> ## Content Index
> Fetch the complete content index at: https://aipriorities.digitalfutureslab.in/llms.txt
> Use this file to discover other available public pages before exploring further.

# Who is the Sovereign? Analysing Indian Policymaker Discourse on AI Sovereignty
- URL: https://aipriorities.digitalfutureslab.in/who-is-the-sovereign-analysing-indian-policymaker-discourse-on-ai-sovereignty/
- Published: 2026-10-03T18:07:21.000Z
- Updated: 2026-10-03T18:07:21.000Z
- Author: Hari
- Tags: #Import 2026-10-12 10:52

> Purnima Rohilla and Anupam Guha

**Abstract**

In recent years, there has been a significant increase in discussions regarding AI sovereignty among Indian policymakers. Before one can build institutions, frameworks, and governance mechanisms for AI in India around the question of sovereignty, there is a need for empirical work to understand what AI sovereignty means to the Indian policymaker. This essay operates within the broader research problem of how Indian policymakers conceptualise the governance of AI. It recovers themes of dominant discourse, including recourse to techno-solutionism and legalism, points of convergence between policymakers, extant vulnerabilities, and concepts that are vital but missing from Indian policy discussions on AI sovereignty. It uses Lok Sabha proceedings, PIB releases, and IndiaAI mission texts. The essay ends with a commentary on the extant policy discourse and potential new directions for AI policymaking.

**Introduction**

Among organisations and individuals engaged in debates on technology policy in India, one concept recurs with particular urgency: “AI Sovereignty.” Policymakers repeatedly argue that India requires sovereignty in its use of AI. Discussion papers from state-aligned think tanks like NITI Aayog and public consultation documents from ministries like MeitY enthusiastically invoke this concept when listing out their organisational goals. Even where the phrase is not explicitly used, related aspirations emerge around developing “AI infrastructure,” i.e., compute, datasets, and models “accessible” to national businesses in the context of developing self-reliance and state capacity. Despite the prevalence of this discourse amongst policymakers, there does not exist any single statement, white paper, or strategy document issued by any Indian government ministry that comprehensively defines the concept of AI sovereignty, identifies its red lines, or outlines a roadmap for its realisation. In these discussions, AI Sovereignty is uncritically treated as a coherent and self-evident virtue whose urgent realisation requires no justification. This essay examines what the concept means to Indian policymakers and the constellation of ideas within which it is discussed. It concludes by proposing alternative ways of framing the debate that place the rights of the people – the ultimate source of sovereignty – at its centre. 

**What is AI Sovereignty?**

An understanding of AI Sovereignty in Indian policymaking requires situating the concept within existing literature on AI sovereignty, and the larger debate around technological or digital sovereignty. Since Barlow’s techno-utopian manifesto in 1996, which theorised the concept of cyberspace beyond the notion of state sovereignty, the debate has evolved considerably. Contemporary scholarship has raised sophisticated questions about sovereignty in cyberspace, and debated who controls data and where it should be stored, whether the Internet should be a territory of state control, and if it should be fragmented along state lines. More recent literature recognises the vague and market-like usage of “AI” and the significant overlap of AI and the merely digital. This literature on AI (or digital) sovereignty can be broadly divided into two strands.

