Donald Trump has changed the vocabulary of America’s AI debate. His administration now speaks increasingly of “Super Intelligence”, and on September 29 the phrase acquired a remarkable cast of characters at the White House. Elon Musk, Mark Zuckerberg, Sundar Pichai, Jensen Huang, Dario Amodei, Satya Nadella, Jeff Bezos and Open AI president Greg Brockman were among a gathering representing much of the machinery on which the AI revolution presently depends: models, chips, cloud computing, platforms, data centres and enormous pools of capital. The immediate outcome was a voluntary accord on frontier AI safety. Companies agreed to stronger internal controls, independent external audits and board level scrutiny.
Trump called the commitments “morally binding” and spoke approvingly of “tremendous self policing”. Yet the President made equally clear that he would “never stifle” a technology he believes could exceed the Industrial Revolution in importance. In those two apparently contradictory impulses lies the real Washington story: America wants safeguards around AI, but it has no intention of allowing those safeguards to cost it the technological race. The meeting itself had surfaced publicly only days earlier. Reuters reported on September 24 that Trump, US House Speaker Mike Johnson and technology chiefs would meet five days later, with further details then unavailable. By the eve of the gathering, Johnson was publicly defining the objective: Washington did not need a moratorium or “hyper-regulation” that could cause America to “lose the race to China”; it needed, he said, to find the right balance between innovation and oversight.
Then came a piece of political theatre that captured the argument better than another policy document might have done. What began as a closed-door luncheon spilled unexpectedly into public view when Trump emerged onto the White House driveway, an area rarely used by the President for such encounters, with Johnson and some of the world’s most powerful technology executives lined up shoulder to shoulder behind him. Trump took questions from journalists for roughly half an hour as the CEOs stood there. Washington has been gripped by renewed concern over the risks of increasingly powerful AI, but the tableau outside the White House conveyed the administration’s answer rather neatly: the builders would be asked to police themselves, government would look for guardrails, but America would keep moving.
FROM ARTIFICIAL INTELLIGENCE TO ‘SUPER INTELLIGENCE’
The terminology may sound characteristically Trumpian, but the technological change underneath it is serious. AI is moving from systems that principally respond to human prompts towards agents capable of carrying out sequences of actions with progressively less human supervision. The debate is consequently shifting from what an AI system can say to what an AI system should be permitted to do.
That distinction helps explain the growing divide not merely in Washington but within the technology industry itself. Some frontier laboratories have warned that increasingly autonomous systems require stronger safeguards, while others argue that a broad slowdown could surrender technological ground to competitors. The disagreement is no longer between people who believe in AI and those who do not. It is increasingly between those who fear moving too fast and those who believe that, in a strategic competition with China, slowing down carries risks of its own.
The political fraternity in Washington is confronting much the same dilemma. Lawmakers are debating how much oversight is necessary without choking innovation, while technology journalists and the foreign press are increasingly examining not merely the extraordinary capabilities of the models but who should be accountable when autonomous systems behave in ways their creators did not anticipate. On Capitol Hill itself, senators attempted on September 29 to advance legislation that would have created an AI safety board and enforceable standards for evaluating frontier models; the effort was blocked in the Senate.
For India, however, the argument reveals something even more immediate. The race has already moved beyond the model.
THE REAL AI RACE IS UNDER THE BONNET
Until recently, the easiest way to follow AI was to compare models. Who reasoned better? Which chatbot generated better code? Who was ahead in multimodality or agents? That scoreboard is becoming inadequate because the glamorous part of AI is the model; the expensive part is everything underneath it.
Frontier systems require GPUs, semiconductor fabrication, enormous data centres, electricity, cooling, networks, datasets, researchers and capital on a scale that would have appeared extraordinary even by Big Tech standards a few years ago. Nvidia’s Jensen Huang therefore matters to this story as much as the people building the models themselves.
