Sovereign AI: Why India Wants Its Own AI Infrastructure and Models

I. Introduction

The governance of artificial intelligence presents constitutional democracies with a challenge that is simultaneously technical, juridical, and geopolitical. As AI systems migrate from the periphery of commercial enterprise into the operational core of public administration, financial services, and national security, the question of who controls these systems and pursuant to what legal authority has assumed first-order constitutional significance.

The concept of "Sovereign AI" denotes the affirmative capacity of a State to exercise comprehensive jurisdictional authority over artificial intelligence infrastructure deployed within its territory, extending to the physical substrate of computation, the data upon which models are trained, and the normative parameters that govern model behaviour. This concept, which existed as a theoretical construct in academic discourse as recently as 2024, achieved practical and legislative salience in early 2026 as States across the democratic world treated algorithmic dependence on foreign platforms as a structural vulnerability analogous to energy dependence capable of producing what one strand of scholarship has termed "algorithmic colonisation."¹

India's engagement with this question is both instructive and distinctive. Possessing the world's largest digital public infrastructure, a constitutional commitment to the welfare state, and a legal tradition that has long contested the distinction between economic sovereignty and political sovereignty, India has pursued Sovereign AI through a tripartite strategy: (i) the physical construction of nationally controlled compute infrastructure; (ii) the codification of data trusteeship as a mode of public law governance; and (iii) active participation in the formation of multilateral accountability norms. Each of these dimensions raises distinct questions of legal analysis that this article addresses in turn.

Part II examines the doctrinal foundations of computational sovereignty, situating India's GPU expansion programme within the broader public law framework of critical national infrastructure. Part III analyses the legal architecture of data governance, with particular reference to the Digital Personal Data Protection Act and the Bhashini linguistic initiative. Part IV considers the semiconductor dimension the so-called "Silicon Realism" turn and its implications for supply-chain law and international trade obligations. Part V examines the regulatory framework governing AI deployment in sensitive public sectors, including the intermediary liability amendments and the New Delhi Declaration's democratic diffusion model. Part VI offers conclusions and identifies outstanding questions of constitutional law.

II. The Computational Pillar: Public Infrastructure, State Capacity, and the Prevention of the Compute Divide

A. Theoretical Foundations

The State's interest in controlling physical computational infrastructure admits of several distinct legal justifications. The most immediate derives from the administrative law concept of critical national infrastructure a category of resources whose disruption would materially impair the State's capacity to discharge its constitutional obligations. In Indian constitutional doctrine, the welfare state mandate enshrined in Part IV of the Constitution of India (the Directive Principles of State Policy) has been read by the Supreme Court as imposing on the State a positive duty not merely to refrain from rights-violating action but to actively constitute the material conditions under which rights may be exercised. Where AI systems become integral to the delivery of public services, the denial of access to adequate computational resources becomes a question of constitutional rather than merely economic significance.

A second justification, rooted in principles of anti-monopoly and economic sovereignty, addresses the structural risk of "compute concentration" a state of affairs in which the processing capacity required for advanced AI development is controlled by a small oligopoly of private actors, the majority domiciled in foreign jurisdictions. The legal concern is not merely economic but constitutional: a State whose public agencies and research institutions depend upon commercially provided compute from entities beyond its jurisdictional reach is structurally exposed to supply-chain weaponisation, the exercise of contractual termination rights as a form of strategic coercion, and the imposition of licensing conditions incompatible with domestic legal obligations.

B. The IndiaAI Mission: Legal Character and Institutional Design

India's legislative and executive response to the compute challenge is centred on the IndiaAI Mission, established with an outlay of ₹10,371.92 crore and designed to construct and operate a nationally controlled GPU cluster of significant scale.² The Mission's juridical character whether it constitutes a public undertaking, a public-private partnership, or an exercise of the State's residual regulatory capacity has material implications for procurement law, subsidy regulation, and accountability. The scheme's administrative structure treats compute access as a subsidised public service, offering high-performance processing capacity at approximately ₹65 per GPU-hour to registered startups and academic researchers³ a pricing structure that engages principles of non-discriminatory access, State aid doctrine, and, potentially, obligations under the Agreement on Subsidies and Countervailing Measures administered by the World Trade Organisation.

