New Delhi: The National Informatics Centre (NIC) is setting up an artificial intelligence data centre—the government’s first—at Shastri Park in the capital, ThePrint has learnt. With a capacity of 1.1 AI exaflops, the facility is expected to be ready this year. Purchase order for the equipment has already been placed. The centre will house 248 Graphics Processing Units (GPUs) and hold government data, a senior government official aware of the matter said.
The centre is being built even as NIC prepares to launch two projects—one database for Parliament, and another for legal records. The parliamentary database will be accessible only to Members of Parliament, while the legal one may be thrown open to the public, the official said. Both databases are set to be launched by next month, according to the official.
An exaflop is a measure of computing speed. A flop is a single calculation, and an exaflop is a quintillion of them every second: a one followed by 18 zeroes. The ‘AI’ prefix is significant. Conventional supercomputers are rated on high-accuracy maths used for tasks, such as weather and climate modelling, whereas AI does not require that level of precision and relies on rougher, faster calculations, allowing the same hardware to crunch far more operations a second.
An AI exaflop is thus an easier benchmark to reach than a supercomputer exaflop, and the two are not directly comparable. The 1.1 figure reflects what the centre can do for AI work, not scientific computation.
Shastri Park already hosts an NIC data centre, but of a different kind. The existing National Data Centre at the Delhi Metro Rail Corporation (MRC) Information Technology Park runs 462 racks, and hosts websites and applications for Central government departments and several states. Built to store data and keep government sites running rather than to perform AI tasks, it carries no exaflops rating; its capacity is measured in racks. The 1.1 AI exaflop figure applies only to the new facility.
The GPUs are needed largely because of the sheer scale of the government’s paperwork.
NIC runs 800 instances across the country, and its offices hold vast volumes of correspondence files. Before these can be read or queried, they must be run through optical character recognition (OCR), which converts scanned documents into searchable text. That conversion, and the question-answering built on top of it, demands the kind of GPU capacity the new centre is meant to supply.
Both the parliamentary and legal projects follow the same template—existing paper and scanned records are digitised, put through OCR and made searchable, so that a user can pose a question and receive answers drawn from the documents themselves.
The development comes as the Centre ramps up its own AI compute base. Under the IndiaAI Mission, backed by an outlay of Rs 10,372 crore, more than 38,000 GPUs have been onboarded through a common compute portal and offered to startups, researchers and academic institutions at subsidised rates—well past the initial target of 10,000.
At the India AI Impact Summit in February, the government announced it would add 20,000 more GPUs, taking public compute capacity beyond 58,000 units, priced at Rs 65 per GPU-hour. The stated aim is to cut dependence on foreign infrastructure and build sovereign AI capacity.
An NIC data centre dedicated to AI workloads fits that push, giving the government in-house capacity to process its own records instead of routing sensitive material through external cloud providers—in keeping with data-sovereignty requirements.
(Edited by Nardeep Singh Dahiya)
Also Read: India’s AI Mission is flying blind without technocrats