IndiaAI, Where are Your Research Papers?

For a minute, let’s forget that India has a GPU and funding problem.

While the country entered the compute race late and there is still a shortage of high-end chips, India has at least begun solving the bottleneck. The government has committed ₹10,372 crore towards building India’s AI ecosystem, and more than 62,000 GPUs have been empanelled.

But where are the research papers backing that progress? Remarkably, little work from Indian AI-native startups, especially those empanelled under the IndiaAI Mission to develop sovereign foundational models, has entered the scientific record.

For example, Sarvam, during its latest Epoch flagship event, announced major updates and performance milestones of its vision and audio models, as well as open-sourced its LLMs. But beyond system overviews, the company has not yet published a research paper to lay out its direction and technical capability.

The same is true of BharatGen, Tech Mahindra, Shodh AI, Genloop, Avataar, or any other IndiaAI firm. None of the major IndiaAI-backed efforts has yet produced the kind of landmark research publication that shapes the global AI conversation.

India Needs DeepMind

Historically, companies that pioneer scientific leadership begin by first contributing to knowledge. The Transformer architecture became the foundation of modern AI only after Google researchers published Attention Is All You Need paper.

India, however, seems to be taking a different path. It is trying to build the products before contributing to science. That concern is no longer coming only from academics.

“There has been no Indian contribution” to the Transformer architecture that powers today’s LLMs, Nikhil Malhotra, Chief Innovation Officer and Global Head of AI and Emerging Technologies at Tech Mahindra, tells AIM. “India's ambition should be building more research-based startups rather than just startups.” 

The admission is notable since Tech Mahindra is part of the IndiaAI Mission.

Malhotra believes the country is chasing the wrong milestone with the GPUs. “The end goal is to actually produce something which only India has the ability to produce,” he adds.

But what exactly has India produced that has fundamentally changed the direction of AI? The uncomfortable answer is very little.

The papers that have defined modern AI continue to come largely from laboratories in the United States and, increasingly, China. 

“I am as guilty as anyone else for not putting papers on arXiv,” Malhotra admits. “Corporates also need to allow people to put in more and more of these research papers because we look at patentability faster. We don't put our research first.”

Malhotra believes India does not lack talent. It lacks a culture that rewards research.

He argues that the country needs organisations built around long-term scientific work, similar to Google DeepMind, where researchers spend years pursuing difficult problems rather than chasing quarterly product launches.

Somewhere along the way, Indian technology companies became convinced that publishing research weakens competitive advantage. Researchers slowly disappeared from boardroom conversations, replaced by product managers and sales teams.

Amit Sheth, Founding Director and CEO of the Indian AI Research Organisation (IAIRO), believes this is not unique to AI. It is the defining weakness of India’s innovation ecosystem.

“They do not hire researchers, typically those who would keep up with the latest scientific advances, requiring keeping up with the literature,” he remarks.

Instead, companies have developed a protectionist instinct. “They believe they need to protect every IP and do not want their competitors to know what they are working on,” Sheth observes.

The result is that companies stop participating in science altogether. “They do not want to contribute to the scientific community. They think they can take care of all the research they need internally by hiring a few people.”

The Mindset Problem

Sheth points to the pharmaceutical industry as an example.

"We have a very poor innovation ecosystem," he laments. "It is unbelievable that Indian pharma cannot produce a single new drug in a year. Management is happy doing low-cost drugs rather than growing new pipelines."

The AI industry, Sheth warns, risks repeating exactly the same mistake.

CoRover founder Ankush Sabharwal believes India is measuring success through the wrong metrics. "The measurement should not be in how much megawatt data centre we are creating, how many tokens we are spending or how many GPUs we are using.It should be what value we are adding," he asserts.

According to Sabharwal, India already knows what it needs to build. The problem is that the country keeps getting distracted by global narratives instead of trusting its own instincts. "Sometimes people do things because it's vanity. It is important to gain a position so that others know we are also doing it."

Sabharwal's criticism goes beyond the current debate around sovereign AI. He argues that India has repeatedly allowed global technology trends to dictate its own priorities instead of building on its historical strengths.

"We have to reclaim the glory and the research appetite we had," he said, pointing to India's long tradition of scientific discovery. 

India's gross expenditure on R&D has hovered at around 0.6–0.7% of GDP, significantly lower than countries such as the US, China, South Korea, and Israel, all of which invest several times more as a share of their economies. The government aims to increase that share to 2% of GDP, but only by 2035. 

Beyond that, there’s a dearth of research converting into real business outcomes. According to a recent research published in the Journal of Science and Technology Policy Management, only 2.5% of all Department for Promotion of Industry and Internal Trade-registered active startups are backed by patent-based inventions.

Research Needs Patience

BharatGen’s Ganesh Ramakrishnan echoes that philosophy from a research perspective. While leading one of India’s most ambitious foundation model efforts, he argues that the country's long-term advantage will come not from endlessly scaling compute but from making AI more data-efficient, collaborative, and affordable. 

BharatGen may actually be the closest example India has of a research-first AI effort. Its consortium structure, multilingual datasets, and emphasis on data efficiency reflect the kind of scientific collaboration that has historically produced breakthroughs elsewhere. But even here, much of that work is still waiting to appear in peer-reviewed publications or widely cited research.

But research is expensive, and it often takes years to produce anything commercially useful. Most importantly, it demands patience—something India's technology industry has historically struggled with.

Abhishek Upperwal, Founder and CEO of Soket AI, which is also empanelled under the IndiaAI Mission, believes that is where the real problem lies. 

"If you're expecting hardcore R&D to come out of India, it can't come in a short time period. That has been the misstep," he observes. Upperwal has little patience for the argument that India should remain focused on applications while someone else builds the frontier models.

"The new effort that needs to be taken up is the slightly difficult one, which is: can India start investing in core research and development? That's a big, big problem."

He is careful to point out that this is not simply about money.

"It's a mindset problem. Indians typically are not accepting of long-term investments, putting money into R&D and then seeing some sort of outcome after five or six years," he comments.

That observation cuts to the heart of India's AI dilemma. 

Not Done Overnight

India is now trying to build laboratories while also simultaneously commercialising them. That inevitably creates pressure to show products before the underlying research has had time to mature. 

Upperwal believes India should stop aiming lower. He pushes back against the increasingly popular idea that Indian companies should settle for building "frontier-minus-one" models, coined by Sarvam and BharatGen. 

"Our vision should always be looking at the stars," he asserts. "Why should we not be building an equivalent of Anthropic? We have to do it."

That confidence, however, comes with a caveat.

"It can't be done in one shot." It has to be iterative, funded, and above all, it has to be treated as research rather than simply another software project.

That is exactly where Sheth believes government policy must evolve. Instead of rewarding companies primarily for manufacturing, deployment, or incremental process improvements, the IARIO founder argues that incentives should be directly linked to genuine scientific research.

Beyond GPUs, funding, and data centres, India’s next AI need is far simpler: a research paper the rest of the world cannot ignore.