As of February 2026, the national cluster had been expanded to 58,000 GPU units, inclusive of a tranche of 20,000 high-specification units announced by Minister Ashwini Vaishnaw at the India AI Impact Summit.⁴ This expansion is best understood as the State's response to a dual market failure: the under-provision by private markets of computational resources accessible to small and medium enterprises, and the strategic externality arising from concentrated foreign control of such resources.

III. Data Governance: Trusteeship, Localisation, and Linguistic Sovereignty

A. The Doctrine of Data Trusteeship

The emergence of data as the primary productive input of artificial intelligence systems has necessitated the development of new legal doctrines governing its ownership, custody, and deployment. Classical property law, constructed around the paradigm of rivalrous and excludable assets, provides an inadequate framework for digital information, which may be replicated without diminution and whose value is in significant part a function of aggregation rather than individual content. Indian data law has responded to this analytical deficiency through the development of the doctrine of "data trusteeship" the proposition that the State exercises a fiduciary-like custody over citizen data, holding it not as owner but as trustee for the benefit of the data principals whose information constitutes the national data estate.

The Digital Personal Data Protection Act (DPDP Act) represents the legislative codification of this principle. The Act regulates the processing of digital personal data in a manner that ensures, inter alia, that Indian citizen data is not extracted for use in training foreign AI models without consent.⁵ The trusteeship concept operates to characterise the State not as a passive regulator but as an active custodian with affirmative obligations to prevent the exploitation of the national data estate by entities whether foreign or domestic that would extract value without accountability.

B. Bhashini: Linguistic Data as a Constitutional Matter

India's linguistic pluralism encompassing 22 constitutionally scheduled languages and several hundred dialects presents a distinctive dimension of the data sovereignty question. The deployment of AI systems trained exclusively or predominantly on high-resource languages (principally English and Mandarin) in Indian public administration raises constitutional concerns under Articles 29 and 350A, which protect the linguistic rights of minorities, as well as under the equality guarantee of Article 14, to the extent that linguistically biased AI systems systematically disadvantage speakers of low-resource regional languages in their interactions with public institutions.

The Bhashini initiative addresses this constitutional imperative through the construction of large-scale multilingual datasets across India's scheduled languages.⁶ The legal significance of this programme lies not merely in its technical output but in its normative framing: linguistic and cultural data is treated as a matter of national security, insulating it from the standard commercial logic under which training data is assembled by private entities for private purposes. The migration of the Bhashini platform to the indigenous Shakti Cloud infrastructure completed in February 2026 with zero data loss across more than 3.5 billion files⁷ represents a decisive exercise of the State's regulatory authority to prevent extra-territorial processing of data constituting a national cultural asset.⁸

IV. Silicon Sovereignty: The Hardware Dimension and the Limits of Software Independence

A. The "Hollow Sovereignty" Problem

A central insight animating India's most recent phase of AI industrial policy is that software sovereignty, unaccompanied by hardware independence, is juridically incomplete. The legal concept of "remote disablement" whereby a foreign vendor may, pursuant to contractual terms or governmental compulsion in its home jurisdiction, disable or restrict the functionality of hardware or software deployed within Indian territory illustrates the structural vulnerability. An AI system whose inferential capacity depends upon GPUs subject to foreign export control regimes and embedded with remote management capabilities cannot be characterised as "sovereign" in any meaningful sense; it is, at best, a conditional asset whose operational continuity is contingent upon the tolerance of a foreign State.⁹

