Recent headlines about India's AI model, Sarvam, speak volumes about our desperation to join the mile-high AI club.
Many articles highlighted it as the AI that beat Google Gemini and ChatGPT. Exciting, isn't it? Except when you read the article in depth, you realise Sarvam AI was best in the category of Optical Character Recognition—OCR, particularly in the Indian languages. This is like a new car manufacturer claiming they have the world's best windshield, better than Mercedes, Ford, Toyota, and Jaguar. It might be true, but the best windshield does not make the entire car better than every other.
I am not belittling Sarvam's achievement. Their capabilities will go a long way in driving AI adoption in regional languages. But anyone hyping it to claim they are better than the world's best LLMs is doing a disservice to Indian AI developers who are indeed working to create systems that could truly be the best someday.
But this also becomes a metaphor for our AI ambition: we are so hungry for success in the field that we hype even our smallest achievements. As the global India AI Impact Summit gets underway in Delhi, I suspect we will do more of it. But stay with me as I show you a better way to truly compete with the best in the world.
THE AI IMPACT SUMMIT: The AI Impact Summit is being projected as the first global AI summit in the Global South and a major stage for India to talk about an inclusive "AI for public good" narrative. The optics will be world-class with global leaders at Bharat Mandapam, glossy expos, hackathons, declarations about responsible AI, and even an AI film festival (by the same folks I wrote about in a previous column).
However, if this remains largely a fair and fanfare, India will walk away with empty speeches instead of strong capabilities. Because truly catching up to developments in AI is not about a two-day event, but needs a national vision to last a decade and ruthless execution on compute, capital, research, and state capacity.
WHERE DOES INDIA ACTUALLY STAND IN THE AI RACE: The correct answer to this question is between a rock and a hard place. We are highly visible in rhetoric, ranking and talent, but underweight in what really matters: hardware, deep research and globally competitive products.
On paper, we have climbed to third place in Stanford University’s 2025 Global AI Vibrancy rankings—behind only the US and China—because of improvements across research, talent, and policy. We also recorded the world's highest year-on-year growth in AI hiring in 2024 at 33.4 per cent, and have become one of the top contributors to AI-related GitHub projects. This is big and shows real momentum in the AI field.
Yet, if you look at the money and infrastructure, the perspective changes. Between 2013 and 2024, India attracted about 11.1 billion dollars in private AI investment, ranking seventh globally, while the US pulled in around 470.9 billion and China 119.3 billion. In 2023 alone, India ranked tenth with about 1.4 billion dollars of private AI funding: far behind the top two. A World Economic Forum-linked analysis suggests that India's accumulated AI investments since 2010 amount to just 1.2–1.8 per cent of its 2024 GDP, compared to 3.4–5.1 per cent for the US and 3.1–4.6 per cent for Singapore.
To sum up: India has joined the AI chatroom, but the gap in what we can contribute compared to China, the US, and even Europe is huge. Summits could be a start, but they won't be the only thing that will fix it.
THE DEFICIT IN COMPUTE, CAPITAL, AND DEEP RESEARCH: Building brilliant LLMs is not about hiring geniuses. It is about GPUs, data centres and electricity. A Deloitte-linked report notes that although India generates about 20 per cent of global data, it holds only around 3 per cent of the world's data centre capacity. To meet AI-driven demand by 2030, India will need an additional 45–50 million square feet of data centre real estate and 40–45 terawatt hours of incremental power: numbers that imply massive infrastructure build-out in less than a decade.
The government indeed understands this, approving, under the IndiaAI Mission, an outlay of about ₹10,371 crore in March 2024 to build a publicly funded AI compute backbone using over 34,000 high-performance GPUs, with a target of about 50,000 GPUs through successive tenders. This is one of the world's largest state-backed AI programmes ever, with subsidised access for startups and researchers at rates as low as ₹65 per GPU hour.
But even this may not be enough.
The Economic Survey 2026 warns that India's AI ambitions may be held back less by demand or funding, and more by limited access to GPUs as global giants corner the supply. A broader survey of India's GPU landscape highlights additional constraints: power and cooling bottlenecks, data centres not designed for AI-grade loads, and grid limitations, especially in Tier 2 cities targeted for expansion. At the same time, independent analyses indicate that our initial national targets, such as 10,000 GPUs, were overly conservative given India's strategic goals.
Even when it comes to talent, we do produce them, but we cannot retain them. The proof is that the best AI companies in the US are helmed by Indians, and so are their grassroots developers. The Stanford AI Index 2025 notes that India leads the world in AI hiring growth, but lags in private investment, research quality, and intellectual property strength. Another review of the same data highlights that while India's AI talent supply is strong, it struggles with talent retention and IP creation compared to advanced economies.
