The New Delhi Declaration, signed by over 90 nations and organisations at the India AI Impact Summit 2026, articu­lates a shared commitment to shape Artificial Intelligence as a tool whose benefits must be distri­buted across societies rather than concentrated among a cluster of influential nation-states and corporations. Yet, democratising AI for social welfare, economic inclusion and human capacity building sits uneasily within a competitive global order. Advanced AI depends on scarce resources: high-end chips, large-scale computing infrastructure, proprietary data and specialised talent. These are treated as strategic assets by nations. AI firms, too, guard intellectual property to scale their advantages. As such, access to AI is shaped by market leverage, geopolitical heft and bargaining power.

The idea that AI can function as a public good thus requires scrutiny. By definition, a public good is non-rival and non-excludable. AI systems do not usually meet these criteria. Training and deploying models require finite compute and energy. Access to leading systems is restricted through pricing, licensing and infrastructure controls. Even open models demand capital, expertise and connectivity. The free-rider problem of public goods — people benefiting without paying for such goods — also applies to AI. Countries and companies may gain from shared safety resear­ch and common standards without contributing proportionately to costs or complying with constraints. It is therefore premature to conclude that AI can be turned into a public good in the truest sense.

However, certain components of AI, such as open benchmarks, safety protocols and interoperable digital infrastructure, can approximate public goods if supported by State finance. Whether that occurs depends on political will and sustained investment. Regulation is central to democratising AI as well. Effective oversight must address such concerns as opaque model development, cross-border deployment, liability for harm, data protection, competition, and the environmental costs of requisite infrastructure. Moreover, as deduction-based AI grows more capable of reasoning and autonomous action, control over such systems will determine where power resides. Democratisation in this context demands mandatory audits, independent evaluation, transparency obligations and clear restrictions on high-risk applications. Public authorities will need to be armed with technical competence and legal authority to scrutinise advanced models. Without these guardrails, the language of shared benefit can actually mask continued concentration of capability. The NDD signals intent. The outcome of its pledge will rest on whether regulation and governance keep pace with technological enhancement.