The GPU gap in India's AI dream

India has entered its AI awakening, driven by talent, investment and a fast-modernising digital population. But the country's ambitions face an undeniable bottleneck: the near-total reliance on foreign-made Graphics Processing Units (GPUs) and accelerators.

India's AI momentum is real, but it is also fragile. The country seems to be entering a rare convergence of talent, capital and policy intent. AI-first startups are emerging fast, global investors are paying attention and the government is also moving fast on deep-tech infrastructure. As per a paper released by Bessemer, the Indian IT sector is projected to reach $400 billion by 2030, with AI playing a key role in this growth.

However, the foundations of the AI boom rest on a narrow and vulnerable layer: the access to computer or hardware resources that make AI models work. For decades, India's tech dominance has come from scale through millions of engineers, global outsourcing contracts and cost efficiency. AI is reshaping that equation.

Ranjit Tinaikar, CEO of Ness Digital Engineering, says, "GenAI is transforming the entire software development industry by significantly increasing productivity and accelerating innovation cycles, ultimately reducing time to market. However, its potential will not be fulfilled if it is viewed as a mere code generation tool, a flawed approach that is common in the world of software development."

The new winners are not service providers but platform builders, like companies that own models, data pipelines and automation systems. With players like Sarvam AI, NxtGen, and BharatGen, India is beginning to build core AI IP, foundation models, agentic platforms and developer tools.

That shift is already visible. AI-native firms such as Graph AI and Leena AI are building software-first businesses with minimal human intervention. Hybrid players like Crescendo and Shopdeck are combining automation with human oversight to increase speed and reduce costs. Infrastructure firms like Scale and Turing are becoming essential enablers by supplying data pipelines and model evaluation layers.

At the same time, a new class of founders, smaller teams, younger builders and solo entrepreneurs, competing directly in global categories, is emerging. NxtGen entered the AI space, bringing India's first full-stack agentic AI platform, "M," while powering the IndiaAI Mission alongside Jio Platforms, Tata Communications, Yotta and others. Solo founder Bishal Saha started building the Duolingo alternative, backed by AI and Bhashini datasets. Prominent AI-first founders include Bhavish Aggarwal (Krutrim), who is building Indian language LLMs, Maithra Raghu (Samaya AI) for financial AI and Sharbel Cherian (KeyValue Software), known for bootstrapped AI products.

This signals a break from the traditional outsourcing mindset and the beginning of an IP-led ecosystem.

Global capital is responding quickly. Google and Accel have announced investments of up to $2 million in Indian AI startups through a new partnership. In September, Google also partnered with Sarvam AI, Soket AI and Gnani.ai, while IIT-Bombay's BharatGen initiative focussed on advancing AI for Indian languages. Such initiatives show that India is being taken seriously as an AI development hub. Telecom players are also repositioning themselves as AI distribution platforms, bundling models into consumer services and accelerating adoption at scale.

Yet this expansion is running into a structural limit, namely hardware.

India's AI growth today depends almost entirely on foreign-made GPUs from NVIDIA, AMD and Intel. The IndiaAI Mission relies on roughly 34,000 GPUs, a modest number compared to what leading AI labs consume globally. And unlike software, hardware resources cannot be scaled overnight.

Export controls, geopolitical tensions and supply concentration mean that GPU access is not guaranteed. Analysts warn that tighter controls, similar to those imposed on China, could eventually restrict availability or inflate costs.

"Today, GPUs and other AI accelerators are under export control and are already in place in certain countries [adversaries]. As model capabilities continue to increase, there is a world where those controls could be extended further," Myron Xie, an analyst from SemiAnalysis, has said.

In that case, Xie said, companies like NVIDIA would still have to comply, even if it hurts their business. He added that, as seen with China's export controls, this could lead to a black market where dishonest players illegally resell GPUs.

If that happens, India's AI ambitions could slow sharply, not due to lack of talent or capital, but due to lack of silicon.

Meanwhile, as global big tech eyes the Indian AI consumer, telecom giants seem to be quietly becoming AI companies. Reliance Jio, Airtel and others are no longer just connectivity providers. They are becoming AI distribution platforms. In October, India got free access to ChatGPT when OpenAI offered a year's free ChatGPT Go access countrywide. Google partnered with Reliance Jio to offer its Gemini Pro plan worth Rs 35,100 free for 18 months to millions of users. Before that, Perplexity did the same with Airtel and OpenAI announced the rollout of free ChatGPT Go for Indians via Jio.

Telecoms could become India's most powerful AI gatekeepers even as big tech squeezes into the Indian market. ChatGPT's parent OpenAI plans to open its first India office in New Delhi later this year, deepening its push in its second-largest market by user numbers.

This creates a deeper concern and a widening AI divide.

Could India's 'GPU gap' create a new form of digital inequality?

Large conglomerates such as Jio, Tata and Yotta can afford dedicated GPU clusters and long-term infrastructure bets. Startups, researchers and smaller builders often cannot. As GPU prices rise and cloud access tightens, innovation risks becoming centralised among a handful of players.

That imbalance could determine who gets to build foundational models and who is forced to operate at the application layer.

At the same time, India's rising AI consumption, driven by free or subsidised access to tools from OpenAI, Google and others is turning the country into one of the world's largest AI user bases. For example, Perplexity and OpenAI both prioritised India early in their free rollouts. This could make India a must-win consumer market for global AI firms.

This gives India leverage as a market, even if it lacks hardware sovereignty. Global firms now treat India as too important to ignore.

India's advantage lies in its timing. With China constrained by export controls and western markets saturated, India has emerged as a geopolitical middle ground, a potential "third pole" in global AI development.

But that window will not stay open indefinitely. To sustain its momentum, India must move quickly on three fronts -- expanding domestic compute capacity, building skills beyond prompt usage into model engineering and reducing dependence on foreign hardware over time.

The ambition is evident. The talent is present. The policy push is unusually fast. What remains uncertain is whether India can turn this moment into long-term AI sovereignty or whether its next growth phase will once again be limited by infrastructure it does not control.