Rhodium: India's $300 billion AI data centre rush outpaces local demand

India is building computing capacity for artificial intelligence (AI) faster than its own businesses and government can use it, according to a report by the US-based research firm Rhodium Group, covered on Monday, 2 November 2026. The report, titled "Hyperscaler Heaven? A Reality Check on India's AI Ambitions" and written by Ankit Tiwari and Reva Goujon, says investment commitments for AI data centres in India have passed $300 billion over the last year. It argues that India's strongest AI opportunity lies in services delivered by its IT industry rather than in the data centres themselves.

The Rhodium report does not dispute the size of India's data centre investment. Its point is about who uses the capacity once it is built. If most of the demand comes from foreign clients, India becomes a host for other countries' AI work rather than a large user of AI itself.

A few terms help. A GPU is a chip originally made for graphics that is now the main workhorse for training and running AI models. A "hyperscaler" is a very large cloud provider that runs data centres around the world. "GPU-as-a-service" means renting computing time on GPUs by the hour instead of buying the chips. Utilisation is the share of time those chips are actually doing work; low utilisation means expensive hardware is sitting idle. Under the IndiaAI Mission, the government subsidises access to GPUs so that Indian start-ups, researchers and agencies can use them at lower rates. A reported utilisation of about 20% suggests that even cheap access has not yet created enough domestic demand.

The report's alternative is the services layer: building, adapting and running AI systems for clients. That is the business Indian IT companies already know, and Rhodium says value in AI is shifting towards it. For IT firms and their enterprise clients, the report shifts attention from how much capacity is built to how much AI work is actually delivered.

There is also a geopolitical side. The report argues that compute based in India makes the country central to US AI export strategy and economic security planning, despite India's stated non-alignment, and expects Washington to attach tighter economic security conditions as India grows into a global AI compute hub.

Peepals calculation: Commitments of $300 billion over one year work out to about $25 billion a month on average ($300 billion ÷ 12). A utilisation rate of about 20% means that, on average, roughly four of every five subsidised GPUs are idle at any time (100% − 20% = 80%). And if nearly 80% of Yotta's revenue comes from foreign clients, about one rupee in five comes from Indian customers. Our calculation from the reported figures.

Short-term: The report adds to questions about whether new AI data centre capacity in India will find enough paying users, especially Indian ones.

Long-term: If the report's view holds, India's main gains from AI would come through IT services exports and enterprise AI projects, while data centres serve largely foreign clients under conditions influenced by US export rules. The report identifies government purchasing rules as one barrier to domestic demand.

Who is affected: Data centre developers and operators, GPU-as-a-service providers such as Yotta, Indian IT services companies, AI start-ups that use subsidised compute under the IndiaAI Mission, state governments offering incentives, and enterprises planning AI projects.

Questions and Answers