The lights are dimming at Bharat Mandapam. Three days of panels, lukewarm buffet coffee, and the relentless hum of air conditioning have come to an end. India’s big AI showcase is over. But the one thing everyone actually came for—the New Delhi Declaration—is nowhere to be found. It’s been pushed back. Always tomorrow. Just one more meeting. One more edit to a paragraph about "responsible innovation" that likely means nothing to the people actually writing the code.
The delay is the most honest thing about the whole summit.
For seventy-two hours, New Delhi tried to position itself as the sensible middle ground between Silicon Valley’s "move fast and break things" nihilism and Brussels’ "regulate it until it dies" bureaucracy. The pitch is simple: India has the data, the developers, and the scale. It’s the "Global South" logic applied to silicon. But the friction behind the scenes is palpable. You can smell the tension through the press releases.
The sticking point isn't the technology. It never is. It’s the liability.
Western tech giants are terrified of being held legally responsible for the hallucinations of their chatbots. They want a "voluntary" framework—the corporate equivalent of a pinky swear. Meanwhile, the Indian government is eyeing a different path. They’re talking about "Sovereign AI," which is a fancy way of saying they don’t want to be digital sharecroppers for Microsoft and Google. They want their own models, trained on their own languages, running on their own terms.
But building a sovereign brain is expensive. NVIDIA doesn't accept "aspirational leadership" as a form of payment. Those H100 chips cost $30,000 a pop, and you need thousands of them just to get a seat at the table. The Indian government has floated a plan to spend billions on a national compute cluster, but in a country where the power grid still groans during a heatwave, the trade-off is brutal. Do you buy GPUs, or do you fix the pipes?
The showcase featured plenty of local startups. Most of them were just wrappers. They take GPT-4, slap a regional language interface on top, and call it a revolution. It’s a thin layer of paint on a house they don't own. When the New Delhi Declaration finally drops tomorrow, expect it to be a masterclass in linguistic gymnastics. It will try to appease the Americans by promising not to stifle profit, while nodding to the Europeans about safety, all while insisting that India won't be bullied.
It won’t work. You can’t have "safe AI" that is also "completely unregulated" and "sovereign" all at the same time. Physics and economics don’t allow for that kind of magic.
The delegates are heading to the airport now, clutching their gift bags and swapping LinkedIn profiles. They’ll tell their bosses the summit was a success. They’ll point to the "spirit of cooperation." But the real story is the silence of that missing document. The delay suggests that when the talk about "AI for all" stopped and the talk about "who pays when this goes wrong" started, the room got very quiet.
India wants to be the world’s back office for intelligence, just like it was for software. But code is predictable. AI is a black box that occasionally lies, cheats, and hallucinates legal precedents. Regulating that isn't a matter of signing a declaration; it’s a matter of deciding who gets sued.
We’ll see the text tomorrow. It’ll be polished. It’ll be professional. It’ll use a lot of words to say very little about the actual cost of doing business in a world where the algorithms are smarter than the laws.
If the most powerful tech players in the world need an extra twenty-four hours to agree on a few pages of non-binding text, what happens when they actually have to build something that doesn't break?