AI sovereignty has become the most sought-after strategic project of our time, fuelled by fears of structural overreliance on a handful of technology providers concentrated in the United States and China. Despite this momentum, the concept remains systematically underspecified. This under-specification is not incidental: it leaves the concept elastic enough for political actors to invoke as a consensus-building device, obscuring the trade-offs that separate real sovereignty from performed sovereignty.
That vacuum is increasingly filled by technology firms such as NVIDIA, Google Cloud, and Microsoft Azure, which are commodifying sovereignty through Sovereignty As-a-Service arrangements. The relevant question, therefore, is not whether to depend, but on whose terms dependence is structured.
Sovereign AI questions become more complex when mapped across the AI stack. For India, sovereignty across the entire stack is neither feasible nor necessary, as sovereignty matters the most where dependence creates exploitable chokepoints.
The economy spans multiple sectors — manufacturing, healthcare, finance, and agriculture among them — each drawing artificial intelligence (AI) investment at varying rates and towards distinct objectives. This spread underscores the structural necessity for sovereign control over the AI layer, since a fault or foreign dependency in one part of the system can propagate across the whole.
The defence sector’s capital outlay of Rs 2.19 trillion for FY27 reflects its overall investment priorities. However, the annual allocation for AI within the sector remains approximately Rs 100 crore, underscoring that sovereign AI capability is still at an early stage relative to the sector’s scale.
Currently, both areas face major roadblocks, ranging from heavy reliance on imported chips to fragmented data systems operating in silos. This creates a pressing need to recalibrate sovereign AI posture in these two sectors. The challenge for both lies in identifying which layers require control and strategic dependence, and which can be leveraged through partnership.
For India, this reorientation is pertinent, as it brings initiatives such as layered stack control and DPI-AI integration, alongside selective partnerships that bridge gaps across the AI stack. The question, though, is whether this posture is being built with sufficient calibration across the defence and economic sectors.
How Sovereign is India’s AI Economy?
The concentration problem in AI is structural, not transitional, because each core input — compute, capital, talent, data, and energy — exhibits increasing returns to scale. This means each unit of advantage an incumbent gains makes the next unit cheaper and more accessible to acquire, while raising the cost of entry for everyone else. Hyperscale operators like Amazon, Microsoft, and Google control 58 percent of all hyperscale capacity worldwide.
Hyperscaler companies based in the US and China, in addition to processing their own AI workloads domestically, also rent out compute capacity to the world through services such as Microsoft’s Azure, Google Cloud, and Alibaba Cloud, thereby capturing revenue, data-handling leverage, and the standard-setting power that comes with owning data centres. Dependent countries, meanwhile, retain less control over their own AI sector — over cost, availability, and continuity decisions.
Given transformative predictions about AI’s impact on the economy, it is imperative to devise a sovereign AI framework for sectors across the AI stack. Localised AI integration is projected to add US$500-600 billion to India’s GDP by 2035, making an end-to-end sovereign AI stack — encompassing semiconductors, chips, applications, and compute — vital to protecting core economic sectors from external supply chain disruptions.
The Banking, Financial Services, and Insurance (BFSI) sector has emerged as India’s leading sector in the deployment of artificial intelligence. As these systems are integrated into high-stakes applications such as credit approvals, fraud detection, and compliance, this adoption also carries incremental risks.
India currently lacks domestic high-performance GPU fabrication capability, a constraint shared by most nations outside Taiwan and the US. For India specifically, though, this means that virtually every sovereign AI model trained on Indian financial data depends on hardware whose export is subject to a foreign government’s licensing decisions, not India’s own.
India’s Financial DPI presents a sovereignty paradox. NPCI operates the transactional rails underpinning the UPI infrastructure, which processes over 241.6 billion transactions annually — a scale that makes the RBI’s 2018 data localisation mandate legally consequential, anchoring sovereignty claims at the data and application layer. In February 2026, NPCI announced a partnership with NVIDIA to train its in-house FiMI (Finance Model for India) on NVIDIA’s Nemotron open-weight models and the NeMo framework. This highlights a critical structural dependency that escapes the localisation mandate: India currently lacks domestic high-performance GPU fabrication capability, a constraint shared by most nations outside Taiwan and the US. For India specifically, though, this means that virtually every sovereign AI model trained on Indian financial data depends on hardware whose export is subject to a foreign government’s licensing decisions, not India’s own.
The RBI’s own trajectory toward an Indian Financial Services Cloud addresses the cloud hosting dimension of this dependency, but the chip layer remains unresolved.
“Sovereign AI” in India highlights a structural dichotomy between algorithmic independence and reliance on foreign hardware. The India AI Mission has deployed roughly 34,000 NVIDIA H100, H200, and Blackwell GPUs, which served as the foundation for Sarvam AI’s flagship open-source models, including Sarvam-30B and Sarvam-105B.
