidly expanding GPU capacity and targeted startup support, hurdles remain. Deloitte’s survey highlights the high cost of infrastructure, the difficulty of measuring AI’s business impact, and an evolving regulatory environment as limits on rapid scale-up. Data quality, integration and change management are persistent constraints that must be tackled for agentic systems to move beyond pilots into trusted, nationwide operation.

Conclusion
India’s articulated “sovereign, agentic AI” strategy weaves together state-led compute and dataset programmes, private investment and industry-led tooling to pursue control, localisation and responsible operation of autonomous agents. The approach recognises that technical scale, GPUs, data centres and models, must be matched by live governance, trustworthy data pipelines and human oversight. As major global players and domestic startups race to build capabilities, the balance between strategic sovereignty, commercial partnerships and public accountability will determine whether agentic AI delivers inclusive benefits or entrenches new dependencies.

Source: Noah Wire Services