India’s AI Bet: Sovereign Technology or Technological Colony?
In February 2026, the country took a big step towards India AI sovereignty as it hosted the world’s most important AI summit for the first time. Prime Minister Modi inaugurated it. French President Macron flew in. UN Secretary-General Guterres addressed the hall. Sam Altman, Sundar Pichai, all present. $250 billion in investment pledges were announced. A Guinness World Record was set for AI responsibility pledges in 24 hours.
And then someone at Galgotias University presented a Chinese robot dog as an indigenous Indian invention.
That one episode where a university representative showcasing a commercially available Unitree Robotics product from China as a homegrown innovation, getting caught on social media within minutes, and being thrown out of the expo captured something real about where India’s AI ambitions actually stand in 2026. Enormous aspiration. Real gaps. And a world watching closely to see which one wins.
This article is an honest assessment of that question.
What India Actually Built
Start with what is real, because it is genuinely impressive.
Sarvam AI a Bengaluru startup founded in 2023 by Vivek Raghavan and Pratyush Kumar launched two foundational AI models at the summit. Sarvam-30B and Sarvam-105B. Both were trained from scratch in India using government-backed computing resources not fine-tuned versions of foreign models, but genuinely indigenous systems built under severe capital and compute constraints.
The 105B model uses a Mixture-of-Experts architecture, the same technique that made DeepSeek competitive at a fraction of the cost of American models. Sarvam claims it outperforms DeepSeek R1, a 600B Chinese model, on certain cultural and Indic reasoning tasks. The 30B model is designed to run on basic phones without an internet connection, a design choice that makes sense for a country where hundreds of millions of potential users live in low-connectivity settings.
BharatGen also launched at the summit, a 17-billion parameter model supporting 22 Indian languages, developed under the broader PARAM-2 initiative with IIT Bombay, IIT Madras, IIIT Hyderabad, IIT Kanpur and five other institutions. The government announced plans to add 20,000 GPUs to India’s existing base of 38,000. An $11 billion fund for semiconductor manufacturing was confirmed. Reliance Industries Chairman Mukesh Ambani pledged a ₹10 trillion investment in AI infrastructure.
These are not small things. As Raghavan put it, the perception that breakthroughs like this can happen only in Silicon Valley or China is precisely what needs to change.
The Gap Nobody Wants to Talk About
Now for the honest part.
India’s R&D intensity sits at 0.64 percent of GDP. The US spends 3.5 percent. China spends 2.4 percent. That gap does not close with summit pledges. It closes with sustained public investment over decades and India has not historically done that.
Mohandas Pai, founder of Aarin Capital and one of India’s most respected technology investors, was direct at the summit itself: “The key thing to do after this summit is to put money to work. We need ₹30,000-40,000 crore to be invested in AI every year. We need to fund companies like Sarvam.”
That number ₹30,000 crore per year is roughly $3.6 billion annually. China is spending multiples of that. OpenAI’s last fundraise alone was larger. India’s entire IndiaAI Mission is a ₹10,370 crore programme spread over multiple years.
The structural challenges are real. Instead of becoming a core innovation hub for foundational AI, India risks emerging as a consumption and testing market for global models. That risk is not hypothetical. It is what happened with smartphones India has 700 million smartphone users and almost none of the underlying technology is Indian. The app layer is India. The operating system is American. The chips are Taiwanese. The manufacturing is Chinese.
AI sovereignty requires winning at a different layer the foundational model layer and that is both harder and more expensive than anything India has attempted in technology before.
The China Comparison That Should Keep India Awake
The summit was framed, explicitly and repeatedly, as India’s answer to Chinese AI dominance. The framing is right. The gap is larger than the framing suggests.
Chinese AI models operate at roughly one-sixth to one-fourth the cost of comparable US systems, driven by open-source strategies, efficient architectures, and domestic infrastructure. Following DeepSeek R1 in January 2025, Chinese LLM global usage share surged from approximately 3 percent to 13 percent within months, reaching 10-30 percent in various markets by late 2025.
Five major Chinese AI developers timed significant model upgrades around the Lunar New Year in February 2026 Zhipu AI, ByteDance, Alibaba, Baidu and others in what analysts described as a coordinated show of force timed to coincide with India’s summit. The timing was not subtle.
