PIB TrackerIndia has 40 supercomputers and 68 petaflops against a target of 50 and 123
The National Supercomputing Mission backgrounder publishes its progress against its own target: eleven years in, four-fifths of the machines and a little over half the capacity.
What happened
- The National Supercomputing Mission reports 40 supercomputers delivering 68 petaflops as of September 2026.
- The stated plan is 50 supercomputers with cumulative capacity exceeding 123 petaflops across academic and research institutions.
- NSM was launched in April 2015 with an outlay of about Rs 4,500 crore, steered by DST and MeitY and implemented by C-DAC Pune and IISc Bengaluru.
- The build approach has three concurrent phases: assembly, manufacturing, and design and manufacturing.
- For scale, the backgrounder notes the world’s fastest supercomputer at about 2.19 exaflops in 2026, and that India generates nearly 20 per cent of the world’s data.
For Prelims
- FLOPS: floating point operations per second. 1 teraflop = a trillion; 1 petaflop = a quadrillion; 1 exaflop = a quintillion. The ladder is thousand-fold at each step, so 68 PF and 2.19 EF are not close.
- National Supercomputing Mission: launched April 2015, outlay about Rs 4,500 crore, jointly steered by the Department of Science and Technology and MeitY - one of the few missions with two parent ministries.
- C-DAC: the Centre for Development of Advanced Computing, Pune, set up in 1988 after India was refused a Cray supercomputer. It built PARAM 8000 in 1991, which is why the series is read as an answer to technology denial.
- The PARAM series: PARAM 8000 (1991) at 1 gigaflop; PARAM Yuva at 54 teraflops; the NSM-era systems carry names such as PARAM Shivay, PARAM Pravega and PARAM Rudra.
- The build approach: three concurrent phases - assembly of imported components, manufacturing of subsystems in India, and design and manufacturing of indigenous systems. Localisation deepens phase by phase.
- What the applications are: computational biology, weather prediction and climate modelling, disaster management, big data analytics and astrophysics - the five domains the Mission names.
- Parallel processing: many processors working simultaneously on one calculation, connected by high-speed interconnects. Interconnect bandwidth, not processor count, is usually the binding constraint.
- ESTIC-2026: the second Emerging Science, Technology and Innovation Conclave, 27-29 October 2026 at Bharat Mandapam, organised by CSIR across twelve thematic areas.
For UPSC: A backgrounder that publishes a target and the distance still to it is unusually usable. Deploy it on indigenisation and technology denial, on the C-DAC story as the standard example of an embargo producing domestic capability, on research infrastructure as a public good, and on the AI compute question where India’s data share and its compute share point in opposite directions.
What it is NOT: No completion date is given for the remaining ten systems or the missing 55 petaflops, and no year-by-year series is published, so the pace cannot be read. Expenditure against the Rs 4,500 crore outlay is not stated. Utilisation is not reported anywhere - no job counts, no queue times, no user numbers - so whether the installed 68 petaflops are busy is unknown. Nothing is said about who makes the processors, which is the part of a supercomputer indigenisation has not reached. And no separate figure is given for AI-capable accelerated computing, which is where demand has shifted.
For Mains
Syllabus: GS3.13 · GS3.12 · Linkage L1
Anchor
The National Supercomputing Mission has 40 supercomputers delivering 68 petaflops as of September 2026, against its own plan of 50 systems with cumulative capacity exceeding 123 petaflops. The Mission was launched in April 2015 with an outlay of about Rs 4,500 crore, is steered jointly by the Department of Science and Technology and MeitY, and is implemented by C-DAC Pune and the Indian Institute of Science.
Substantiation (data)
Eighty per cent of the planned machines are in place but only about 55 per cent of the planned capacity, which means the systems still to come are the large ones. The backgrounder supplies its own scale check: the world’s fastest supercomputer runs at about 2.19 exaflops in 2026, roughly 32 times the whole Indian national capacity. Against that, India generates close to 20 per cent of the world’s data.
Position
The build approach is the substantive policy here, and it is staged rather than declared. Three concurrent phases - assembly, then manufacturing, then design and manufacturing - let capability deepen while machines keep being delivered, instead of waiting for indigenous design before installing anything. The indigenous output claimed so far is servers, software, cooling and networking: everything in a supercomputer except the processor.
Counterpoint
A mission in its eleventh year reporting four-fifths of its machines and a little over half its capacity has no completion date attached to the remainder, and no year-by-year series that would show whether the pace is rising or falling. Nor is utilisation reported. Installed capacity is an input; what matters is whether the queues are full, and no job count, queue time or user number appears anywhere.
Way forward
The useful disclosures are cheap ones: expenditure against outlay, a delivery schedule for the remaining systems, and utilisation data from the installed base. The harder question is composition - demand has moved towards accelerated computing for AI training, and a petaflop figure that does not separate general-purpose from AI-capable capacity no longer describes what researchers are queueing for.
Conclusion
The honest reading is a mission that has delivered steadily, built real subsystem capability, and remains two orders of magnitude from the frontier it is measured against. The PARAM line from one gigaflop in 1991 to 68 petaflops today is the achievement; the 2.19 exaflop comparison printed in the same release is the reason the target matters more than the milestone.
Deploys into: Indigenisation and technology denial · Research infrastructure as public good · AI compute capacity and data sovereignty · Mission-mode programmes and their targets
PIB Backgrounder · 2026-10-04 · PRID 2318908 · PIB source ↗