India's Sovereign AI Gap Exposed as Anthropic Export Controls Cut Frontier Model Access

India’s artificial intelligence strategy depended on a simple bet: leverage the country’s massive IT talent pool to build applications on top of foreign foundational models. That bet broke on June 12 when Anthropic disabled access to its Fable 5 and Mythos 5 models for foreign nationals, complying with a U.S. government export control directive.

The fallout is now forcing a reckoning across India’s tech sector, according to CNBC’s Inside India newsletter, which spoke with founders, investors, and industry experts about what the sudden cutoff means for the country’s AI ambitions.

”Diversification Buys Time, Not Independence”

Saket Dandotia, co-founder and CEO of enterprise AI application maker Onetab.ai, told CNBC that his business would have broken had he not already diversified across multiple models. “The fact that frontier access can vanish overnight on a foreign government’s order is the whole problem,” Dandotia said. His assessment of model diversification as a mitigation: “Diversification buys time; it doesn’t buy independence.”

That framing captures the core vulnerability. An ADP Research report cited by CNBC found that 41% of Indian workers use AI nearly every day, higher than 26% in China and 19% in the U.S. For a country with that level of daily AI adoption, and one of the world’s largest IT workforces actively building agent-enabled applications and workflows, sudden loss of frontier model access creates immediate operational risk.

The Infrastructure Gap

India does not produce cutting-edge chips domestically. It does not have a frontier-scale foundation model comparable to leading U.S. or Chinese systems. Its data center capacity, while growing, lags considerably behind the U.S. and China.

Government programs exist on all three fronts: an $18 billion semiconductor mission, an AI mission, and tax incentives for global hyperscalers to build data centers in the country.

The private sector is responding too. On Monday, sovereign AI company Sarvam AI announced a $300 million Series B at a $1.5 billion valuation, with HCL Technologies and Bessemer Venture Partners joining Khosla Ventures and Peak XV Partners. HCL Tech’s investment of 14.27 billion rupees ($151 million) was less than 10% of what it paid shareholders as dividends in the fiscal year ending March 2026.

Industry experts told CNBC the efforts are insufficient. Prominent venture capitalist Mohandas Pai urged Prime Minister Modi to expand the AI mission, calling existing programs “too slow, way too small to make any large impact.”

The Scale Problem

Building a competitive foundational model for a country India’s size requires trillions of parameters, according to experts cited by CNBC. Sarvam’s flagship model currently has roughly 100 billion parameters. Closing that gap demands capital and computing power that India’s investment ecosystem has not historically directed toward deep-tech.

Indian startups raised $10.5 billion in 2025, third globally after the U.S. and U.K., according to a Tracxn report. But most of that capital went to enterprise applications, retail, and fintech, not deep-tech companies working on foundational AI infrastructure. Manish Agarwal, co-founder of physical AI data company Humyn Labs, told CNBC that India’s strength is its domestic market and talent pool, but it “lacks the capital that is available to sovereign AI companies in the U.S. and China.”

Downstream Risk for Agent Builders

The sovereign AI gap has direct consequences for anyone building agent infrastructure on Indian platforms or for Indian enterprises. The country’s IT services sector employs millions of developers who increasingly build agent-enabled workflows, automated business processes, and agentic applications for global clients. When frontier model access disappears, those applications degrade or break entirely.

The risk compounds further if the U.S. extends chip restrictions to India. Indian sovereign AI models currently run on Nvidia architecture. A Blackwell chip restriction, similar to limits already placed on China, would constrain India’s ability to train competitive domestic models, leaving the country dependent on foreign infrastructure at every layer of the AI stack.

For now, India’s AI strategy stands exposed: high adoption, deep talent, growing demand for agent-enabled workflows, and a foundational layer it does not control.