Home-made chips: India's GPUs set for year-end trials
Summary
The indigenous GPU will be a general-purpose semiconductor chip, and can be used in the long run for various tasks such as local cloud platforms, data analytics, AI training and more.India will begin trials of its first homemade graphic processing units before the end of this year, two senior officials familiar with the matter said, crossing a milestone in the development of components at the heart of artificial intelligence (AI) systems.
The chips will be general-purpose GPUs, or GPGPUs, with the potential use across a wide range of tasks. In the long run, the goal is to replace chips from global giants such as Advanced Micro Devices (AMD), Intel, Nvidia and Qualcomm in data centres, AI model training, analytics, cloud platforms, supercomputers and critical communications infrastructure, among others.
“About 29 prototypes of these indigenous GPGPUs will be produced in India by the end of this year. Following this, agencies associated with the development process will collate data on performance, reliability and efficiency, since the idea will be to ensure that the product in question is not just indigenous, but also at par with global standards set by the US," one of the officials cited above said on the condition of anonymity.
Also read | No GPUs, no problem: Ziroh Labs can run AI models just with CPUs
On 5 February, Mint reported that the Centre was in the process of designing its own ‘AI chip’, setting a tentative deadline of 2027 for local production. The project is headed by Meity, with the Centre for Development of Advanced Computing (C-Dac) and National e-Governance Division (NeGD) working on it. The GPGPUs are being built on an existing indigenous semiconductor architecture developed by C-Dac, based on open-source standards.
After trials, the GPGPUs will be finetuned before being put up for mass production and commercial scaling by 2029, the two officials said on the condition of anonymity.
“What an indigenous GPGPU will do is help India reduce its dependency on global firms, and their ability to restrict access to critical computing infrastructure in crucial moments. It’s not possible for India to do everything independently—a GPU typically uses myriad associated patents and intellectual properties (IPs). But, what the indigenous chip will do is that India will no longer need to outright buy GPUs for projects such as sovereign cloud deployment, or training of AI models at an academic level," the second official said.
Read this | Nvidia's Blackwell GPU may reach Indian shores as early as October
Currently, the US has an outsize impact on the global market for semiconductor patents. In 2024, the US earned 72% of the global revenue related to semiconductor design, data from market researcher Gartner showed. In its 2019 trade tiff with China, the US fully leveraged its leadership position in semiconductors to hurt Huawei Electronics, then one of the top global tech conglomerates.
It is this situation that India eventually wants to avoid.
Custom GPUs can be deployed in homegrown supercomputers, improving access to central computing infrastructure for academics and researchers, the second official said. "In the long run, the Centre plans to use domestic GPUs in homegrown cloud platforms and supercomputers which, in turn, could help India closely protect crucial technology patents. Beyond 2030, most strategic developments around the world will be hinged on technology, and every technology has semiconductors at its core—they simply cannot be avoided," the second official added.
And read | Murugappa's chip testing plant to begin supplies next year, says JV partner
Kanishka Chauhan, senior principal analyst at Gartner, said that challenges behind building an indigenous chip are manifold.
“Chips are designed on electronics design automation (EDA) tools, which have significant licensing fees that run into millions of dollars. Plus, design validation is an iterative process and is extremely time consuming, which necessitates sound financial backing for any entity pursuing it. GPUs also lack a standardized architecture unlike CPUs, which makes the engineering process required to build GPUs even more complicated," he said.
Ashok Chandak, president of India Electronics and Semiconductor Association, added that the key thing to consider would be domestic value addition through the creation of IPs. “Having your own production facilities as well as patents would be key to generating greater value. Most of the percentage of the cost of manufacturing in electronics is accounted for by semiconductor chips—having our own patents will help India further climb up the value chain," Chandak said.
Read this | Mint Primer: Why India’s first locally made chip will be a big deal
topics
