The New Delhi AI Impact Summit brings the US back to the table (2026)
At the first global AI summit held in the Global South, more than 80 countries including the US, China and the UK endorse a non-binding declaration on democratising AI, while Washington openly rejects any global governance of the technology.
A year after the Paris AI Action Summit exposed a split over global rules, India hosted the AI Impact Summit at Bharat Mandapam in New Delhi. The five-day meeting, which ran from February 16 to 20, 2026, with leaders gathering on February 18 and 19, was described by Reuters as the first global AI summit in a developing country. Prime Minister Narendra Modi framed the theme as welfare and happiness for all.
Seven pillars, all voluntary
The summit produced the New Delhi Declaration on AI Impact, dated February 19, 2026, organised around seven “Chakras” (pillars): democratising AI resources; economic growth and social good; safe and trustworthy AI; AI for science; access for social empowerment; human capital; and resilient, efficient, innovative systems.
The annex listed voluntary outcomes such as a Charter for the Democratic Diffusion of AI, a Global AI Impact Commons, a Trusted AI Commons, and guiding principles on resilient AI. Nothing in the text was binding. Countries could endorse the vision without committing to shared enforcement.
India’s press office reported more than 80 countries and organisations on the signatory list in the days after the summit, including the United States, China, the United Kingdom, the European Union, Russia, Germany, and France. Exact counts shifted as late endorsements arrived; later tallies ranged into the high eighties and low nineties, so contemporary reports often used the conservative phrasing “more than 80.”
Everyone signs, Washington still says no
The reversal against Paris was striking. In February 2025 the US and UK had refused to sign the Paris declaration. In New Delhi, both appeared on the endorsement list. Yet Michael Kratsios, head of the US delegation and director of the White House Office of Science and Technology Policy, told the summit on February 20 that international talk had moved “from AI safety to AI action” and now to “AI impact,” but warned against fear-driven bureaucracy. “We totally reject global governance of AI,” he said, arguing that adoption must stay local and that countries should pursue sovereign AI built on trusted stacks rather than centralized rulebooks.
Kratsios linked the shift to Vice President JD Vance’s Paris speech and announced US export and standards initiatives under the American AI Exports Program. The tension carried the article: a signature on a declaration about democratising AI, and a speech rejecting the idea of governing AI globally.
From safety to impact
The diplomatic series that began at Bletchley Park in 2023, continued in Seoul in 2024, and fractured in Paris, had changed its label again. Kratsios himself described the arc from safety to action to impact. New Delhi put access, infrastructure, and the Global South at the centre, connecting to debates about compute and the EU AI Act as alternative models of control.
Swiss president Guy Parmelin used the summit to announce that Switzerland would host the next meeting in Geneva on June 21 and 22, 2027.
Why it matters
The Paris article ended by asking whether the world could agree on a single rulebook. New Delhi offered a partial answer: broad signatures, little obligation. More than 80 countries could endorse language about democratising AI while the United States insisted that real governance stay national.
For a technology built across borders, that is both progress and a limit. The summit widened who convenes the conversation and shifted the vocabulary from existential risk toward impact and access. Whether the declarations change how models are built depends on follow-up, not on the text alone. The Geneva meeting in 2027 will be the next test.