Tech Explained: Here’s a simplified explanation of the latest technology update around Tech Explained: Govt to rejig PLI 2.0 sops on high GPU prices, AI server tech in Simple Termsand what it means for users..
GPU costs are set to contribute almost 90% of the cost of servers, up from 70% earlier, with users migrating to AI servers, which still have significant potential for domestic value addition, said Sushil Pal, joint secretary at the electronics and information technology ministry (MeitY).
“We need to be aware of how we can capture this value chain,” he said. “We need to review if the PLI IT hardware scheme still fits the requirements for AI servers.”
The scheme provides incentives based on incremental turnover or net sales of manufactured goods. This structure needs to be changed if up to 90% of the cost for components is not manufactured locally, Pal said, noting that commercial high-performance GPUs are not manufactured in India.
The first PLI scheme for IT hardware was notified in 2021, while the second iteration currently in place was notified in 2023 to create a robust domestic manufacturing ecosystem for hardware including laptops, tablets, and servers.
“The AI server story wasn’t completely understood at that point in time,” Pal explained. “Given the different nature of AI servers, we need to examine this. Talks with the industry will begin.”
He was speaking at the launch of a new Made-in-India AI server portfolio by Altos Computing.
The Acer Group subsidiary is targeting $2 billion in revenue from the Indian server market over the next two years and aims to become one of the top three server brands in India within three years, CEO Jackie Lee told ET.
Memory prices are likely to remain elevated for the next few quarters and the prices of Altos’ own products may rise, Lee said, noting that costs for high-performance components are rising 3-5% every week. “This is driven by a shortage of DDR memory, as major manufacturers like Samsung, Hynix, and Micron have shifted capacity to accommodate the explosive demand for high bandwidth memory (HBM),” he said.
Altos also plans to invest in R&D resources and hire software engineers in India.
The company’s latest Altos Brainsphere R300 AI series servers, designed for next generation AI workloads, will be ready for commercial supplies within a few weeks, Lee said.
Altos India director Priya Krishnamurthy said the company’s compute products can hit between 60-80% local value addition, depending on the CPU used, as compared to the industry standard of 20-21%.
In line with the government’s sovereign AI push, the company has developed a series of “air-gapped” products, which ensure servers are located within captive networks behind firewalls, and no data is transmitted via cloud, she said.