Arguing that India “can’t construct its AI future on borrowed fashions”, Bernstein analysts Venugopal Garre and Nikhil Arela cautioned {that a} technique constructed on renting compute for overseas LLMs and incomes data-centre lease may go away the nation dangerously uncovered. In a report the analysts flagged current US export-control actions, together with restrictions on entry to frontier fashions for non-US residents, as an early sign that AI is shifting from a globally shared know-how to a tightly managed strategic useful resource.
“Foundational fashions will now not be SaaS merchandise,” the report notes, likening cutting-edge AI to “fighter jets” in a brand new period the place entry to the perfect fashions is guardrailed and rationed. In Bernstein’s base case, AI settles into “a world of stratified entry”, with superior nations reserving bleeding-edge capabilities for home defence, intelligence and significant infrastructure, and exporting older, downgraded variations to rising markets.
Within the excessive state of affairs, India may “get up and discover that essentially the most highly effective functions in finance or navy are now not working, and the APIs and weights they depend on have vanished in a single day”.
The priority is amplified by India’s deep know-how dependence throughout the stack, from CPUs, GPUs and cellular SoCs to working programs, cloud platforms, enterprise software program and shopper web rails. Throughout core compute, working programs, browsers, public cloud, social media and productiveness suites, India overwhelmingly depends on US and different overseas distributors, with home initiatives restricted to pockets and missing scale.
Bernstein warns that repeating this sample in AI—letting world corporations personal India’s intelligence layer whereas native gamers keep within the utility margins—may entrench dependence “for many years”.
The absence of an Indian “DeepSeek second”, regardless of India’s huge information reservoir, is described as structural. India’s tech ecosystem has been “services-led”, with out large-scale, consumer-facing platforms in search, social or messaging that generate wealthy, organized information for coaching frontier fashions. This has meant no sustained expertise pipeline or educational depth round foundational fashions, whereas IT providers have rewarded fine-tuning overseas software program slightly than constructing core platforms. Bernstein notes that many Indian institutional leaders have argued India doesn’t want its personal LLMs and might give attention to functions, calling these views “extra reflective of the trail India has taken” than a deliberate strategic selection.
Coverage efforts thus far are characterised as “too lumpy, too little” and “defocused, thinly unfold”. The India AI Mission, launched in 2024 with an outlay of round INR 104 billion (about USD 1.2 billion over 5 years), recognized compute, information, analysis, {hardware}, functions and foundational fashions as key pillars.
However allocations have been risky, with a notable discount in AI spending within the revised FY2025 finances, and solely about USD 220 million—lower than 20% of the full—earmarked for foundational fashions. Round 44% of the envelope is directed towards compute, together with imports, which Bernstein suggests dangers reinforcing dependence slightly than constructing home functionality.
The brokerage contrasts India’s broad-but-thin strategy with extra sequenced methods within the US and China. The US leveraged entry to compute and deep non-public capital to push speedy advances in mannequin efficiency, whereas China targeted on mannequin effectivity and semiconductor capabilities underneath export stress.
Over time, each ecosystems expanded throughout the stack from an preliminary base of concentrated funding, whereas India is making an attempt “to do quite a bit from much less sources” and throughout a number of layers concurrently. Bernstein additionally reminds traders that India has traditionally been topic to Western export controls in nuclear, house and defence, arguing AI is more likely to observe the identical sample of sanctions and restricted flows.
The report underlines the strategic threat of India’s AI stack sitting “on the mercy of another person”. If core enterprise, defence, house and monetary programs had been to run on overseas LLMs, a geopolitical disruption may curtail entry “in a single day”, halting crucial functions. Even with out a laborious cut-off, India may very well be locked into N‑1 or N‑2 fashions, working one or two generations behind, with native corporations competing in opposition to world startups bootstrapped by “vibe coders” who’ve direct entry to frontier programs. “India can’t afford to function on another person’s trapdoor,” Bernstein warns, calling deep AI capabilities a “necessity, not a luxurious”.
Towards this backdrop, Bernstein outlines why it needs India to pursue its personal “DeepSeek” and a sovereign AI stack. The core suggestion is to pivot from horizontal dependence on overseas fashions to proudly owning and controlling high-quality, domain-specific datasets in sectors akin to Industrials and healthcare. These vertical datasets, if retained and shielded from unrestricted entry by world platforms, can underpin smaller, specialised LLMs constructed by Indian know-how corporations, on which defensible functions are layered. Bernstein sees the chance extending past software program into the bodily financial system, together with coaching industrial robots and next-generation humanoid programs the place industrial-context information might matter greater than bleeding-edge GenAI.
Coverage levers, in Bernstein’s view, should not easy however two stand out. One is to “prohibit or tempo entry to world AI fashions” whereas channelling capital and expertise into India-native LLM capabilities. The opposite is to mandate or incentivise localisation—requiring overseas corporations to construct and function India-based AI stacks insulated from geopolitics. Each choices are imperfect, however they body the essential trade-off between entry and autonomy. Bernstein suggests the federal government’s function needs to be primarily policy-driven—akin to China’s “nice firewall”—so long as Indian firms are prepared to take part within the LLM race; if nobody needs to construct fashions and everybody focuses on functions, even sturdy public funding will wrestle to vary the end result.