B. Pax Silica and the International Law of Technology Supply Chains

India's formal accession to the Pax Silica coalition on February 20, 2026 represents an assertion of supply-chain sovereignty through multilateral rather than unilateral means.¹⁰ The coalition which provides participating States with priority access to critical minerals and advanced lithography equipment functions as a form of collective technology security, analogous in structure to collective defence arrangements but operating in the domain of strategic industrial inputs. The international legal questions raised by such arrangements are significant: to the extent that coalition membership confers preferential access to dual-use technologies, tensions may arise with WTO Most-Favoured-Nation obligations and the non-discriminatory access principles that underlie multilateral trade law.¹¹

At the domestic level, the advancement of semiconductor assembly facilities and the Tata Electronics partnership with global chip manufacturers represent the physical instantiation of the principle that chips deployed in critical national infrastructure must be manufactured under conditions of domestic oversight.¹² The relevant legal instruments industrial licensing conditions, security clearance requirements, and procurement rules favouring domestically manufactured components collectively constitute an emerging body of "silicon law" that sits at the intersection of industrial policy, procurement regulation, and national security law.

V. The Accountability Framework: Democratic Diffusion and the Governance of Sovereign AI Systems

A. The New Delhi Declaration and the Democratic Diffusion Model

The deployment of AI systems in sensitive public sectors healthcare, justice, social welfare, immigration, and security raises accountability questions of constitutional magnitude. The administrative law principle of procedural fairness, the right to reasoned decision-making under Article 21 of the Constitution as interpreted in Maneka Gandhi v. Union of India,¹³ and the emerging international norm of algorithmic accountability together generate a legal environment in which the State cannot simply deploy sovereign AI without appropriate governance mechanisms.

The New Delhi Declaration, endorsed by 88 nations at the close of the India AI Impact Summit on February 21, 2026, establishes what its text denominates a "Charter for Democratic Diffusion" of AI governance.¹⁴ The charter is framed as a voluntary and non-binding framework to promote access to foundational AI resources, support locally relevant innovation, and build resilient AI ecosystems within the constraints of national law.¹⁵ The legal character of these commitments whether they constitute standards with persuasive authority or soft-law norms capable of domestication is a question of ongoing scholarly debate. India's position as host nation and primary architect of the Declaration strongly suggests their progressive incorporation into domestic regulatory instruments.

B. The Intermediary Guidelines and Domestic Accountability

The newly notified Intermediary Guidelines, issued pursuant to Section 79 of the Information Technology Act, 2000, constitute the principal domestic instrument for giving effect to the accountability aspirations of the New Delhi Declaration. The Guidelines impose obligations on AI systems deployed in specified sensitive categories, require a degree of algorithmic transparency sufficient to enable regulatory review, and establish a complaints-based enforcement mechanism under the supervision of the competent authority.

The legal questions raised by this framework are not trivial. First, the scope of "sensitive" deployment categories requires authoritative definition; the exercise of definitional discretion by the executive without parliamentary guidance may be challenged on grounds of excessive delegation under the established doctrine of Delhi Laws Act, 1912, In re.¹⁶ Second, the substantive standards of "fairness" and "alignment" are not self-defining, and their operationalisation through regulatory guidance will require both technical expertise and normative judgment. Third, the enforcement mechanism must be designed to avoid the dual risks of regulatory capture the subordination of audit functions to the interests of regulated entities and regulatory overreach that chills legitimate innovation in AI development.

VI. Conclusions and Outstanding Questions

The foregoing analysis establishes that Sovereign AI, as pursued by India in 2025–2026, constitutes a legally coherent and constitutionally grounded project of jurisdictional assertion over critical digital infrastructure. The project rests on three mutually reinforcing pillars computational sovereignty, data trusteeship, and international accountability each of which engages distinct bodies of domestic and international law.

