‘We are Not in the Race to Build Data Centres’: Nikhil Malhotra

While many technology companies race to build data centres and boast trillion-parameter models, Tech Mahindra is deliberately choosing a different path. Rather than entering an infrastructure arms race, the company says it will double down on partnerships, responsible deployment, and outcome-driven AI.

At the heart of this strategy is a simple proposition: AI must be “delivered right”.

“We are not in the race for building data centres,” Nikhil Malhotra, Chief Innovation Officer and Global Head of AI at Tech Mahindra, says candidly. “We will continue to work with our partners, whether they are chip manufacturers or hyperscalers, to ensure that we provide the right services.”

That decision is both strategic and philosophical. While global giants invest billions in owning compute capacity, Tech Mahindra sees its role differently. “AI has probably become the invisible infrastructure,” he explains in a conversation with AIM on the sidelines of the India AI Impact Summit. “Most of our deals today have an AI component built in.”

Rather than owning the pipes, the company wants to control how intelligently they are used. “There’s a lot of confusion in the market about bigger GPUs and bigger infrastructure,” he notes. “But sometimes the best solution runs on a PC.”

He highlights Project Indus as an example. Powered by NVIDIA, it is a Hindi-first, education-focused large language model built to offer Indian linguistic and cultural contexts. “The deployment of that model actually goes on a local system. We can take it to parts of the country where there is no internet. That’s impact.”

Where heavier compute is required—such as for multimodal systems—Tech Mahindra will rely on ecosystem partners. “We will continue to partner with infrastructure providers. That is not our battleground.”

Project Indus and the Foundational Debate

With Project Indus, Tech Mahindra entered the LLM arena ahead of many Indian peers. Critics argue India lacks original foundational research capability.

“I don’t need to argue,” he responded. “We put the model on Hugging Face. People can see the code.”

Malhotra says Indus-1 is unique for being trained only on Hindi tokens. “There is still no model in the world that takes only my dialect of Hindi. Everything from tokenisation up was ours.”

While the architecture drew from known frameworks, the build was indigenous. “We can now build LLMs and SLMs from the ground up.”

Yet he acknowledges limitations. “I would have liked us to come out with a completely new algorithm. That is the next frontier.”

From Cost Optimisation to Revenue Growth

For India’s IT majors, AI initially promised productivity gains and automation-led savings. But where will the real revenue come from? The answer, Malhotra suggests, lies in embedding AI deeply and responsibly into enterprise transformation.

The first pillar is productivity, but with pragmatism. “There is a lot of talk about 60–70% productivity gains. We are very cognisant that in brownfield systems, there is a limit. We benchmark everything.”

The second pillar is transformation. “It could be stack transformation, technology transformation, or process transformation. We help customers reimagine the system.”

The third is expectation management. “When the rubber hits the road, what is expected rarely happens. We guide clients on what AI can and cannot do.”

The fourth is assurance. “We want to make AI responsible. We have our own system, TechMVerify, which looks at bias, ethics, and toxicity. Responsible AI is not optional.”

On how much of Tech Mahindra’s AI investment goes into foundational research versus enterprise applications, Malhotra, while not disclosing the precise split, emphasises applied innovation anchored in internal research hubs such as Maker’s Lab.

The Workforce Question

Morale, meanwhile, is a recurring concern across the IT industry as automation reshapes roles.

“There are going to be changes in jobs,” he admits, while also rejecting the narrative of mass obsolescence. “If you learn physics and mathematics correctly, if you understand the basics of machine learning, you will always have a job in the IT industry.”

In fact, he argues AI is creating new roles. “When we apply AI in brownfield systems, unless there is a human holding its hand, it doesn’t work well. Enterprises are complex. We cannot leave everything to AI.”

Tech Mahindra is investing in upskilling and rotational exposure. “People move from delivery to Maker’s Lab, work on cutting-edge systems, then go back. They bring that value with them.”

The future workforce, he suggests, will be hybrid. “You may become someone who understands prompts. Or someone who interprets AI outputs and applies them in enterprise integration. The roles will shift, not disappear.”

Cutting Through Buzzwords

In a market obsessed with agentic AI and RAG frameworks, Tech Mahindra’s approach is consultative. “First, what problem are we solving? Second, what is your current estate? Third, what financial return will this generate? Fourth, do we agree on the ROI?” Malhotra explains.

In some cases, the company has to double down on pragmatism. “Customers ask for 90% improvement. We tell them, sorry, this is not possible.”

The Trillion-Parameter Question

Last year, in an interview with Moneycontrol, Tech Mahindra CEO and MD Mohit Joshi said that the company was building a trillion-parameter sovereign model. Malhotra now calls that “a misnomer”.

“We have always said 8–12 billion parameters,” he clarified. “One trillion is compute-heavy and energy-intensive.”

At an industry showcase, Tech Mahindra demonstrated an 8-billion-parameter educational model focused on physics from grades six to 12, trained in Indian languages, including Bhojpuri and Hindi. Built in partnership with hardware players, including Intel, the system runs locally. “We did this for the sake of the planet and for reach.”

Commercial viability remains a work in progress. “We will see how commercially viable it is post-April. At this point, we are focused on utility.”

A Pragmatic Future

Tech Mahindra is now repositioning around vertical stacks. “AI is invisible infrastructure. We look at infrastructure, models, data ontology, applications—a vertical slice,” Malhotra says.

Outcome-based pricing and service token models are also under exploration. Yet the transition will not be overnight. “There are still large integration systems that need hands. We cannot automate everything immediately.”

For the company, the AI journey is less about scale for its own sake and more about fit-for-purpose intelligence.

As enterprises sift through AI promises and policymakers grapple with ethics and sovereignty, the company’s message is restrained but firm: responsibility, practicality and partnership will outlast hype cycles.

“AI is not a tick mark,” he concludes. “It is a tool. First, understand the problem. Then decide whether AI is the right answer.”