While these models demonstrate progress in localised data sovereignty, the reliance on foreign hardware exposes a structural gap in India’s semiconductor strategy, as the India Semiconductor Mission (ISM) and Tata’s Dholera fab remain restricted to assembly and packaging. This dependency creates economic vulnerability, as expenditure on computational resources continues to generate revenue for foreign hardware suppliers while remaining susceptible to global price fluctuations and export-control measures.
Consequently, India’s objective to scale to 100,000 public GPUs by December 2026 ties its future to foreign supply chains and volatile geopolitical dynamics.
Silos and Silicon in India’s Defence AI
AI is no longer peripheral to warfare; it has become its grammar. It extends into the electromagnetic spectrum, data networks, financial infrastructure, satellite constellations, and critical supply chains. The India AI Impact Summit 2026 demonstrated India’s readiness, showcasing dual-use platforms and other initiatives aimed at establishing a secure, networked, AI-driven ecosystem. A fundamental question remains, however: is New Delhi prepared to recognise AI as a core pillar of national security?
For FY 2026-27, India has allocated Rs 7.85 lakh crore to defence, including Rs 29,100.25 crore for DRDO research and Rs 1.85 lakh crore for capital acquisitions to integrate artificial intelligence, drones, and advanced platforms. Though this allocation, alongside the Defence AI Council’s (DAIC) mandate since 2019, signals intent, it does not close the structural gap. In Operation Sindoor, India demonstrated its capability to deploy AI on the battlefield. It is therefore timely to ask whether, under sustained conflict, export controls, supply chain pressure, or hardware-level interdiction, India’s AI-enabled military would continue to function — and on whose terms its decision logic ultimately runs.
To sustain strategic deterrence, India must align its capabilities with the defining features of contemporary conflict: data-centric operations, persistent hybrid engagements, grey-zone escalation, rapid decision-making, and autonomous human-machine collaboration. Without interoperable data systems, the goal of establishing a “Joint AI-Fused Single Operating Picture” will remain unfulfilled.
Sovereign AI in defence is unattainable without sovereign cloud infrastructure. India’s current initiatives, including MeghRaj, the Army Cloud, Project Sanjay, and the DRDO-RailTel system, remain fragmented, undermining the effectiveness of real-time, AI-driven command. To sustain strategic deterrence, India must align its capabilities with the defining features of contemporary conflict: data-centric operations, persistent hybrid engagements, grey-zone escalation, rapid decision-making, and autonomous human-machine collaboration. Without interoperable data systems, the goal of establishing a “Joint AI-Fused Single Operating Picture” will remain unfulfilled.
The Design-Linked Incentive (DLI) Scheme has advanced specialised RISC-V and Edge AI architectures, which are crucial to developing “silicon sovereignty“. A significant gap persists, however, between civilian innovation and military deployment. This is most evident in the ruggedisation gap, where chips validated for stable civilian environments have not been adapted to withstand the heat, shock, and vibration of battlefield use. The MIL-SPEC qualification process is therefore critical for fitting these chips — manufactured under the DLI scheme and fabricated by third-party foundries such as TSMC — onto defence platforms. This strategic bottleneck risks leaving frontline military systems anchored to vulnerable, foreign-proprietary architectures amid geopolitical instability.
Strategic Autonomy in Building a Sovereign AI
The understanding to emerge from the India AI Impact Summit 2026 is that the pursuit of sovereign AI is not a problem to be solved but a condition to be managed, both strategically and continuously, with a focus on identifying where the real leverage lies.
India must strategise to become a selective leader in key sectors aligned with its core economic strengths. AI sovereignty should be understood not as full-stack self-sufficiency, but as the capacity to adapt as technologies evolve and, with that, to renegotiate the terms of dependency as national priorities shift.
India must strategise to become a selective leader in key sectors aligned with its core economic strengths. AI sovereignty should be understood not as full-stack self-sufficiency, but as the capacity to adapt as technologies evolve and, with that, to renegotiate the terms of dependency as national priorities shift.
Dependencies are inevitable; the real question is whether they are diversified, reciprocal, and negotiable. Both challenges and possibilities lie here for India and the wider Global South — amid the AI boom, the task is not to become part of great-power rivalry, but to develop sufficient industrial depth, logistical resilience, and policy autonomy. This prioritisation and goal-setting will prevent that rivalry from determining development outcomes.
Sovereignty for India, then, is a process through which it not only navigates the global AI order but also engages with it on terms it has helped set. The question that needs deliberation is not whether India will be part of the global AI order — it already is. The question is: on whose terms?
Anubhuti Jain is a Research Intern for the Observer Research Foundation.
Disclaimer: Claude was used for language refinements.
The views expressed above belong to the author(s). ORF research and analyses now available on Telegram! Click here to access our curated content — blogs, longforms and interviews.