India's AI problem: Gaps in compute, hardware, infrastructure, chips affect readiness

In India, like in much of the rest of the world now, talking about Artificial Intelligence makes one suddenly feel important. But is India AI-ready? The Asian giant is slowly coming to terms with and addressing its AI problem. It has several aspects, but the main gaps are in software, hardware and infrastructure. The AI skill gap among youth is tied to slow curriculum updates from primary school to college. AI regulation is still evolving. India's traditional IT sector is bleeding as global AI tools eat up the work it used to bill for. Slow AI adoption by these firms is making things worse. Here is a snapshot of India's AI problem.

Where India stands on AI readiness

International benchmarks place India in the middle. The IMF's AI Preparedness Index put India at 0.49 and ranked it 72nd of 174 economies, trailing China (0.63, rank 31), Singapore (0.80), Indonesia (0.52), and the US (0.77). The IMF warned AI could affect about 24 per cent of jobs in emerging economies like India's. India ranks higher on talent metrics. The Stanford AI Index 2025 noted India leads on AI hiring rate, at about 33 per cent annually.

India's AI compute and hardware gaps

Global breakthroughs in hardware, software, frontier models, agentic AI and chip design are the result of two decades of work, if not more, mainly in the US and China. India's AI-related compute was near zero until recently. Under the IndiaAI Mission, national common compute crossed roughly 34,000 GPUs by 30 May 2025: initial 18,417 GPUs plus 15,916 new units, according to state media. A third tender took the cluster past 38,000 units, including 1,050 Google Trillium TPUs.

At the AI Impact Summit in February, Minister Ashwini Vaishnaw pledged 20,000 more sovereign GPUs, lifting public capacity up to 58,000 units. The government now targets 100,000 public GPUs by December 2026. Private deployments by Reliance, Tata and hyperscalers would match that, pushing total national capacity past 200,000 by year-end.

Access to compute remains cheap in India. Subsidised rates run around 65 to 150 rupees per GPU-hour, roughly 42 per cent below market. But here is the catch: Not a single one of those GPUs is designed or made in India. They are mostly NVIDIA H100, H200 and Blackwell cards.

Frontier AI models and AI chips: India lags behind

India trails on frontier models and advanced chips despite the buildout. India lags in producing large-scale frontier models mainly due to a lack of access to advanced chips, the scale of modern data centres, and the complexity of training the models across India's multiple languages. Vaishnaw said India aims to attract $200 billion in data-centre investment. Reliance is building a 1-gigawatt data centre, expandable to 2GW, in Gujarat on NVIDIA Blackwell processors, at an estimated $20–30 billion cost.

How much of India's AI infrastructure is actually sovereign?

India's current AI infrastructure is mostly foreign-funded and foreign-chip-dependent. Foreign companies have, or are planning to, invest in AI in India. Microsoft pledged $17.5 billion for India on December. Google committed $15 billion for an AI hub in October. Amazon Web Services committed $12.7 billion through 2030.

Chip import dependence and slow fabrication: India's weak points

India meets close to 85 to 90 per cent of its chip needs through imports, and forms only about 3 per cent of the global market. More than 60 per cent of imported chips come from China and about a fifth from Japan. India is strong in design, holding about 20 per cent of the global semiconductor design workforce, but weak in fabrication and commercialisation. The fabrication gap will not close fast. McKinsey noted that India is pushing for fabrication at nodes above 14 nanometres by 2030, while sub-10-nanometre technology will take longer. High capital costs, limited access to advanced manufacturing technology, and supply-chain gaps in high-purity gases, special chemicals and ultrapure water are among the reasons. The India Semiconductor Mission's roughly $9.1 billion commitment has catalysed about $18.2 billion in investment. But the real movement is mainly in packaging and legacy nodes, not AI chips.

India's AI factories: A snapshot

Indian Budget 2026-27 allocated 8,000 crore rupees to the newly launched semiconductor mission, the largest single-year outlay yet. On the ground, the progress results are only starting to show. Micron's assembly and test facility in Sanand, Gujarat, which reached initial commercial production and was inaugurated by PM Narendra Modi on February 28, is India's first modern high-volume chip facility. But it handles assembly and packaging of memory, not wafer fabrication. Kaynes Semicon's OSAT facility in Sanand opened in March. The Tata–PSMC fab at Dholera targets first silicon by the end of this year, with capacity for 50,000 wafers a month. None of these plants targets AI chips.

AI vs jobs: What India's Chief Economic Advisor said about the nation's AI situation

AI, robotics and automation threaten entry-level service jobs, as millions of Indians with traditional degrees enter the job market each year. The remarks of India's Chief Economic Advisor V Anantha Nageswaran in the past few weeks are noteworthy in this scenario. He said AI is a growth and stability question, not just a jobs question. In his view, AI cannot be adopted passively and must be deliberately aligned with creating jobs at mass scale. Calling it a decision about the future of growth and social stability, he said the window is “open but not indefinite”. In his view, India can lead globally only by aligning AI adoption with mass employability through education reform, pedagogy change and stronger foundational skills. India must create at least 8 million jobs each year for the next 10 to 12 years to reach developed-nation status by 2047, he said.

But are Indian youth learning the AI skills required to enter such a job market? That's a story for another day.