Most conversations around AI in India are positive — new apps, innovative startups, and rising productivity. That optimism is understandable. But beneath the excitement, a more serious shift is taking place. The core infrastructure of AI — the chips that power it, the cloud systems that run it, the foundational models that guide it and the data systems that feed it — is becoming concentrated in the hands of a very small number of global companies. When the same few firms control all four layers, it becomes extremely difficult for new players, including entire nations, to compete — no matter how skilled or ambitious they are. In the last eighteen months, regulators in the United States, United Kingdom and European Union have all raised the same warning: this kind of concentration can create long-term dependency and limit who gets to shape the future of AI.
For India, this is not just a market concern—it is a sovereignty concern. Our hospitals, schools, banks, courts and defense systems will soon rely on AI for everyday decisions. If the computing power and the core models behind these systems are owned and operated abroad, then India will, in effect, be running critical parts of our society on borrowed intelligence. That would mean becoming dependent by default, before we even realise it has happened.
Where the power is concentrating: The real concentration of power in AI is happening across three layers: chips, cloud and core models. To begin with, training large AI systems requires thousands of specialised high-performance chips and today one company holds the clear majority of this market. European regulators have already raised concerns that this dominance may be reinforced through software and system-level bundling. Even if India secures these chips, the biggest training facilities are located inside the data centres of just three global cloud giants—Amazon, Microsoft and Google. In regions like Europe, these three already account for nearly 70 per cent of all cloud spending, which shows how quickly domestic capacity can be overshadowed once these platforms become entrenched.
At the top of the stack are the foundation model developers, many of whom are financially and technically linked to these same cloud and chip providers. The US Federal Trade Commission has formally examined these partnerships and warned that such ties could quietly limit competition and make it harder for new countries or companies to catch up. Meanwhile, the European Union has introduced rules for general-purpose AI under the EU AI Act, with key obligations taking effect in 2025 — standards that will likely influence the global market India interacts with. For us, this isn’t just a regulatory observation; it’s a signal that if we don’t secure our own capabilities now, we may have to adapt to decisions made elsewhere.
India’s starting point: The Union Cabinet’s IndiaAI Mission, with Rs10,372 crore and a plan to set up 10,000 GPUs, is an important beginning. It will help Indian researchers and startups get access to computing power. But this alone is not enough, because the world’s biggest chip makers and cloud companies still control most of the advanced AI infrastructure. If that does not change, India will remain dependent when it comes to training the largest and most powerful AI models.
On the policy side, the Government has understood that old competition laws are too slow for the digital age. A new Digital Competition Bill has been proposed to prevent big tech companies from blocking fair competition early on. Meanwhile, the Competition Commission of India has already studied where AI power may concentrate. This preparation is essential if India wants strong and independent AI capability.
Why this matters for everyday India
Think about the next decade. AI will help forecast monsoon variability for crop insurance, triage patients in district hospitals, assist judges with research to cut case backlogs, monitor freight and power flows on the grid, and watch our coasts. If the core models and compute live in someone else’s legal jurisdiction —and if they can change licensing terms or turn down access — then we don’t really control the system that thinks on our behalf. That is not a tech inconvenience. It’s a Governance risk.
What India should do now
This is not a call to close our doors to the world. India has always grown stronger by learning from the world, not by isolating itself. The real goal is to ensure that we do not become dependent on systems we cannot change, influence, or even fully understand. To secure our future, India now needs a
clear and practical path — not slogans, not vague ambition.
The first step is to build a National Compute Grid. Think of it as India’s “AI railways.” Just as rail networks connected markets and people, this grid would connect compute power to everyone who needs it — startups in Bengaluru, researchers in Hyderabad, MSMEs in Surat, defence labs in Pune, universities in Guwahati. Facilities already exist across C-DAC, IISc, IITs, ISRO and key research labs. What we need is to stitch them together into one federated network, governed transparently, with fair access. International bodies like the OECD have advised countries to treat computing capacity as critical national infrastructure. India should do so before scarcity turns into dependency.
Second, we must avoid being locked in. When Government departments adopt AI systems, they should not be tied to a single cloud provider or a single model ecosystem. Every public AI system must allow: Switching between cloud platforms, Moving data easily when required and Running on different hardware. The UK and US have already warned that exclusive AI partnerships can quietly squeeze out competition. India can avoid this trap simply by putting portability rules into government procurement standards. If public systems remain flexible, the entire economy remains flexible.
Third, India must treat AI safety and sovereignty as two sides of the same coin. Transparency reports, security testing, and red-teaming are important — yes. But in India’s context, one more safeguard is essential: for critical sectors such as elections, healthcare, banking, welfare delivery and national security, any fine-tuning of models must happen inside India and under Indian law. This is how we ensure accountability without slowing innovation. Finally, India must build models that speak India’s languages, understand India’s contexts, and reflect India’s lived experience. Our linguistic and cultural diversity is not a burden—it is a superpower. No country in the world has our depth of multilingual, code-mixed, socially layered communication. If we treat language datasets and sector-specific knowledge in agriculture, law, climate and public health — as national knowledge infrastructure, we can create open, India-trained AI models that understand how Indians speak, work, negotiate, teach, argue and decide. That is how AI grows with Indian society, not over it.
The bottom line
If we drift, the likely outcome is simple: a few firms abroad will control the chips, the clouds, and the most capable models. Indian companies and agencies will pay a silent tax every month — through rents, restrictions
and roadmaps we don’t control. That is the cartel risk in plain language. If we act now, we can write a different story: a National Compute Grid that any accredited Indian builder can use; procurement rules that keep us portable; language and sector models that reflect our society and regulators with the technical depth to keep markets contestable. We’ve done it before. Aadhaar, UPI and CoWIN were built because India decided to make digital infrastructure a public mission, not a private moat. AI needs the same clarity and scale. The global window is open — but it won’t stay open long. Let’s build an India that owns its intelligence, instead of renting it.
Author is a theoretical physicist at the University of North Carolina, United States; views are personal
