The cost of building artificial intelligence (AI) data-centre capacity in India could rise by 8-12% if reported price increases by Nvidia on its AI server systems are fully passed through, adding pressure to an infrastructure build-out that remains heavily dependent on imported hardware.
Nvidia has reportedly told some of its largest customers that prices for servers built around its Grace Blackwell and upcoming Vera Rubin platforms will rise by more than 15% for systems shipped from early 2027. Contract manufacturers that build these systems for companies such as Microsoft, Google and Oracle have also begun notifying customers of the coming increases, which will vary depending on chip generation and memory configuration.
Hardware Price Inflation
The impact on India, however, will depend on how much of the increase is ultimately passed through to data-centre operators. Servers typically account for a little over half of the total cost of an AI data centre, with the rest going towards land, power infrastructure, cooling, networking and other facilities.
“Servers generally account for a little over half of the total cost of an AI data centre. The remaining expenditure goes towards land, electrical infrastructure, cooling, networking and other facilities,” Ravi Kumar Etamsetti, founder and chief executive officer of TierX Data Center Solutions, said.
This means a 15% increase in server costs would not translate into a similar increase in overall project expenditure. Even so, the effect could be significant for an industry that requires large upfront investments and is still working towards improving utilisation of expensive AI compute capacity.
The first impact may be felt in project economics rather than customer pricing. Large hyperscalers have greater bargaining power and can spread higher hardware costs across larger customer bases, while smaller GPU-cloud operators have thinner margins and less room to absorb increases.
“Broadly, overall Nvidia GPU prices in India are already inflated. So, suppliers have a huge margin,” Raghav Gupta, founder of Futuresense, said.
Market Reconfiguration
Higher hardware costs could therefore make operators more selective about adding capacity ahead of demand. Instead of simply acquiring more GPUs, the focus could increasingly shift towards getting more computing output from existing infrastructure through better utilisation, orchestration and workload optimisation.
“The era of building enormous GPU capacity simply because demand is expected to arrive may become harder to justify,” Krupesh Bhat, founder and chief executive officer of Melento, said. “The next wave of investment will likely favour utilisation, orchestration, and workload optimisation alongside physical capacity.”
This could also give alternative accelerator platforms a greater opening in India. Companies such as AMD and Intel, as well as cloud-specific chips such as Google’s TPUs and Amazon Web Services’ Trainium and Inferentia, offer alternatives to Nvidia hardware.
However, price alone is unlikely to be enough to dislodge Nvidia, particularly for large-scale AI training. Its established software ecosystem remains a significant advantage, while workloads such as inference are likely to be easier to migrate across competing hardware platforms.