The first strand builds frameworks defining AI (or digital) sovereignty and measures how a state performs against these frameworks. Within this literature, sovereignty becomes a stand-in for state capacity, measured against a list of developmentalist features like governance capacity, infrastructural capacity, electricity, security measures, etc. One such example is the KASE framework, which examines AI development and governance on the basis of a list of factors theorised as enabling digital sovereignty. Various commentaries examine how India’s AI policy measures up on this framework, and how its features can be modified for the Indian context. Another study critically examines what the authors call India’s AI Industrial Policy. Collectively, these works use as their data sources documents from sources like NITI Aayog and MeitY. They converge on the view that policymaking around sovereignty is primarily concerned with the development or purchase of computational power, acquisition (via enclosure) and “making available” of large collections of data, and the development of connectivity. At the same time, they recognise that there are concrete gaps between what is stated in their data sources and what is technically feasible, while also highlighting the limited attention paid to legal and notional rights. The second strand of the literature on digital sovereignty comprises theoretical discussions that explore the meaning of the concept, question its legitimacy, and critically examine the assumptions on which it rests. While sovereignty under capitalism has never been absolute, a substantial body of work argues that the meaning of “sovereignty” for the digital has significantly diluted the notion of state control. The literature recognises that, discursively, the concept of digital sovereignty is used flexibly to protect notions of rights of collectives and individuals by framing the threat as encroachment by a foreign actor. The word “sovereignty” itself has been used to legitimise action by weaker agents against various kinds of hegemony. At the same time, the literature indicates that the word is also used to legitimise and reinforce the status quo of a global alliance between states and tech capital. This tension is particularly visible in debates on AI sovereignty. Global technology firms like NVIDIA and OpenAI seemingly become champions of “sovereignty,” with firms like Alphabet and Microsoft undertaking “digital sovereignty pledges” and offering “sovereignty-as-a-service.” Yet the US Clarifying Lawful Overseas Use of Data (CLOUD) Act casts doubt on such claims, since it permits the US government to compel access to data held by US cloud providers regardless of where the relevant data centres are located. 

This apparent contradiction becomes easier to understand when viewed from the perspective of the relationship between states and AI firms. AI firms need extraordinarily large monetary investments and other forms of assistance from host states that extend beyond the assistance states typically provide capital. In exchange for helping build their sovereignty, states offer these companies everything from data enclosure policies to building data centres that are resource-intensive and impose on the rights of local residents. To attract companies, states risk creating dependencies and vendor “lock in.” While the states involved gain discursive power from these policies – using sovereignty to legitimise policies of development, securitisation, enclosure, and surveillance – the consequence may be the disciplining of citizens and the erosion of their rights.

These and other contradictions have led some scholars to question, and even reject, the category of digital (and thus AI) sovereignty. Some find the category sufficiently distinct from Westphalian sovereignty and advocate a study of the agendas behind its framing, while others characterise it as a mechanism to obfuscate state overreach and reproduce a violent national and global order. Still others view it as a form of mystification designed to facilitate the very global digital capitalism that it claims to regulate, by promoting the urgent and blind adoption of exploitative and unproven technology systems. Given this literature, this essay adopts the methodological position that AI sovereignty should be analysed as a distinct concept rather than assumed to correspond to conventional understandings of sovereignty when examining its meaning in the Indian policy context.

**AI Sovereignty and the Indian Policymaker**

AI policymaking in India is an evolving field. For the purposes of this essay, we constrain our definition of an “AI policymaker” to three central government sources: Members of Parliament, official press, and the government’s primary AI “mission.” Accordingly, our data is collected from three primary sources connected to the Indian government: 1) the Question Hour of the Lok Sabha, the lower house of the Indian parliament; 2) Press Information Bureau releases, the PIB being the nodal agency of the Government of India to disseminate information to the media; and 3) articles from the IndiaAI mission website. The IndiaAI mission serves as the Indian government’s primary vehicle for signalling policy intent and disbursing funding for AI. A limitation of this study is that it does not analyse texts from various projects in state and local governments, or from executive bodies, statutory bodies, and the judiciary, all of which are also actively shaping AI policymaking in India. We limit our discovery process to documents between the 17th and 18th Lok Sabha sessions, corresponding to the period from June 17, 2019 to the time of writing. This period begins after several institutional developments, including the formation of the AI Task Force by the Union Ministry of Commerce and Industry and the framing of its vision document in March 2018, the launch of MeitY’s roadmap for a national AI programme in February 2018, and NITI Aayog’s release of its National Strategy for Artificial Intelligence in May 2018\. 