For the average American, this increasingly has a tangible dimension. A data centre is no longer an abstraction when it arrives in a community consuming vast quantities of electricity and residents begin asking what it means for their utility bills, water resources and local infrastructure. The opposition has become sufficiently visible and bipartisan for Trump to address it directly on Tuesday, arguing that technology companies should provide greater benefits to the communities hosting these facilities. The International Energy Agency estimates that electricity consumed by data centres worldwide could rise to roughly 945 terawatt-hours by 2030, more than twice the 2024 level. The irony of the AI revolution is that something experienced through an almost weightless screen is developing the physical appetite of heavy industry.
That should immediately alter the Indian conversation. India has spent years speaking, legitimately, about its software talent, startup ecosystem and enormous digital market. But the AI era will punish countries that confuse the ability to use sophisticated software with control over enough of the infrastructure required to sustain it.
45,000 GPUs ARE A BEGINNING
India has made a serious start. The IndiaAI Mission, approved with an outlay of approximately Rs 10,372 crore, had expanded its shared compute capacity to more than 45,000 GPUs by June 2026. By August, 237 projects had received subsidised computing support covering 93.18 lakh GPU hours. From 506 applications, the government has also selected 20 indigenous foundation-model proposals, including 12 large multimodal models and eight small language models.
These numbers matter because compute is expensive. An Indian startup with an excellent idea cannot be expected to compete with a hyperscaler’s balance sheet simply to test a model. Making computing capacity available to universities, researchers and young companies therefore removes one of the earliest barriers to innovation.
But access to computing power and control over the technology producing that computing power are not the same thing. The Modi government itself acknowledged in Parliament in August the risks created by limited domestic capabilities in semiconductor manufacturing, computing infrastructure, foundation models and advanced research. That recognition is important because dependence is not inherently a weakness; dependence that cannot be substituted when geopolitical circumstances change can become one.
THE SILICON TEST
This is why India’s semiconductor programme and its AI programme can no longer be treated as two separate policy stories.
In July, the Modi government approved Semicon 2.0 with an outlay of Rs 127,500 crore. The programme extends the effort beyond fabrication into Indiandesigned chips, advanced packaging, materials, gases, manufacturing equipment, research and skills. The latest government figures say five semiconductor units have commenced commercial production, while India employs nearly 20% of the global semiconductor chipdesign workforce.
Those numbers demonstrate both India’s advantage and the distance still to travel. Indian engineers already occupy an important place in the global semiconductor industry, but India does not yet manufacture the most advanced chips at anything approaching the scale the AI age may demand.
That distinction once belonged principally to industrial policy. AI has made it strategic. The US has already demonstrated that access to advanced semiconductors can become an instrument of foreign policy. If tomorrow’s most capable models require ever larger quantities of sophisticated processors, a country cannot compensate indefinitely for constrained hardware simply by possessing excellent software engineers.
India does not need semiconductor autarky; no serious technological power has it. It does need to decide, with considerable precision, which dependencies are commercially sensible and which could eventually constrain national choices.
THEN THERE IS THE ELECTRICITY BILL
There is another constraint that receives far less glamour than models or chips. India is simultaneously industrialising, electrifying transport, expanding manufacturing, cooling a warming population and building digital infrastructure. AI now arrives as another potentially enormous claimant on the grid.
That does not mean India cannot generate sufficient electricity. It means AI planning can no longer be separated from power planning. Where should major computing clusters be located? Which states possess adequate transmission capacity? What combination of renewable energy, storage, conventional generation and, over the longer term, nuclear power should support them? And how do we prevent the cost of supplying enormous private computing facilities from becoming a burden on ordinary consumers?
America is encountering those questions after data centre construction has already collided with local politics. India has the advantage of seeing the problem before its own AI build-out reaches comparable scale.
The technology conversation therefore has to extend beyond the IT ministry into power, finance, education, infrastructure and the states.
INDIA’S RACE NEED NOT LOOK LIKE AMERICA’S
There is nevertheless a danger in judging Indian progress entirely against American benchmarks.
India’s most consequential AI opportunity may emerge from problems Silicon Valley has little commercial reason to solve first.
