Concern is growing that India, despite its early lead in IT services, has not built the base needed for frontier AI.

A recent suspension of access to US frontier models under export controls showed how much control lies with the countries that build them.

• Services era: Indian firms prospered by customising and maintaining software built in advanced countries, using a large pool of software talent.
• AI era: AI can now write, test and maintain code on its own, which strikes at the labour advantage on which India built its services exports.
• Global capability centres (GCCs): They are booming and have kept the sector resilient, but they do not build frontier models.
• IndiaAI Mission: Launched in 2024 with a five-year budget of about $1.2 billion, it has expanded shared compute to over 45,000 GPUs at subsidised rates for start-ups and researchers.

The gap comes from low research spending and a business model built on services.

• Low R&D: At 0.84% of GDP, India’s research spending is far below China’s and the US’s, so it has not given frontier AI a base.
• Scale of resources: The IndiaAI Mission’s budget of about $1.2 billion over five years is small against annual training and inference costs of $45.8 billion for OpenAI and $19 billion for Anthropic in 2026.
• Services habit: Indian firms earned by customising software built elsewhere, so they did not need to build core technology.

• Entry-level pressure: An ICRIER survey of 651 IT firms found that 55% reported a fall in entry-level employment and 35% an increase, while senior-level jobs stayed largely stable.
• Layoffs and hiring: Nomura’s analysis shows AI-linked hiring still exceeds layoffs, but job losses are concentrated in support and entry-level roles.
• Takeaway: The first effects fall on young and junior workers, who form the base of India’s IT workforce.

Hosting data centres without a presence elsewhere in the AI value chain may lock India into the lowest-value link.

• Thin gains: The hardware is imported, most jobs are in construction, and the models running in data centres may be foreign-owned.
• Power demand: AI data centres may add about 26.3 GW of demand by FY32, but they need round-the-clock power that renewables alone cannot give without storage. The night-time gap may be filled by coal, and the cost may fall on other consumers.
• Semiconductors: Of the 12 approved facilities, nine only assemble, test and package chips, and none makes leading-edge AI chips.

• Research base: Low R&D spending and a services-led model limit the ability to build home-grown models.
• Cost of compute: Frontier models need very large training and running budgets, far above a public mission’s budget.
• Job risk: AI is cutting into entry-level work, which is where many young graduates start.
• Grid strain: Large and round-the-clock data-centre demand may raise costs for other consumers, and demand swings can also strain slow-ramping coal plants.
• Value chain position: Without commitments from investors, India may stay at the lowest-value end of the AI chain.