As documents relevant to sovereignty discourse may not explicitly use language relevant to it, we first identified documents containing AI-related text, and filtered out those irrelevant to our analysis. We searched for documents containing one or more of these key phrases (accounting for capitalisation): artificial intelligence, machine learning/ ML, generative AI, AI governance, AI infrastructure, Large Language Models/ LLM, Sovereignty/ Sovereign, national AI, AI nationalism, Indian AI, AI strategy, localisation/ localization, Data sovereignty, Make in India, decolonial, and global south. This keyword list was developed from the literature reviewed in the previous section. Using this approach, we identified 52 documents from the 17th session of Lok Sabha, 36 documents from the 18th session of Lok Sabha, 90 PIB releases, and 146 IndiaAI articles, yielding a corpus of 324 documents. All of these documents were then manually reviewed to filter out irrelevant text, resulting in a final dataset comprising 11 documents from the 17th session of Lok Sabha, 10 documents from the 18th session of Lok Sabha, 12 PIB releases, and 22 IndiaAI articles – yielding a total of 55 documents.

Given the sparsity of data and the prevalence of common bureaucratic jargon, quantitative or computational methods like topic modelling and frequency of bigrams were unlikely to provide meaningful insights. We therefore applied a variant of thematic analysis on the documents, searching for common themes or discourses that we coded, clustered, and identified in the text. Guided by the literature reviewed in the previous section, we identified the following themes for analysis.

**Largesse for Compute-Heavy LLM Development**

This is the most prevalent theme in the Lok Sabha questions directly asking about the details of the IndiaAI mission, and statements like *“The financial outlay of the mission is Rs 10,371.92 Crore”* become the centrepiece of the discussion. Although the documents record frequent revisions to budget allocations, they contain little discussion of concrete policy goals or subsequent implementation. Details on building of domestic compute production and progress in machine learning research are missing, despite the existence of the India Semiconductor Mission as well as production- and design-linked incentive schemes. A PIB release in December 2024 announced the intent of acquiring 10,000 GPUs.This position was later revised. Rather than procuring GPUs directly, the government announced that approximately half of the IndiaAI Mission budget, around Rs 5000 Crore, would subsidise GPU procurement by Indian firms, although this fact, announced to the press by MeitY, was not present in any of the documents in our dataset. Finally, a PIB released on 30 December 2025 announced that *“India has now achieved 38,000 GPUs, providing affordable access to world-class AI resources,”* though the ownership structure of this compute capacity is not clarified in the data; the documents consistently conflate the Indian state with private ownership. On the issue of models, the documents mention that *“Four (4) startups – Sarvam AI, Soket AI, Gnani AI, and Gan AI –have been selected (so far) to build India’s foundation models”* and although the contentious industry terminology of “foundation models” is used, it is asserted that these models will be eventually made open source. A broad indication of strengthening academic research exists in 2019: the documents talk of NITI Aayog’s proposal of building *“Centers of Research Excellence in Artificial Intelligence (COREs) to specialise in creating new knowledge through basic research and source fundamental knowledge /technologies that will be needed to keep India prepared for the next generation of technologies.”* A funding plan for five years also sees mention. This, later in 2024, is modified into a plan to fund *“three Centres of Excellence (CoE) in Artificial intelligence (Al), one each in the areas of health, agriculture and sustainable cities with a total financial outlay of Rs. 990.00 Cr over the period of FY 2023-24 to FY 2027-28.”* The documents do not contain discourse around building basic or theoretical AI research, nor is there any intent of bolstering research in academia, with high research imagined to be the development of industry-led, compute-intensive LLMs. The documents constantly conflate and promotes AI research with LLM development, e.g., in April 2025 a Lok Sabha answer states *“Till 15th March, IndiaAI Mission has received a total of 187 proposals aimed at building India’s foundation models, with contributions from both established startups and new teams of researchers & academia. Along with funding support, a wide range of GPUs have been requested by teams submitting these proposals”.*

