In October 1985, the United States refused to allow the sale of a supercomputer to India fearing the government may use it for progressing nuclear research and building strategic defence resources. Pushed to a corner, India created the Centre for Development of Advanced Computing (C-Dac) in March 1988 with the singular purpose of building an advanced supercomputer as quickly as it could. By August 1991, C-Dac had built Param-8000—India’s first supercomputer.
India found itself in similar circumstances last year. In January 2025, the US decided to regulate global access to Nvidia Corp.’s high-performance computing chips. India was grouped along with Israel, Singapore and a few other countries on a list allowing limited access to the American semiconductor giant’s chips.
A few months earlier, presciently, India’s ministry of electronics and information technology (Meity) had begun discussions on an indigenous AI chip project to avoid being vulnerable to geopolitical pressures. Shortly after the US restriction, in February 2025, India announced it had begun work on an AI chip.
C-Dac, primarily a research institute under Meity, had been assigned a lofty mission similar to its founding years. It has to build an AI chip that can help India reduce its dependency on other countries for strategically important chip deployments, data processing, and AI usage such as in defence, aerospace, secure communications, and public services.
Four senior officials Mint spoke with said C-Dac was seen as the right fit for two main reasons. One, since August 2017, C-Dac has run Meity’s low-key Microprocessor Development Programme—a project that has led to the development of semiconductor products such as the ‘Vega’ processor cores and Dhruv64, India’s first dual-core processor. And two, in building the Param family of supercomputers, C-Dac gained expertise in deploying high-performance computing infrastructure that can be used for powerful tasks requiring vast tranches of information to be processed.
“C-Dac is the sole body with the expertise and know-how to create strategic chips, because we have done it before,” one of the officials said. “The engineering process could be steep and tricky, but only our engineers have such talent and capability in the country.”
Meity has assigned C-Dac’s AI chip project a target budget of $200 million (about ₹1,900 crore) over five fiscal years—from FY26 to FY30, according to the officials, who declined to be identified given the strategic nature of the project. For FY26 and the ongoing FY27, the budget allocation to C-Dac was ₹275 crore and ₹280 crore, respectively.
C-Dac’s AI chip may not necessarily become a commercial offering. But from the purview of sovereign independence, India will hope to deploy it in academia, research-led workloads, and local data centres to run public services. The goal, in the long run, would be to have an option to fall back on if geopolitical sanctions are ever applied on India.
A major step
Late in 2024, C-Dac began assembling a team of nearly 30 engineers, led by senior engineers with doctorates, former research scientists, and visiting faculty of top US engineering institutes, for the AI chip project. They were based largely in a seemingly regular office complex of C-Dac’s lush Bengaluru campus on Old Mysore Road.
C-Dac has built semiconductor chips before, such as Dhanush64 and Dhruv64 that are meant for general purpose tasks such as in smart meters and low-voltage sensors. The challenge before its team now is steeper. They have to design and build a chip that will be at par with the graphics processing units (GPUs) of Nvidia—the world’s most valuable company with a market capitalization above $5 trillion driven by global demand for its AI chips.
Modern artificial intelligence relies on high-performance chips like graphics processing units (GPUs) to process massive datasets and run large language model (LLM) algorithms, enabling computers to read, write, code, and converse like humans.
C-Dac’s task is to not emulate an Nvidia chip, but create India’s own chip design patent for an AI chip. It should be akin to chips that companies such as Nvidia-acquired Groq in San Jose have built—capable of high performance that’s required to generate results from generative AI interfaces, but with low energy consumption and low cost of manufacturing.
When this journalist visited C-Dac’s office in Bengaluru in June last year, the team was in multiple phases of technical deliberations over the development of this chip. The person leading this team, who requested anonymity since the government has not authorized the institute to divulge specific technical details, said the Centre has assigned “strategic importance” to the project.
Most of C-Dac’s early chips, including for its 2017 Microprocessor Development Programme and the 2022 Digital India Risc-V programme, too were built keeping strategic autonomy in mind. Most of these were built to find usage in sectors such as agriculture and earth observation. C-Dac, along with Meity’s Semi-Conductor Laboratory, currently India’s sole semiconductor chip fabrication plant, have also designed chips that the Indian Space Research Organisation (Isro) used in missions such as Chandrayaan and Mangalyaan.
