The government is set to revisit its production-linked incentive (PLI) scheme for IT hardware to better align with the rapidly evolving economics of AI servers, a senior Ministry of Electronics and IT (MeitY) official said on Monday. The move underscores both the opportunity and the structural constraints in building a domestic AI hardware ecosystem, where a large share of value is concentrated in high-end chips and GPUs.
Speaking at the launch of Altos AI servers, Sushil Pal, Joint Secretary at MeitY, said that the current PLI scheme for IT hardware, launched in 2023, did not fully anticipate the rise of AI servers.
What else did Sushil Pal say?
“The scheme mentions servers, but not AI servers specifically,” he said, adding that AI servers differ significantly from traditional servers in both architecture and value distribution. Unlike conventional hardware, where value addition is more evenly spread, AI servers are heavily dependent on high-end chips and GPUs.
“About 70% of the cost of an AI server today comes from GPUs or silicon. With newer platforms, this could rise to 90–92%,” Pal said.
This skew raises concerns about how much value India can realistically capture domestically under a turnover-linked incentive model.
“If 90% of the value is not being created here, we need to examine what exactly we are incentivising,” he said, adding that the government will soon begin consultations with industry stakeholders to reassess the scheme’s relevance.
The official confirmed that AI servers currently fall under the existing PLI framework for IT hardware but indicated that their distinct nature warrants a closer policy review. The PLI 2.0 scheme for IT Hardware was launched in May 2023 with a budgetary outlay of 17,000 crore for a six-year period. It covers laptops, tablets, all-in-one PCs, servers, and ultra-small form factor devices.
Pal added that data centres and AI servers are a critical infrastructure for the economy, with India aligning itself not just for domestic demand but as a global manufacturing hub.
The country is on a path to emerge as a manufacturing base for AI hardware, Pal said, noting that around 70% of certain AI server manufacturing processes are currently being carried out in the country. This share could rise to 80% in the near future, he said.
However, Pal flagged energy availability as a bottleneck to scale AI infrastructure. “Anything that could constrain this sector going forward is energy,” he said, pointing to the high power demands of data centres and AI workloads.