Consider language. A frontier model may perform spectacularly on an English coding benchmark and still be of limited use to an Indian citizen who wants to describe an electricity complaint orally in Kokborok, obtain agricultural advice in Assamese or understand a government entitlement in Bengali.
A useful Indian AI system cannot presume that its user will type an elegant English prompt. It must understand voice, accents, dialects and code-switching. It has to operate cheaply enough for mass deployment, connect securely with public or private databases, recognise uncertainty and know when to transfer a conversation to a human being.
India AI is already supporting indigenous foundation models designed around Indian languages and needs, and AI Kosh had more than 14,000 datasets and 331 AI models available by July.
The opportunity is not to produce a poorer imitation of an American chatbot. It is to make artificial intelligence useful to an Indian population at Indian scale.
THE AMERICAN OPENING, AND THE CHINESE SHADOW
China runs through Washington’s AI debate even when no Chinese official is in the room. Semiconductors, critical minerals, supply chains, models and data centres are increasingly being discussed as elements of strategic competition rather than merely commercial sectors.
That creates an opening for New Delhi. Washington wants trusted technology partners, diversified supply chains, engineering talent and large markets. India wants capital, frontier technology, advanced chips, research partnerships and greater participation in semiconductor manufacturing. The strategic overlap is unusually strong.
But access should never be confused with capability.
The presence of Sundar Pichai and Satya Nadella among the technology leaders in Washington inevitably resonates in India. Their ascent is an extraordinary testament to Indian technological talent. The next ambition should be larger: not only Indians occupying the commanding heights of corporations created elsewhere, but more globally consequential technologies and enterprises created in India itself.
For decades India has supplied some of the people who built the global technology economy. The AI age will test whether it can own more of what they build.
CAN THE BUILDERS POLICE THEMSELVES?
That returns the debate to the question Trump sought to address on September 29: what happens when safety collides with competition?
The American public is plainly uneasy. The administration is simultaneously confronting demands for stronger safeguards and pressure not to hobble companies engaged in a global technological contest. Congress itself is divided over how far regulation should go, while the industry possesses both the technical expertise governments need and commercial interests governments cannot afford to ignore. The voluntary compact therefore represents less a settlement of the argument than Washington’s present attempt to live with it.
India will eventually confront its own version of that dilemma. Its answer cannot simply be imported from Washington. AI will increasingly operate inside Indian public services and across a population whose ability to challenge an automated decision varies enormously.
Privacy, accountability, accuracy and human escalation will matter greatly, but regulation written principally out of fear could impose another cost. A country still building technological capability cannot regulate itself out of the race before it has properly entered it.
THE MESSAGE FROM WASHINGTON
That is what makes September 29 more than another White House technology event. Trump calls it “Super Intelligence”. Behind the branding lies something much more concrete: America is attempting to bring chips, compute, data centres, capital, energy, frontier models and government policy into the same strategic conversation.
India has begun work on almost every one of those fronts. What it does not yet possess is the luxury of treating them as separate conversations.
The next Indian AI debate therefore cannot stop at how many GPUs have been acquired, how many startups have been funded or how many Indian-language models can be demonstrated.
The harder test is whether semiconductors, compute, electricity, research, capital, language technologies and public deployment can be made to reinforce one another.
The scene on the White House driveway offered an unusually vivid glimpse of where the American debate has reached: a President determined to accelerate, political Washington searching for acceptable guardrails, technology chiefs promising greater responsibility and China looming over the entire conversation.
India need not reproduce America’s AI strategy, nor should it. But as the age of “Super Intelligence” moves from slogan towards technology, infrastructure and geopolitical competition, New Delhi will have to answer three increasingly consequential questions: what does India intend to build, what is it prepared to depend upon others for, and what can it no longer afford not to control?
*Manoranjana Gupta is a senior journalist, columnist, & media entrepreneur with over three decades in print & broadcast. She is co-author of “Ink, Saffron & Freedom” (2025)
