To understand the real friction within India’s AI ambitions today, you need only look at two numbers.
- 4,096: This is the number of Nvidia H100 computing chips that the government has allotted to Bengaluru-based Sarvam AI. These chips have become the world’s most sought-after AI computing infrastructure. Costing up to Rs 30 lakh each and manufactured by the same company in Taiwan, they have become indispensable for training frontier AI models.
- 1%: This is the estimated stake that the government is expected to acquire in the company if the subsidy offered by Sarvam is converted into equity on completion of the funding round.
For the first time since the Internet entered our lives, the Indian state is preparing to become more than a regulator of technology. If the proposed conversion goes through, it would become a shareholder in a private AI company. Vivek Raghavan, co-founder of Sarvam AI, declined to comment for this article.
Around the world, AI is increasingly being regarded as a strategic asset. The United States has tightened access, Europe is pouring public money into building its own AI ecosystems, and China has spent years supporting domestic technology champions.
In that backdrop, New Delhi’s support for Sarvam is not just a financial transaction; This reflects a widespread belief that India cannot depend on foreign companies for technology that could shape everything from public services to national security.
Read this also Sarvam AI, and India’s role in the future of high technology
DigiLocker co-founder Amit Ranjan has no objection to the government’s objective. In fact, he doesn’t see anything particularly unusual about the proposed equity structure. He points out that the Government of India is already an indirect shareholder in the startup through entities like SIDBI. So, this kind of stake is a more direct version of the existing idea. Ranjan’s restlessness is somewhere else.
Think about the technology you use. Some people choose email services or video-conferencing platforms for nationalistic reasons. They choose it because it works better than the alternatives. So, Ranjan asks, why would people apply different standards to AI? If Sarvam wants to compete with the world’s best, it should ultimately be judged by the quality of its models, not the passport it holds.
On the other hand, Harish Mehta, one of the founders of NASSCOM and founder of Onward Technologies, is less interested in how Sarvam positions itself than whether governments should try to create national AI champions. “India has already lost the race to build large language models (LLMs),” he says.
Not because of a lack of talent, but because frontier AI has become a race defined by extraordinary concentrations of capital, compute, and research capacity – advantages that already exist on a large scale elsewhere.
Mehta says governments are accustomed to creating durable assets. Roads, airports, power plants and other such facilities become more valuable with use. Frontier technology works differently. LLM is not a ready asset but a living system that must be continuously improved. Its actual value does not reside on the server. It lives in the researchers who refine it, the engineers who optimize it, and the ideas they generate. Those people go. When they do this, a significant portion of the profits go away with them.
“People move from one company to another,” argues Mehta. “So how can you do an LLM and expect people to stick to it?” This is an observation that goes far beyond sarvam. This questions whether AI can ever be treated like traditional infrastructure. Governments can finance research, they can subsidize computing costs, they can even become shareholders, but they cannot take for granted the curiosity, judgment and ambition that border the frontier model. So, the debate on sarvam is not really about sarvam.
Nor is it about whether the government ultimately owns a percentage of a startup. Those details matter because they reveal a larger change already underway. For nearly three decades, governments left software largely to entrepreneurs and the market. AI is changing that relationship.
Around the world, states have come to regard intelligence as a strategic capability worthy of public intervention. India is ready to join that experiment. Should the state actively support a handful of companies that it believes can build leading edge capacity? Or should it focus on creating the conditions in which multiple competing companies emerge? Amit Ranjan and Harish Mehta arrived at different answers. Yet they agree on one thing: India needs companies capable of competing with the world’s best.
Their disagreement is over how to create them. That’s why two numbers matter at the beginning of this story.
The capacity is about 4,096. 1% is about the role of the state. Together, they ask a question that India will continue to grapple with long after Sarvam’s funding round is forgotten: When intelligence becomes a strategic national capability, where should the market end and the state begin?