The deficits, then, are clear: insufficient high-end compute relative to ambition, underinvestment as a share of GDP, and a thin layer of globally leading research labs and products.
INDIA'S STRATEGIC ADVANTAGES—TALENT AND DIGITAL PUBLIC INFRASTRUCTURE: The good news is that India is not starting from zero. We have had three structural advantages that very few countries possess.
First of course is talent. The Stanford AI Index 2024 finds that India ranks first globally in AI skill penetration and has seen AI talent concentration grow by over 260 per cent since 2016. Subsequent analyses also find that we lead in AI skill penetration and talent concentration. This means we have the people; the challenge is to give them the tools and incentives to build frontier systems here, not elsewhere.
Second is digital public infrastructure (DPI). Over the last decade, India has built one of the world's most expansive DPI stacks: Aadhaar for identity, UPI for payments, DigiLocker and CoWIN for documents and health, ABHA for health IDs, and ONDC for commerce. This "India Stack" has shown how open, interoperable, population-scale data can enable both public service delivery and private innovation. Add a layer of AI atop this DPI, and you have an intelligent, personalised, responsive and multilingual system that can truly transform everything in India.
From a political perspective, we have a vision. Our early national AI strategy, led by NITI Aayog, proposed an AI for All formulation, positing India as an "AI Garage" for 40 per cent of the world, especially the Global South, for whom we can make inclusive, affordable AI solutions. The idea is there, all it needs is implementation... ruthless execution.
WHAT SUMMITS CAN, AND CANNOT DO: One of the stated goals of the India AI Impact Summit 2026 is responsible AI governance, global cooperation, and impact-driven outcomes that are anchored in the principles of People, Planet, and Progress. This is valuable, especially for shaping global AI norms from a Global South perspective. But we must remember that summits are multipliers, not substitutes. Without hard reforms, we could end up becoming the Town Square for AI diplomacy rather than a powerhouse of AI capability, which we must be to meet our own strategic needs.
So what would it mean to use the summit as a catalyst for work rather than one for cacophony?
The first is to look at and fund our GPU push as we do our highways, ports, and railways, which, interestingly, is also how the USA is looking at its AI infrastructure. The current plan of 34,000–50,000 GPUs deployed through empanelled cloud providers is a strong start, but to compete with the US and China, we need to scale AI-ready data centre capacity with clear, long-term policies about land use, power generation and renewables so that private hyperscalers and Indian firms can plan long term, instead of reacting to immediate changes in policies. Startup labs need to receive heavily subsidised GPUs and other facilities, with a special focus on building regional AI hubs outside local metros, so that clusters of innovation can crop up in universities and democratise access.
The next is to fund deep, open research and sovereign models. Krutrim and Sarvam AI are examples of our ambitions and achievements, especially in our own languages. But we need more of them, a lot more. And not just those who are taking APIs of other foundational models and building an application atop it, but we need a national open model program for the government to fund foundational models for everything, including language, vision and speech. This can only happen with a strong industry-academia partnership with joint labs, shared IP, co-funding and focus on multimodal models and safety. If we want to leapfrog rather than "adapt", we must invest in creating new knowledge, not just deploy imported models on Indian data.
For that, we need centres of research excellence and application-focused AI centres with at least five years, and ideally 10 to 15-year funding horizons, rather than three-year project grants.
But none of it will mean anything until we retain our talent via elite AI fellowships and competitive compensation, offer tax breaks, and offer a fast-track visa for foreign AI researchers and entrepreneurs. We need to reskill mid-career software engineers to move them to AI product development and research.
What the government must do, hence, is set up standards and goals that the industry can aim for. Take AI in healthcare in India. Via research from think tanks and white papers, the government could suggest and nudge industry to focus on drug discovery, triage and diagnostics. In education, personal tutoring in local languages could be what the GoI could push for. In agriculture, prediction and advisory could be the drive.
Of course, regulatory clarity and guardrails must be pushed under an AI framework, especially for high-risk uses, but without it being a roadblock to AI research and development.
A serious Indian AI strategy for catching up with the world must involve hard metrics, from GPUs deployed on the ground and used by Indian teams to AI papers cited globally, models originating in India, AI investment as a share of GDP, and the languages in which the models are available, etc.
But this can't happen just by using AI as a spectacle; it must become part of our statecraft and industrial policy. We have the talent and, in the recent past, even the governmental will. What we lack is a sustained, boring, infrastructure-heavy grind that turns our positive aspects into working products. And if the AI Impact Summit 2026 becomes the moment when India commits to that grind, resolves to compete on compute, research, regulation, and state capacity, this 2026 AI mela won’t be a jhamela, but a global khela which India competes in, and perhaps even wins.
