In mid-February, Delhi hosted heads of state, ministers, technology executives, and some civil society representatives for the India AI Impact Summit. The meeting was the fourth in a series of yearly international AI Summits, held for the first time in a Majority World country.
The Delhi event was framed as an opportunity to “move beyond aspirational frameworks toward concrete, measurable impact that addresses both AI’s promise and its perils.” But examining the outcomes of the Delhi gathering reveals a tension between this (aspirational) rhetoric and the structural constraints facing so-called middle powers—states that seek greater strategic autonomy in AI but lack control over semiconductor supply chains, hyperscale cloud platforms, and frontier model labs that currently determine the trajectory of AI development.
Declarations
The principal formal outcome of the summit was the Leaders’ Declaration, endorsed by participating governments, including the European Union.
The Declaration puts forth that “AI’s promise is best realized only when its benefits are shared by humanity,” framing AI as a vehicle for broadly distributed economic and social gains, if done well. Its rhetoric emphasizes collaboration, trust, and social good, and avoids direct reference to structural power asymmetries. (Notably, the only instance in which the word “power” appears in the Declaration is in the term “empowerment.” )
The Declaration does not impose binding legal obligations. Amnesty International criticized the summit for failing to introduce enforceable measures addressing harmful uses of AI by governments and corporations, including biometric surveillance, predictive policing systems, and opaque algorithmic decision-making in public administration. The divergence between these concerns and the promise-driven tone of the Declaration reflects a broader pattern in global AI forums: shared principles are articulated and “perils” are invoked in general terms, but concrete and binding mechanisms to address them are often absent.
Voluntary Commitments
In addition to the Declaration, the Indian government announced the New Delhi Frontier AI Commitments, a set of voluntary agreements endorsed by Indian and global frontier AI firms. The Commitments focus on two areas.
The first commitment, “Advancing Understanding of Real-World AI Usage,” centers on generating anonymized and aggregated data on how AI systems are deployed across sectors. Participating organizations commit to collaborating on evidence generation regarding AI’s impact on jobs, productivity, skills, and economic transformation. The stated objective is to enable “data-driven analysis of how AI is being deployed across sectors” in order to “help governments and institutions craft informed strategies that maximize benefits while mitigating risks associated with technological change.”
This is a problematic framing.
While crafting informed strategies is undoubtedly a worthwhile objective, in the case of AI deployment, the scale of the current infrastructural buildout cannot be ignored. The New Delhi Frontier AI Commitments naturalize AI deployment and treat it as if it were a consequence of a natural phenomenon, of something that will happen regardless, rather than as the result of concrete decisions. As a consequence, evidence generation becomes a retrospective exercise used to rationalize and tweak AI adoption rather than a forward-looking effort to anchor AI in concrete, problem-driven priorities. The commitment shifts the focus from whether and where AI should be deployed to how its impacts can be measured and managed after the fact. Prioritizing the supply of AI technologies without clearly articulated needs accelerates the race for scale and infrastructure, rather than grounding these costly developments, financially, socially, and environmentally, in meaningful priorities.
This is a pernicious dynamic that we investigate in our Steering AI work.
In contrast to the first, the second commitment, “Strengthening Multilingual and Contextual Evaluations,” tackles a concrete and identifiable challenge. It addresses disparities in AI system performance across languages and cultural contexts. Under this commitment, participating companies agreed to collaborate with governments and local ecosystems to develop datasets, benchmarks, and expertise to evaluate AI systems in underrepresented languages and contexts.
Investment Announcements
The summit also featured major investment announcements focused on infrastructure, energy, and computing capacity. India’s Electronics and IT Minister projected over $200 billion in AI and deep-tech investment over the next two years. Major Indian conglomerates announced large-scale commitments to build AI and data infrastructure. For example, India’s Adani Group pledged about $100 billion to develop AI-ready data center infrastructure, while Reliance Industries announced plans to invest around $110 billion over the next several years to build sovereign AI infrastructure and gigawatt-scale compute capacity. American companies expanded their commitments as well. Microsoft indicated it is “on pace to invest” $50 billion across the Global South by 2030, and AMD strengthened its infrastructure partnership with Tata Consultancy Services, while Google made multifaceted commitments focused on infrastructure, connectivity, AI research funding, and collaboration with the Indian government.
These announcements highlight the physical foundations of AI development: data centres, advanced processors, grid capacity, and large-scale infrastructure investment. They also reinforce the central role of US-based technology firms in supplying critical components across the AI stack, particularly at the hardware and cloud orchestration layers. They point to a deepening integration between US and Indian industry.
The Sovereignty Gap
At the Delhi Summit, representatives from both the EU and India expressed their desire for a more sovereign approach to AI development. For now, however, the ‘AI stacks’ of both the EU and India are firmly embedded in supply chains that are largely controlled by a handful of US technology companies, with additional dependencies on Chinese-controlled critical mineral processing and hardware manufacturing.
The different nature of the US-India and EU-India partnerships announced in Delhi illustrates this reality. Cooperation between the US and India centres on infrastructure, corporate partnerships, and supply chain integration. Agreements have been made to link Indian firms with US frontier AI companies, with a focus on providing access to advanced processors and large-scale computing capacity. This will lead to India becoming more deeply integrated into existing global AI supply chains. This trend is further illustrated by India’s joining of Pax Silica, a US-led initiative that aims to strengthen coordination and resilience across AI supply chains among selected partners focused on reducing what the declaration calls “coercive dependencies” across the semiconductor and critical minerals ecosystem. While China is not named explicitly, it is the clear target of this coalition.
EU-India collaboration on AI operates at a different level. The presence of many European heads of state or government at the summit emphasized a strong political commitment, with public statements characterizing the relationship as long-term and value-based. European AI executives, including Arthur Mensch of Mistral AI, expressed concerns about dependence on external providers and the concentration of AI power in a small number of US companies. Mensch’s argument captured a broader European anxiety about sovereignty that provides context for the EU’s growing engagement with India.
The conclusion of negotiations on an EU–India Free Trade Agreement (pending ratification) and the signing of a Security and Defence Partnership provide an institutional framework for cooperation in digital trade, data governance, and technology-related procurement, all of which could, over time, enable more structured AI collaboration. Bilateral initiatives, notably between France and India, further signal intensified cooperation. For now, however, EU-level cooperation in AI is modest in scale (compared to India-US) and consists primarily of policy dialogue and initiatives such as the European Legal Gateway Office focused on ICT talent mobility rather than large-scale joint AI infrastructure or compute integration.
Challenges and Opportunities Ahead
Both the EU and India share concerns about AI sovereignty. Addressing this challenge faces obstacles related not only to limited technological capacity, but also to geopolitical realities. Michael Kratsios, the White House representative present in Delhi, framed AI sovereignty as compatible with deep integration into US-led technology ecosystems, confirming the US’s unwillingness to cede dominance over the “AI stack”.
India’s joining and the European Union’s half-in, half-out position on Pax Silica—the EU participates in discussions but is not a signatory, and the Netherlands, despite hosting ASML, has not signed either—reveal how elusive the desire for a more sovereign AI is. This was made visible once again at the Delhi Summit. Despite the technological and geopolitical challenges, the AI middle powers—including the EU—cannot simply acquiesce to the US position or afford a “business as usual” approach. Unless they begin to address the dependencies across the AI hardware and software stack and chart their own path for AI, they will become even more entrenched in the development paradigm set by the US AI labs, thereby exacerbating the pressures they already face.
