Who Controls India's AI Boom? Firms, Funds and Data

Indian companies are moving AI from experiments into production, turning control over data, infrastructure and spending into a boardroom decision. Navtez Singh Bal, Red Hat's Vice President and General Manager for India and South Asia, said organisations now have to ask where data is stored, who can access it and how it trains models. “Show me the money. Why are you spending this much? Can you actually show what the output is?” he said, describing the cost questions companies put to AI projects.

Who controls the data and the bill

In an interview published October 28, 2026, Bal said data protection had become mandatory in AI discussions as organisations consider scaling their systems. He cited India's Digital Personal Data Protection Act and requirements covering personal-data safeguards, consented uses, algorithmic accountability, auditing and cross-border transfer restrictions. Data isn't just a technical input here; access and use sit within rules companies have to navigate.

Bal called reports of AI systems bypassing internal guardrails “the tip of the iceberg.” Companies want autonomous agents that can perform tasks, he said, but agents may access unintended data because people don't anticipate every search needed to complete a task. “My sense is we can all get better. We will actually use machines to, by the way, figure out the guardrails as well, so we will use AI for AI,” he said, adding that every boardroom should consider the issue.

Microsoft, IBM and Red Hat are implementing AI in finance, human resources, marketing, software development and testing, Bal said. As models become more reasoning-oriented, they use more tokens and can generate larger bills. At Red Hat, he said, the cost difference between models for basic productivity and those for complex, first-of-its-kind code can be 10 to 100 times. Indian enterprises are adding governance and infrastructure, monitoring costs and trying to prevent “shadow AI.”

Infrastructure, ownership and dependence

Bal said semiconductor issues have caused hardware shortages and rising prices, pushing IT costs up faster than companies' revenue growth. Businesses increasingly discuss sovereignty not only as local data storage but as ownership and control of data and infrastructure, alongside governance and compliance. “I'm not seeing it only in India, but also across the region,” he said, adding that geopolitics and technology had become closely intertwined.

He called Red Hat Enterprise Linux a “bedrock” and the go-to operating system for enterprise deployments, while saying uptake of automation had grown alongside security and AI. Companies want AI embedded in applications, he said, bringing application development and AI management into shared workflows with corporate and enterprise guardrails. The discussion is about who sets those guardrails—and who pays for the infrastructure beneath them.

Bal identified infrastructure spending, change management and user training, along with Indian service providers' need to rethink their business models, as obstacles to wider implementation. Those changes require a shift in leadership mindset, he said. Companies will also need hybrid setups and internal data-centre buildouts as AI moves beyond pilots into production.

Investor money and the early-stage ceiling

A Tracxn report on Indic AI found that 34 companies building AI for Indian languages raised $657 million across 56 funding rounds between 2021 and October 2026. The data was current as of October 15, 2026; the report recorded $687 million in all-time equity funding. Every recorded round was at seed or early stage; no late-stage round had been recorded. Capital rose in 2026 despite fewer deals, largely because of two large Sarvam rounds. Without those rounds, companies raised $54 million across seven rounds that year.

The report divided the companies into those for which language is the product and those for which language is a differentiator. Eighteen—including Sarvam, Gnani, Soket AI and BharatGen—build speech, translation or language models, or data used to train them. Sixteen, including Skit.ai, Nuplay AI and GreyLabs, sell enterprise products where Indic-language handling is a strength. GreyLabs serves more than 50 institutions, including RBL Bank, AU Small Finance Bank and IDFC FIRST Bank.

Sarvam and Krutrim, both founded in 2023, had each reached unicorn status and together held $464 million, the report said. Sarvam had raised $389 million and was one of the government's four selected sovereign model builders in the group. Krutrim had raised $75 million, partly funded expansion with debt collateralised against Bhavish Aggarwal's Ola Electric shares, paused chip design in late 2025 and shifted toward AI cloud infrastructure.

The report described India's ecosystem as built primarily through competing, investor-owned companies, unlike Singapore's state-funded SEA-LION programme, Africa's LINGUA grants, and state-backed projects in Saudi Arabia and the United Arab Emirates. It compared the group's $687 million in equity funding with an estimated $77 billion cost for Saudi Arabia's HUMAIN project, while warning that the figures measure different things. Bal predicted open-source or open-weight models would do “80% of the work,” with frontier labs handling the balance. The companies, investors and state programmes are already contesting who gets to build, fund and govern the systems everyone else will use.