AI Without Dependence: Modi’s Vision for a BRICS-Led Digital Future Beyond Silicon Valley

The artificial-intelligence century has a default geography: models trained in California, chips etched in Taiwan, clouds domiciled in Virginia. For most of humanity, that means depending on decisions made elsewhere — about what models know, what they cost and whose languages they speak. India’s answer, articulated most forcefully in Modi’s digital-sovereignty push and now echoed across BRICS capitals, is not a wall. It is a parallel stack: AI built at home, for the languages and problems the default stack ignores, offered to the Global South at prices the default stack will never quote.
The sovereignty stack, layer by layer
The foundation is compute. The IndiaAI Mission put GPUs into a national commons — tens of thousands of subsidised accelerators available to startups and researchers at prices no commercial cloud matches — because a country cannot speak about sovereignty while renting its thinking capacity from someone else’s data centre. The second layer is models. Home-grown foundation models, including the Sarvam family, are being trained with Indian languages as first-class citizens rather than translation afterthoughts; the government’s AIKosh dataset platform feeds them the linguistic and cultural corpus that Silicon Valley never bothered to scrape properly. The third layer is applications, where India’s advantage compounds: AI layered on digital public infrastructure — UPI rails, Aadhaar identity, ONDC commerce — produces services that reach a flower seller on a bicycle, not just a laptop owner in a capital city.
Digital sovereignty is not refusing the future. It is refusing to rent the future exclusively from someone else.
The Crescent World
Why a BRICS-led AI bloc is more than rhetoric
The BRICS AI agenda has three pragmatic planks. Compute diversification: members pool accelerator capacity and negotiate chip access as a bloc, softening dependence on a single export regime. Data and language: joint efforts on multilingual models — Hindi, Portuguese, Russian, Arabic, Swahili, Bahasa — create a common research frontier that Western labs have no commercial reason to prioritise. And governance: a Global South voice on AI safety and standards, so the rules of the next technology are not written exclusively by the countries that own the current one. India’s G20 presidency already moved this vocabulary into the mainstream; the BRICS channel gives it standing infrastructure.
- Bhashini and the language frontier. The national translation stack covers India’s 22 scheduled languages and is being extended to partner-country languages — AI that speaks for the majority of humanity, not the median internet user.
- Agri and health AI first. Crop-disease detection by photo, satellite yield forecasts, tele-dermatology for district hospitals — applications aimed at the Global South’s actual productivity frontier.
- Open weights, open prices. India’s policy bias favors open and low-cost models that universities and startups across the Global South can actually deploy and audit.
- Semiconductor ambition. Fab projects in Sanand, Dholera and Jagiroad aim at the mature-node heart of the market — not the bleeding edge, but the chips that run everything real.
What “without dependence” does not mean
Honesty requires precision. The vision is not autarky — India still imports advanced chips, still publishes in shared research communities, and still benefits from open-source models released abroad. The point is optionality: a Global South that has its own compute, its own models and its own governance seat negotiates better prices, audits its own risks and trains its own languages into the future. Dependence is a spectrum, and the Indian strategy moves the centre of gravity without pretending the spectrum disappears.
The honest frictions
- The compute ceiling. Export controls and foundry concentration mean India’s GPU ambitions depend on hardware decisions made in Washington and Taipei; sovereignty stops at the fab door for now.
- Talent exports itself. India trains elite AI researchers who mostly work abroad; domestic labs must outbid global salaries with mission, equity and compute.
- Regulation is unfinished. The AI governance law is still taking shape; getting it wrong either freezes startups or repeats the surveillance mistakes the sovereignty rhetoric criticises.
- BRICS coordination is hard. Members compete on the same chips and distrust each other’s data regimes; a bloc that cannot agree on banking Settlement will find AI standards harder still.
Even with the frictions, the strategic arithmetic stands. The twentieth century’s infrastructure was ports and power grids; the twenty-first century’s includes compute and models, and countries that rent all of it will live by someone else’s terms. India is building the option of not doing that — for itself first, and as a working export for every Global South government that wants the same option without building a Silicon Valley to get it. That is the real offer on the table in BRICS corridors: not a rival empire, but a second set of rails.
Your take: should Global South governments build sovereign AI stacks at national cost — or is open-source dependence on Western models the pragmatic middle path? Argue it in the comments.
Read the companion piece: the moment Silicon Valley noticed — India’s AI Moment: Why Silicon Valley Is Courting the Global South’s Newest Power Player.
Illustration generated with AI (Gemini 3.1 Flash Image) for The Crescent World.



