Amid pressure from Big Tech to control the technology, the Global South excels in rhetoric but avoids triggering collective action and takes dangerous bait. Delhi tilts towards the US. Brazil signs a document Lula viewed with reservations—and may end up isolated.

Nearly 90 countries and participating organisations at the India AI Impact Summit, held in New Delhi last week, signed a leaders’ declaration that fell far short of the needs of the major developing digital markets. Unlike the preceding editions of the event created by the United Kingdom, many had hoped the gathering would mark a turning point in how the world thinks about the governance of artificial intelligence—a genuinely different perspective from that produced in forums led by countries of the Global North. The expectation made sense because, for the first time, a country from the Global South was occupying centre stage. The outcome, however, disappointed observers more closely attuned to the global AI landscape, and was endorsed by the Brazilian government. The Leaders’ Declaration that emerged is more principled than normative and, by remaining silent on the most urgent structural issues, ends up legitimising precisely the state of affairs it ought to challenge. But there is an even more troubling element that remained in the shadows: on the margins of the summit, India signed its accession to the Pax Silica—the American pact on AI and semiconductor supply chain security—consolidating a geopolitical realignment whose consequences for the Global South and for Brazil deserve urgent attention.

Effective sovereignty

The leaders’ declaration organises its vision of AI governance around seven “Chakras”—pillars encompassing human capital, social access, trustworthiness, energy efficiency, science, democratisation, and economic growth. The tone is aspirational and philosophically anchored in concepts from Indian culture, such as Vasudhaiva Kutumbakam (“the world is one family”). Beyond the rhetoric, however, what the declaration proposes are voluntary and non-binding platforms and instruments—a Trusted AI Commons, a Global AI Impact Commons, an International AI Network for Scientific Institutions—as though informal cooperation were sufficient to rebalance a technological race whose stakes are deeply concentrated in the hands of a handful of American and Chinese companies.

The most glaring omission concerns the digital sovereignty of the Global South. The document mentions “national sovereignty” only in passing and in a perfunctory manner, with no concrete mechanism to ensure that developing countries can build AI autonomously—rather than merely adopting whatever is on offer from companies in the Global North. There is no mention of technology transfer, of compulsory licensing of models, of economic regulation of Big Tech’s market concentration, nor of any instrument that might prevent the deepening extraction of data from developing countries without compensation. There is not even any mention of the Global South or developing countries.

The phrase “democratisation of AI resources” is emptied of meaning when the entire proposed structure is voluntary. What has historically reproduced technological dependency is not a lack of inclusive rhetoric, but the absence of binding obligations. The emphasis on “open source where appropriate” is vague enough to commit to nothing. Decentralised infrastructure does not mean technological autonomy, since cloud services may be located in a developing country whilst their control remains under the aegis of a Big Tech firm—as is already happening.

The ignored bubble

Equally worrying is the silence on the risks of the AI financial bubble. The document completely ignores the macroeconomic consequences of the current hyper-inflation of investment in the sector. Billions of dollars have been poured into AI infrastructure on the basis of return projections that have yet to materialise. For emerging economies attempting to “ride the wave” of this speculative cycle—pressured by narratives suggesting that staying out would mean condemning themselves to backwardness—the risk is particularly severe. Previous technology bubbles left a trail of public and private debt in countries that bet on the technology of the moment without any critical cost-benefit assessment.

The declaration does not devote a single line to this risk. By banking on voluntary, industry-led instruments, the document goes beyond mere silence: it implicitly legitimises the self-regulatory model of the very companies whose inflated valuations are pressuring governments to subsidise AI infrastructure without any independent scrutiny.

Unilateral governance

But perhaps the most consequential—and most revealing—choice in the declaration is institutional in nature. The document strips the United Nations of central responsibility for AI governance and places its bets on the multi-stakeholder model, that is, a governance shared among states, companies, and other non-state actors, without any clear hierarchy of responsibilities and with decisions made by consensus. This is not a neutral choice: it is precisely the model that shaped internet governance over the past few decades, with results that historical experience itself allows us to assess.

ICANN, the internet’s principal technical coordination body, was born with the rhetoric of multi-stakeholderism and produced, in practice, a progressive corporate capture, marked by a lack of transparency and control by a single power. The transition from North American state oversight to this model did not represent a victory for globalism, but the replacement of state control with control by the United States’ major technology companies. New domains were sold at prohibitive prices; civil society was treated as a decorative layer; accountability mechanisms proved largely performative; and nation-states have no active voice in its decisions. The result was the consolidation of an internet whose logical, economic, and infrastructural layers remain under the unilateral control of a small group of countries and corporations.

For AI, the risk is analogous and potentially more serious. The extremely high capital requirements for developing large language models make multi-stakeholderism even more fragile than it was in the case of the internet: without billions of dollars in processing power, data, and talent, there is simply no way to sit at the table as an equal. The model proposed by the declaration thus paves the way for bilateral governance between Big Tech and major states—with the Global South reduced to the role of consumer and data supplier. The so-called AI Summits risk repeating the trajectory of the WSIS—the World Summit on the Information Society—a process created in 2003 that dragged on for decades of rhetorical stalemate without any concrete results in terms of redistributing power over digital infrastructure.