**Techno-solutionism/ Techno-legalism/ Techno-optimism**

Throughout the documents, the second most prevalent theme is confidence in the existing legal framework and technological innovation in solving the social issues around building AI sovereignty, and a strong belief that AI provides the pathway to national prosperity and security. AI is always seen as a tool to solve complex social issues; for example a PIB release states *“AI can help us to analyse the mobility patterns of our migrant workforce”*. The Lok Sabha questions demonstrate an intent to govern AI through extensions of existing digital laws rather than any future standalone statute, using intermediary liability and data protection as its main regulatory levers. The recent guidelines drafted by the Subcommittee on AI Governance and Guidelines Development under the Principal Scientific Advisor (PSA) of India, though referencing sectoral rules, does not contradict this tendency. The documents repeatedly assert that the expansion of an AI-led economy will generate more employment than it displaces, e.g., *“AI may result in automating some routine jobs but will also result in job creation in various data science, data curation, etc”*. There is no reflection beyond the binary of job creation and displacement, including on structural issues with wages or quality of employment. While NITI Aayog has drafted a “Roadmap for Job Creation in the AI Economy”, it is not referred to in these discussions. Confidence in employment-related issues being solved via re-skilling and vocational training is seen repeatedly. We found no discourse addressing the transnational nature of AI infrastructure needing any special regulations, a point that needs to be addressed while pursuing sovereignty. Both laws and technology are framed here as levers removing obstacles from capital, and the expansion of private sector AI is framed as a solution to complex policy problems and the harbinger of general prosperity. A PIB release in October 2022 declares *“Artificial Intelligence will help the country reach that destination and bring prosperity to every citizen of this country”*. As before, there is a general theme of optimistic vagueness. In some of the documents, a new organisation called the AI Safety Institute is announced, *“to ensure that the proposed AI Safety Institute adopts a human rights-based approach with a focus on openness, innovation and inclusivity while safeguarding people’s data and privacy”*. This institute under MeitY was launched in January 2025, though it isn’t clear from these documents what its mandates, audit powers, or redressal mechanisms are. 

**Market Fundamentalism and Legitimising Data Enclosures**

There is a general theme throughout the dataset affirming that the State needs to act to protect its sovereignty, but these actions must always be in collaboration with digital capital, with ease of doing business being a primary goal. Privacy is repeatedly qualified by the assertion that it should not become a barrier to development. For example, these two statements from a PIB release from July 2020 simultaneously assert a need for sovereignty and the preservation of global supply chains: *“In the wake of global pandemic like COVID-10, Mr. Prasad strongly emphasised the need for building a resilient global supply chain and shared the vision of Prime Minister of India for making India an attractive destination for investment closely integrated with global supply chains”* and *“The Minister specially emphasised on the data related issues and sovereign rights of countries to protect data privacy and security of its citizens.”* Also, the same document states, *“India is soon going to put in place a robust personal data protection law which will not only address the data privacy related concerns of citizens but also ensure availability of data for innovation and economic development.”* A Lok Sabha answer similarly states,*“ The government has taken an inclusive and innovation-friendly approach to Artificial Intelligence (AI) governance. It is backed by an institutional setup to protect citizens’ rights, ensure data privacy, and promote innovation.”* The data indicates strong dependency on foreign capital like Microsoft and Nvidia, which is relevant given the literature discussed in the last section on these companies’ appropriation of the concept of sovereignty. An implicit assumption running through these documents is that regulation constrains innovation; innovation is understood to occur primarily through the private sector, facilitated by access to community data and public subsidies for domestic firms.

**Geopolitical or Techno-Nationalist Aspirations**

An important underlying theme in this dataset is AI as a source of national pride and culture building. There are multiple invocations to *“India’s leadership”* in AI being important to maintain, though it is not qualified how. *“Global south”* is invoked multiple times, but specifics on this are missing, and the exact relation being sought for with other states is not elaborated. Despite its presence in sovereignty debates in the literature, China is mentioned rarely in this dataset. A PIB release quotes a speech declaring that *“India is the main player in the field of AI talent and AI-related ideas”*. One recurring pattern in the data is the conflation of research capability in AI with “skill” in AI-related tasks among Indian engineers. A repeated data point is quoted: *“The Stanford AI Index 2024 highlights that India tops global AI skill penetration rates and AI talent concentration. India's AI skill penetration is rated at 2.8, surpassing the United States (2.2) and Germany (1.9). Additionally, India's AI talent concentration has seen significant growth, with a 263% increase from 2016 onwards”*. Because this indicator is derived from self-reported LinkedIn profiles, it provides only a limited proxy for national AI capabilities.