C-Dac is working on other unnamed chip projects as well, including under the Digital India Risc-V programme. Mint could not independently determine the number of chips in C-Dac’s development pipeline.
As of August 2026, the first designs of C-Dac’s AI chip had entered trial manufacturing—the first major step towards creating a commercially successful chip. The internal timeline assigned to the project is to have a production-ready version of the AI chip “some time in 2029, and start deploying it in applications by 2030,” said the second of the four officials mentioned earlier. That’s a decent timeline. China’s Kunlun AI chip, developed by Baidu and unveiled in 2018, took seven years from concept to becoming commercial-ready.
Strategic choice
Up until October 2022, GPUs were largely confined to usage in supercomputers and gaming rigs. Unlike central processing units (CPUs), which have been used in mainstream computing devices such as personal computers and calculators since 1971, GPUs are specialized processors designed with a specific task in mind—handle multiple heavy tasks simultaneously.
In 1999, Nvidia, then a little-known chipmaking company based in Santa Clara, California, launched what was the first marketed product with the ‘GPU’ tag. For 23 years, Nvidia largely catered to a niche market—up until the public unveiling of generative AI by fellow Silicon Valley entity, OpenAI, in October 2022.
Generative AI, by design, requires processors that can churn massive chunks of data within very short spans of time to produce the kind of results that top companies such as OpenAI, Google, Microsoft and Anthropic have advertised. GPUs, the world concluded, were the best tool for this, and demand for Nvidia’s chips shot through the roof.
India was caught in a worrying geopolitical situation. For four months between January and May 2025, India was grouped alongside Israel, Singapore and other countries for which access to Nvidia’s GPUs was limited to 50,000 units a year. Among the core objectives of this regulation, titled ‘AI Diffusion Rule’, by the then-outgoing Joe Biden administration in the US were national security (primarily in the context of China and Russia) and global commercial leadership. It was rolled back amid industry pressure on the current US administration led by Donald Trump.
The episode left New Delhi concerned at the possibility of the US again restricting access to crucial semiconductor resources. “There may not be any import restriction on India right now, but that does not mean that we will not build our own technologies,” Union IT minister Ashwini Vaishnaw said during a 15 July media roundtable that included Mint. “Having our own AI chip is a strategically very important part of how India can run AI for a wide range of tasks, many of which will handle highly sensitive data that we will try to retain within our country. Work is happening at the pace it is expected to, and we are on track to match internal timelines for it.”
The AI chip takes shape
C-Dac’s AI chip, according to two officials, is being built using the 2 nanometre (nm) fabrication node. In simple terms, a fabrication node refers to the size of a single chip in a GPU. The smaller the size of the node, the more advanced it is. Nvidia’s latest ‘Vera Rubin’ class of GPUs, which are set to be sold commercially later this fiscal year, are based on the 3nm node.
In comparison, the Dhruv64 chip was built on the 28nm node, which is typically seen in electrical appliances. Private chip design startup Mindgrove Technologies Pvt. Ltd is expected to unveil its proprietary V2600 Vision chips, which it has rated for AI inference workloads, also on the 28nm node.
C-Dac’s AI chip will use an open-source reduced instruction source computer-V (Risc-V) core architecture, which too is a widely accepted industry standard. This, however, “does not mean that India’s AI chip’s core patent will not be Indian”, said the first official quoted earlier.
“The idea is that the 2nm standard will still be cutting-edge by the time India’s own AI chip will be ready. This will ensure that India does not remain import-reliant even for strategically important chips,” said a second official. “But to do that, it is not crucial to reinvent every wheel. A chip has multiple parts, and for things such as networking and other aspects, one need not create everything from scratch. Even an Nvidia chip is not 100% patent-owned, and this is the norm in the industry. India, too, is doing the same.”
The first official added that the chip is being designed “to handle modern-day generative AI and machine-learning use cases at performance levels at par with cutting-edge commercial chips, while reducing the energy consumption cost by making the chip’s architecture more efficient than what is currently available in the market today”.
After about 16 months of design-led development, C-Dac Bengaluru in February floated a tender through the Government e-Marketplace (GeM) platform for the chip’s trial production process, which HCL Technologies Ltd won. The latter is now in process of getting multiple versions of the chip’s design manufactured at Taiwan Semiconductor Manufacturing Co. TSMC, alongside Korea’s Samsung Semiconductor, are the two most advanced chip fabrication companies in the world.