China’s terawatt-scale renewables and sub-dollar AI inference will likely fuel an unprecedented wave of applications throughout the Global South, driving industrialisation on a scale surpassing historical precedents like the Marshall Plan. If that projection is even partially correct, the AI competition is not just about technology. It is about who shapes the digital infrastructure of the next two billion people to come online most of whom live in South Asia, Southeast Asia, and Africa.
India’s model is different from China’s. India’s inclusive, multilingual models and compute sovereignty could establish it as the essential bridge for ethical AI serving billions. That is a real value proposition. The question is whether it translates into actual deployment at scale or remains a conference talking point.
The US Angle And Why It Complicated the Summit
One of the more revealing moments at the India AI Impact Summit was the behaviour of the US delegation.
TechPolicy.Press reported that the United States delegation arrived at the summit with an agenda centred on “domination” rather than cooperation, framing AI as a geopolitical race against China.
That framing AI as a zero-sum competition between the US and China is precisely what India is trying to resist. India’s pitch at the summit was for a third path: frugal, sovereign, multilingual AI that serves the Global South without being controlled by either Washington or Beijing. The American delegation essentially showed up and tried to turn that conversation into a two-player game.
This tension matters for India’s semiconductor ambitions too. Modi discussed semiconductor cooperation with the Netherlands Prime Minister Rob Jetten just this week. ASML the Dutch company that makes the extreme ultraviolet lithography machines without which no advanced chip can be manufactured is central to any serious semiconductor strategy. India getting access to ASML’s technology requires US approval because of export control regimes. That dependency is the quiet constraint beneath every bold announcement about AI sovereignty.
What Sarvam’s Co-Founders Actually Said
Beyond the press releases, Sarvam’s founders said things at the summit that are worth sitting with.
Vivek Raghavan drew a direct comparison to Aadhaar. When India built its digital identity system, all the standards were proprietary and foreign companies owned them. India built something indigenous on open standards and created the India Stack which became the foundation for UPI, DigiLocker and dozens of applications that now serve a billion people. His argument is that Sarvam is doing the same thing for AI — building the foundation layer on which a generation of applications will be built.
Pratyush Kumar was equally direct about the gap with frontier models. Sarvam is not competing with GPT-5 for the highest-end enterprise customer. It is competing for the person who is resource constrained, needs local languages, and operates in low-connectivity environments. That is not a consolation prize. That is a market of hundreds of millions of people that OpenAI and Google have no particular incentive to serve well.
Western AI models charge a 4x “language tax” to process Hindi — meaning it costs four times more to run AI queries in Hindi than in English, because the tokenisation of Hindi is less efficient. Sarvam’s architecture is designed to eliminate that tax. If it succeeds at scale, it changes the economics of AI for every Hindi speaker on the planet.
ThirdPol’s Take
India’s AI sovereignty push is real. It is also insufficient relative to the scale of the challenge.
The honest position is somewhere between the triumphalism of the summit $250 billion pledges, Guinness World Records, Modi testing smart glasses and the cynicism of the critics who saw the Galgotias robot dog episode as the real summary of where India stands.
Sarvam has built something genuinely impressive under genuinely difficult conditions. Sovereignty matters and buffers count and it ultimately rests on building capability at home. The Iran war, which has simultaneously closed the Strait of Hormuz for oil and threatened to cut undersea cables for internet, is a live demonstration of what happens when a country depends on others for critical infrastructure. AI is the next layer of that same vulnerability.
But building sovereignty in AI requires sustained investment at a scale India has never committed to in technology. The ₹10,370 crore IndiaAI Mission is a start. It is not a programme that matches the ambition of the summit that launched it.
The real test is not what gets announced in February at Bharat Mandapam. It is what gets funded in March, deployed in April, and used by real people in villages across UP and Bihar in 2027. Aadhaar passed that test. UPI passed that test. The India Stack is genuinely one of the most impressive technology achievements of any developing country in history.
The question for Indian AI is whether it follows that template patient, indigenous, infrastructure-first, designed for the population that needs it most or whether it follows the more common template of ambitious summits followed by slow implementation and continued dependence on foreign systems.
The answer will define not just India’s technology future, but its strategic autonomy in the age of AI.
Amit Mangal writes on India’s foreign policy and geopolitics at ThirdPol. Follow ThirdPol on X and LinkedIn.