A close reading of the documents in our datasets also reveals the conspicuous absence of certain themes, which are considered important in the literature described in the last section. First, policymaker discourse flattens data into a homogeneous category, and does not specify whose data or what kind of data will be collected and concentrated to enable domestic capital. Second, the future of labour is a mere footnote in discussions on AI among policymakers, with reskilling optimistically seen as benefitting the future of work. While there are multiple instances of questions asked on the economic impact of AI in the Lok Sabha, the responses sidestep the core issue. Finally, the issue of rights and democratic oversight is framed statically as a question of implementing the correct “guardrails” without any vision regarding the expansion of rights in this imagined realm of future prosperity. Ultimately, the sovereign imagined in these policy documents is neither the citizen nor, indeed, citizens' elected representatives. Rather, sovereignty is increasingly vested in a market-oriented technological project whose inevitability is taken for granted and can only be managed, rather than questioned.

**Reframing the Discourse Around the Sovereignty of the People** 

This essay does not seek to provide a definitive or universal account of AI sovereignty, nor does it attempt to determine whether the concept can be sufficiently rescued as an analytically meaningful category. Instead, it advances parsimonious claims that set the boundaries of strategic policymaking. The first is that “foundation models” are anything but foundational beyond modelling text. Framing LLMs as the cornerstone of AI research has been critiqued; and these models might not be central to machine learning research, let alone the path to universal reasoners or the ill-defined notion of “Artificial General Intelligence.” Second, AI research is not dependent on a direct relation with GPU availability. Advances in model architectures continue at a rapid pace that does not map directly onto increases in computational power. Indeed, the LLM architectures tied to GPU access popular now will likely be obsolete before investments of public resources yield meaningful returns. This means directing the bulk of public investments in AI on LLMs and GPU procurement to subsidise companies is a myopic approach. 

Another technical consideration is that machine learning systems are by design stochastic, while sophisticated neural networks are mostly opaque. Public procurement and deployment of these systems need to fulfil two criteria. First, systems must work as advertised and not function as “snake oil.” Second, governance frameworks must ensure that the inevitable errors associated with stochastic systems do not cause harm to individuals and their rights. In short, AI systems cannot be treated as bandaids to social problems. Addressing these concerns should form a central mandate of emerging AI Safety Institutes, which must have actionable regulatory power beyond issuing ethical guidance.

If policymaking is to be aligned with how the technology actually functions, it needs to stop the blind pursuit of priors presented as established facts by the AI industry. Recent AI research demonstrates the continued importance of open source approaches for research breakthroughs. Sustained public investment in technical and humanistic education, academic research, international collaboration with academics and research groups and laboratories working across the theoretical, application, and the social dimensions of AI remains imperative. Equally, the sovereignty of people cannot be protected if their labour power is exploited – whether through the expropriation of the data they generate or via semi-automated management systems. 

Policy responses should therefore include the recognition of platform and gig work as employment accompanied by labour protections, regulatory intervention to stop the farming of datasets where skills are being recorded, the decommodification of AI research via the formation of technology cooperatives, and the strengthening of workplace privacy protections. While an in-depth analysis of Indian data laws lies beyond the scope of this essay, current protections do not extend to data collected within workplaces, where the privacy interests of employers frequently take precedence over those of workers. This imbalance needs to be reversed. The prospect of workers contributing to development in ways that ensure their own obsolescence, as in the case of ghost work done by data annotators, presents a significant long-term policy problem. At a minimum, policymakers should consider stronger labour protections, including minimum wages and social security for data and AI workers. More ambitious interventions, such as worker ownership models or basic income, may also warrant consideration. 

Ultimately, sovereignty cannot be understood as merely a question of industrial or data policy, but one of labour policy. Whether AI companies are global or domestic, the state should avoid enabling them uncritically at the expense of workers’ rights. A structural policy push here will involve institution building which regulates data collection and usage by digital companies, and investigating alternate ownership structures of the production and distribution of AI research, models, and artefacts beyond conventional market logics.