When people talk about technology infrastructure, they usually mean semiconductors, data centres, and AI models. But the layer that decides how well these technologies work is the data underneath them, and India still has to build it. At a panel discussion on 'The invisible infrastructure of the new tech order' at TechSparks 2026, moderated by Chhavi Mahajan, Chief Content Officer, YourStory, the panellists said India’s advantage will come from owning trusted, context-rich data in fields ranging from language AI to biotech. Arvind Pani, Founder, CoreMantle, an AI data intelligence startup, said data is the one part of the technology stack that cannot be swapped out. “You can continue to replace compute. You can continue to replace the application, but you cannot replace the data,” he said. Ankit Bagaria, Co-founder and CEO, Loopworm, an insect biotechnology company, said the gap is especially wide in biopharma, where too much work still happens through trial and error. Better data, he said, could cut the time it takes to bring a product to market. Bagaria said biotech deserves the same attention as AI, quantum and chips, particularly after the COVID-19 pandemic. He noted that India still imports about 73% of its key materials, including active pharmaceutical ingredients and reagents, from China. And although the country supplies a large share of the world’s vaccines by volume, that accounts for only about 5% of the market by value. “We have operated in the biosimilar space for a very, very long time, where somebody else is discovering the molecules for us,” he said. He argued that India now needs to move into drug discovery. Pani said the same need for local context applies to language. India’s internet was built for English users, which put up two barriers at once for most citizens, who had to learn the technology and a new language. Only about 4 to 5% of India’s population is comfortable using English alone, he noted. He added that even English is not a solved problem for AI. Models trained mostly on American English often struggle with Australian or regional accents, and English as spoken across Indian states varies widely. “These are the nuances that nobody else will understand, that India will understand,” he said. Both panellists agreed that a bigger dataset is not necessarily a better one. Pani said AI models built on smaller amounts of enriched, contextual data outperform those trained on large volumes of raw data. Many Indian language datasets are recorded in studios from scripts, he noted, so models trained on them perform well in controlled settings but fail in real-world conversations. Bagaria said that across biological data, “it’s garbage in, garbage out.” Loopworm spent about 12 to 13 months cleaning existing data while building a digital tool that maps the inner workings of the silkworm. The work had to be tailored to Indian silkworm strains and the pathogens that affect them. Loopworm uses silkworms to produce sustainable animal feed and high-value recombinant proteins. Bagaria said Loopworm chose silkworm because silk is over 98% protein, which makes the insect one of nature’s most efficient converters of plant leaves into protein. The silkworm has been domesticated for about 4,000 years, its genome was sequenced in 2007, and around 175 institutes in India teach sericulture and related biotech. He added that this data-led approach has already cut a set of experiments from 10 months and Rs 25 lakh to two and a half months and about Rs 4.5 lakh. “I am sure that in 10 years’ time, all of us would be saying that the largest impact of AI has come on biotech,” he said. Pani said proprietary data is already becoming a competitive moat. He cited legal AI company Harvey, which has changed both its models and its compute while keeping its domain-specific data as its core advantage. On what needs to change, Pani said India must design technology for its own needs. He recalled that his previous company brought Indian language support to Android in 2011 through a partnership with Qualcomm, before Google did. “If we continue to think that somebody else is going to solve our problem, then we will always be behind the curve,” he said. Bagaria welcomed government initiatives such as the BioE3 policy and the Research, Development and Innovation Fund. But he called for more public and foundational funding for basic data generation, such as the in-depth sequencing and characterisation of Indian organisms, because traditional investors are unlikely to fund fundamental science.