With $200 million in funding, engineers at the Centre for Development of Advanced Computing (C-Dac) in Bengaluru have been tasked with developing the chip, according to four senior officials familiar with the matter, three of whom are directly involved in the project.
The first official said that Nvidia’s product roadmap shows that by 2028, cutting-end chips will be based on the 2nm node. “This means that by 2030, the best GPUs in mainstream circulation in data centres and for AI training will be at this standard,” this official said. “That’s what our GPU will achieve too, but at a much, much lower cost.”
To be sure, the smaller the nanometre size, the more advanced the chip. The most advanced mainstream chips of today are of 3 nanometre, such as the ones found in Apple’s iPhones, among other consumer devices.
Since ChatGPT’s debut in 2022, GPUs have become essential to AI—boosting Nvidia’s value tenfold and making it the world’s second-most valuable company. Despite India’s strong chip design talent, it lacks homegrown GPU patents, leaving the country dependent on US firms for core AI technology. It is this dependence that the country is looking to change.
The second official cited above said that an early preview of the chip will be showcased by end-2025, whichMinthad reported last month. However, once the 2nm chip is developed by C-Dac, India is unlikely to get in the next five years a domestic fabrication plant with the capability to manufacture such chips, which is why “we’ll likely be manufacturing it at scale with Taiwan Semiconductor Manufacturing Corp (TSMC)”, the second official said, adding that the Indian GPU will cost “up to 50% less than what Nvidia currently retails its chips at”.
An email requesting comment on the matter from Meity and C-Dac remained unanswered till press time.
A chip customer
To be sure, India has historically been a customer for US chipmakers Intel, Advanced Micro Devices (AMD), Qualcomm and Nvidia, but an executive order by former US president Joe Biden, signed last year, showed that in case of conflicts, the US can restrict access to critical chips to India or any other nation.
The third official cited above said that this order was a key moment for India to start seriously weighing the idea of building its own chip. “Since then, we’ve been engineering an indigenous GPU from the ground-up,” this official said. “By 2030, we’ll be installing it on C-Dac’s cloud servers and supercomputers—making it accessible to academia, researchers and startups to make our own sovereign AI models and run cloud platforms.”
Industry stakeholders have for long urged India to develop its proprietary semiconductor IPs for geopolitical independence. Last month, Ajai Chowdhry, chairman of HCL and cofounder of industry body Epic Foundation, toldMintthat “a domestic GPU patent based on the government-funded research bodies is imperative, especially seeing that almost all chips today are owned by the US”.
Security concern
“The necessity of sovereign technologies also comes from a security concern,” said Ashok Chandak, president of industry body, India Electronics and Semiconductor Association (Iesa), pointing out that much of the chip supply chain is today reliant on China.
“In the long run, being reliant on China can make critical chips such as those used in CCTVs or automation in industrial infrastructure vulnerable to back doors,” Chandak said. “An indigenous chip will address all of these concerns. Having our own GPU chip is also vital, since it can allow India to not only make and train its own AI on such a chip, but also market it to the rest of the world.”
Meanwhile, C-Dac’s objective is to offer the indigenous GPU as a system-on-a-chip (SoC) board, which will work as a full-stack system including memory chips, computing processor and connectivity modems as well.
The body is well-funded, too. In FY24, C-Dac had capital fund allocation of ₹1,056 crore ($122 million) for the year from the Centre, per its annual report for the fiscal. Two of the officials cited above said that this figure has been increased in FY25.
To be sure, the $200-million engineering design cost of the indigenous GPU will be spread over five years—from fiscals FY25 through FY29,according to the first official.