“Once the trial production is done, C-Dac will then take possession of the chips, test them out in its indigenous supercomputing systems, and see how they perform in the real world—as well as what modifications would be required,” the first official said.
HCL Technologies and C-Dac Bengaluru did not respond to Mint’s emails seeking comment. S.D. Sudarsan, executive director at C-Dac Bengaluru, told Mint on the sidelines of Meity’s unveiling of its Semicon 2.0 incentive scheme on 15 July that “work around the indigenous AI chip is progressing as per the expected timeline”.
The next phase, industry stakeholders said, will be to validate performance standards once the different designs of the AI chip are manufactured, traditionally called ‘tape-out’. But the first tape-out versions of the chip, according to Kashyap Kompella, technology analyst and founder of consultancy firm RPA2AI Research, may not necessarily “be perfect and ready for final engineering stages—let alone deployment”.
“The chipmaking cycle typically follows multiple tape-out phases—at least two or three, which is where the biggest challenge lies because each tape-out is not only time-consuming, but also rather expensive as they can run into hundreds of millions each time depending on the volume of chips being made,” Kompella said. “What will be important to see [is] if the chip can move swiftly on to engineering validation and testing (EVT) stages, which is when a chip’s design is locked in and work begins to make a production version of it.”
Complicated but crucial
Jonathan Ross, chief software architect of Nvidia, said in an interview with Mint on 5 June that while making a proprietary GPU is “super difficult” in the current era, it is also important for countries to pursue sovereign chip designs and own chip patents.
“Nvidia’s success in chipmaking comes from the fact that it invests billions each year into research and development. At any given point, we’re likely designing and testing tens of different types of chip design, or maybe even hundreds. Only a handful make it to the final stages, and from that, only one–such as our new Vera Rubin chips–make it to production. This process is extremely capital-intensive, and requires decades of expertise in chipmaking to be able to make a truly cutting-edge commercial GPU,” Ross said.
“It is definitely a government’s imperative to work on sovereign technologies, and everyone should work on sovereign technical architecture today. But rivalling Nvidia or any other similar chipmaking company commercially through public funds may seem difficult—there may not be enough funds to rival private companies in this space.”
Sriram Viswanathan, founding managing partner of venture capital firm Celesta Capital, concurred with Ross, adding that alongside developing core sovereign technologies, India’s semiconductor innovation cycle should design and deploy “some lower hanging fruits first”.
“There are chips to be made in communication modules, power management, radio frequency and other applications. None of these require you to be at the cutting-edge of semiconductor development, which one can solve first before looking at the most advanced chips since the latter involves high amounts of capex (capital expenditure),” said Viswanathan, who’s also a member of the advisory board of the India Semiconductor Mission and vice-president of Intel’s architecture division.
“If we build a leading-edge product today, that will also have to compete with the absolute best in the global industry. I don’t have a definition for how an India-made cutting-edge semiconductor product can be better than what is already available globally, so India is not quite there yet in this regard. As we catch up, it is important to focus on specific areas of national security and commercial interest,” he added.
C-Dac, to be sure, has worked on such projects, including a navigation simulator for Indian submarines and Thejas64, a 180nm chip fabricated in India’s Semi-Conductor Laboratory for use in smart metering systems.
The AI chip, however, could potentially play a key role in public services if C-Dac can prove its performance at scale. Kompella said C-Dac’s track record of credibility is a key factor in this regard. “One of C-Dac’s Param supercomputers is still ranked within the top 250 of the world, and C-Dac also plays a key role in India’s design-linked incentives (DLI) scheme, which is being ramped up now,” he said. “It will now remain to be seen if C-Dac’s AI chip is proven in application. Once it is, the chip in question can potentially be used in strategic uses—even if not at scale.”
India, however, believes it is on the right track for a key strategic victory with its first AI chip. As Vaishnaw said last month, “The most important aspect of this project is to ensure India’s strategic autonomy in face of geopolitical relations. Some products need not be commercial, but they will still play crucial roles in India’s semiconductor journey. The AI chip is one such.”
- $200 million | Total budget allocated to C-DAC’s indigenous AI-chip project over five fiscal years
- 2nm | Target fabrication node for C-Dac’s AI chip, putting it at the cutting edge of semiconductor technology
- 2030 | Target deployment year for C-Dac’s AI chip in real-world applications and strategic use cases