Several outstanding legal questions merit sustained scholarly attention. First, the constitutional status of compute infrastructure as critical national infrastructure under Article 19(6) remains to be authoritatively determined; its resolution will bear upon the validity of the IndiaAI Mission's subsidy structure and pricing regime. Second, the WTO compatibility of the Pax Silica preferential arrangements presents unresolved questions in the law of international trade. Third, the enforceability of the New Delhi Declaration's aspirational commitments, and their relationship to India's existing privacy and IT law framework, requires legislative clarification.

What is clear is that the era in which AI was treated as a purely commercial commodity subject exclusively to private contract and market regulation has concluded. The emergence of Sovereign AI as a legal and policy category reflects a structural shift in the relationship between the State, the market, and the digital infrastructure upon which both increasingly depend. The legal frameworks adequate to this new configuration remain, as this article has sought to demonstrate, incompletely elaborated; their development will constitute one of the defining exercises in public law scholarship and legislative craftsmanship of the coming decade.

The views expressed herein are those of the author(s) alone.

Footnotes

¹ The term "algorithmic colonisation" has gained currency in the literature to describe the structural dependency of Global South states on AI infrastructure controlled by foreign technology entities. See Abeba Birhane, The Unseen Black Faces of AI Algorithms, Nature 589 (2021).

² Ministry of Electronics and Information Technology, IndiaAI Mission — budget and programme overview: indiaai.gov.in; see also DD News, Transforming India with AI: Rs 10,300 crore mission, 38,000 GPUs & a vision for inclusive growth: ddnews.gov.in

³ The subsidised rate of approximately ₹65 per GPU-hour is documented in: Trak.in, AI Mission 2.0: India To Have 58,000 GPUs, Available At Rs 65/Hour: trak.in

⁴ The February 17, 2026 announcement of the 20,000-unit expansion to a total of 58,000 GPUs is reported in: Kashmir Reader, India's AI Infrastructure Revolution: USD 100 Billion, 58,000 GPUs, And A Global Compute Bank: kashmirreader.com; see also CIOL, India AI Impact Summit 2026 Sees $200 Billion in Commitments: ciol.com

⁵ Digital Personal Data Protection Act, 2023. For a summary of the regulatory framework, see the Ministry of Electronics and Information Technology's official communications.

⁶ Bhashini National Language Translation Mission; see DD News, Transforming India with AI: ddnews.gov.in

⁷ Business Standard, India AI Impact Summit: Govt migrates BHASHINI to Indian cloud platform (Feb. 10, 2026): business-standard.com

⁸ Open Source For You, Yotta Replaces Hyperscalers With Sovereign AI Cloud Shakti For BHASHINI: opensourceforu.com; CIO&Leader, Yotta and BHASHINI Collaborate to Enable Sovereign AI Cloud: cioandleader.com

⁹ The concept of "API kills" and supply-chain weaponisation is addressed in the US Bureau of Industry and Security's export control framework for advanced computing items (October 2023).

¹⁰ Fortune, India joins U.S. 'Pax Silica' semiconductor alliance (Feb. 20, 2026): fortune.com

¹¹ Carnegie Endowment for International Peace, India Signs the Pax Silica — A Counter to Pax Sinica?: carnegieendowment.org; India's World, India Joins Pax Silica: The Technology Order Taking Shape: indiasworld.in

¹² Business Today, AI Impact Summit 2026: India signs US led Pax Silica; electronics, semiconductor industry to benefit greatly: businesstoday.in; Press Information Bureau, India Joins Pax Silica at India AI Impact Summit 2026: pib.gov.in

¹³ Maneka Gandhi v. Union of India, (1978) 1 SCC 248.

¹⁴ Press Information Bureau, AI Impact Summit 2026 Concludes with Adoption of New Delhi Declaration: pib.gov.in

¹⁵ Insights on India, New Delhi Declaration — AI Impact Summit 2026: insightsonindia.com; Drishti IAS, New Delhi Declaration on AI Impact: drishtiias.com

¹⁶ Delhi Laws Act, 1912, In re, AIR 1951 SC 332 (on the permissible limits of legislative delegation to the executive).