India is buying AI chips at a rising clip — and the harder question, as The Economic Times reports, is no longer acquisition but whether the country can keep them running once they're installed.

That reframing matters. For two years the story out of India has been procurement: imports of GPUs and AI accelerators climbing as data center operators, cloud providers and government-backed compute programs race to build out capacity. Buying is a purchase order. Running is an operations problem — power, cooling, skilled staff, software stacks and maintenance contracts spread across facilities that were often never designed for kilowatt-hungry accelerators.

What changes when the chips arrive?

Everything, basically. An AI accelerator in production draws far more power and dumps far more heat than the general-purpose servers most Indian data halls were built around. Keeping a fleet of them at high utilization means:

• Stable, high-density power delivery around the clock
• Liquid or advanced cooling retrofits in halls built for air
• Engineers who can diagnose failures at the node, driver and cluster level
• Software and systems talent to keep training and inference workloads from idling expensive silicon

The chip is the input. The output — usable compute-hours — depends entirely on the infrastructure and people wrapped around it.

Why utilization, not volume, decides who wins

If India's AI ambitions are measured in chips purchased, the numbers will look impressive and tell you little. If they're measured in sustained, efficient operation of that silicon, the picture gets sharper — and harder. Idle or badly run clusters burn capital without producing models, tokens or revenue. Well-run ones compound it.

That's the test The Economic Times identifies, and it's the same one every scaling AI market faces once the procurement race cools: the gap between owning accelerators and operating them at competitive efficiency.

The road ahead

India's AI compute buildout is still in its buying phase. Whether the country converts that spend into durable AI capability will be decided less on import invoices than in the data center — in uptime, utilization and the depth of the operating talent India can train around the chips it has bought.