The choice between multilateralism and multi-stakeholderism is not technical; it is political. Multilateralism, for all its flaws, anchors responsibilities in sovereign states and preserves the possibility of binding mechanisms negotiated under conditions of formal equality. Multi-stakeholderism, as the history of the internet has demonstrated, tends to dilute that formal equality into real asymmetries—where the weight of each actor depends less on its democratic mandate and more on its economic and technological power.

Pax Silica: the shadow behind the Summit

On 20 February 2026, on the margins of the multilateral summit, India signed its accession to the Pax Silica—an initiative led by the United States Department of State for the security of AI and semiconductor supply chains. The agreement was signed by the Secretary of India’s Ministry of Electronics and Information Technology, Shri S. Krishnan, and the US Under Secretary of State for Economic Growth, Energy, and the Environment, Jacob Helberg, making India the 12th member of the coalition.¹ At the same time, India and the United States signed an AI Opportunity Partnership, at an event attended by the CEOs of Google, Sundar Pichai, and OpenAI, Sam Altman. The multilateral ceremony ended. Shortly afterwards, the real business began.

The Pax Silica is described by the Department of State as the “principal US effort on AI and supply chain security”, designed to align partner countries across the full technological spectrum—from critical minerals and energy to semiconductor processing, AI infrastructure, and logistics. If the twentieth century ran on oil, the twenty-first runs on computing and the minerals that power it: this is the explicit strategic logic of the initiative. Although Under Secretary Helberg stated publicly that the Pax Silica “is not about China, but about America”, the architecture of the agreement—based on export controls, coordinated subsidies, and reduced dependence on manufacturing hubs dominated by Beijing—leaves very little doubt as to who the implicit adversary is. India’s accession is politically significant for an additional reason: the country is a member of the BRICS, which means the largest democracy in the Global South has just formalised its alignment with the technology bloc led by Washington.

Brazil and the Global South outside the real game

The combination of the two events—the New Delhi Declaration and the signing of the Pax Silica—reveals a geometry of power that the summit’s official discourse deliberately obscures. On one side, 88 countries sign an aspirational declaration, full of non-binding principles about the democratisation of AI. On the other, twelve countries—selected on the basis of their strategic role in the technological supply chain—sign a concrete agreement that will define access to advanced chips, manufacturing partnerships, and the flows of capital that will shape AI over the next decade. Brazil, which signed the New Delhi Declaration, is not in the Pax Silica. The difference is not symbolic: it is the difference between sitting at the table where principles are celebrated and sitting at the table where the rules of the game are actually written.

For countries such as Brazil, a major digital market, the absence from the Pax Silica is not necessarily a strategic disadvantage in the short term—after all, the agreement carries obligations of alignment with export and security policies that entail delicate geopolitical choices. But the absence also means being left out of a “concierge service” that the Department of State itself publicly described as a preferential mechanism of access to advanced American semiconductors, through which US diplomats will actively act as business development agents to facilitate the acquisition of chips by allied countries.² In a world where access to the latest-generation GPUs can determine a country’s sovereign capacity to develop AI language models, being excluded from this service is not an operational detail. It is a structural asymmetry that deepens with every cycle of innovation. For the Global South as a whole, the Pax Silica consolidates a model in which developing countries are invited to sign declarations about “AI for all” whilst real access to the tools that produce AI is managed by a restricted club of countries aligned with Washington.

The India Summit could have been a historic opportunity to break that pattern. The host country had the legitimacy to speak on behalf of a significant portion of the Global South and sufficient political capital to propose governance architectures that did not repeat the mistakes of the internet. It did not do so. Instead, it signed the Pax Silica on the very same day that the multilateral declaration was adopted—signalling that its real priority lay elsewhere. Brazil is another country that could make a difference. President Lula’s speech pointed towards paths less subject to the status quo and could signal a shift in the game of pressure from corporations. But the time available to build alternatives is finite: every round of summits that passes without binding mechanisms and without genuine representation of developing countries at the core of technological decision-making is a lost opportunity that can scarcely be recovered. It is known that Lula was reluctant to sign the summit’s leaders’ declaration, disagreeing with the absence of references to multilateralism and the United Nations, but Brazil ended up endorsing the document at the last minute.

What the New Delhi Declaration delivered is an architecture of cooperation that is unlikely to alter existing power asymmetries—and that remains silent on the most urgent systemic economic risks that any gathering of world leaders ought to address in 2026. The multilateral summit has ended. The Pax Silica has begun. And that distance, far more than any declaration of principles, is what reveals the true geography of power in digital governance for the coming decades. The question that remains is whether there is still room for countries such as Brazil to lead the effort to reverse this logic of segregation—crowned last week in the Asian country—and to place the Global South at the table.