India’s Rising Supercomputing Ecosystem

1. At a Glance

• National Supercomputing Mission (NSM): India's main programme for building high-performance computing (HPC). It was launched in 2015 and has two aims: install supercomputers across research institutions and build the hardware and software in India. [1]
• As of September 2026, there are 40 supercomputers with 68 PF of combined capacity under NSM. The target is 50 systems with more than 123 PF. [1]
• Why it matters for UPSC: it links to Aatmanirbhar Bharat (self-reliance in technology), AI computing, weather and climate forecasting, disaster management and strategic autonomy. It sits in GS-III (S&T) and is a frequent source of Prelims factual questions. [1]

2. Why in the News

• PIB explainer (4 Oct 2026) on India's supercomputing journey: NSM now has 40 systems and 68 PF, and has developed indigenous servers, software, cooling and networking. [1]
• ESTIC-2026 (the second Emerging Science, Technology and Innovation Conclave) is due on 27–29 Oct 2026 at Bharat Mandapam, New Delhi. It is organised by CSIR, covers 12 thematic areas and includes supercomputing. [1]
• PARAM Rudra supercomputer inaugurated at IIT Bombay, extending the indigenous Rudra-based systems. [6]

3. Background & Evolution

• 1991: PARAM 8000, built by C-DAC, marked India's entry into indigenous supercomputing. [1]
• Earlier weather HPC under MoES: the Pratyush and Mihir systems. [9]
• April 2015: NSM launched with an outlay of about ₹4,500 crore. [1]
• ISC 2023: AIRAWAT, the AI compute platform, ranked 75th in the TOP500 global list. [2]
• PARAM Ganga, a petascale system, was set up at IIT Roorkee under NSM. [10]
• 2024: the PM dedicated three PARAM Rudra supercomputers in Pune, Delhi and Kolkata (the Kolkata system is at S.N. Bose National Centre for Basic Sciences). [2][7]
• 2024: the PM inaugurated the MoES weather and climate HPC systems Arka and Arunika (₹850 crore). These took MoES capacity from 6.8 PF to 22 PF. [3]
• NSM capacity over time:
• 37 systems / 40 PF [5]
• 38 systems / 44 PF (January 2026) [4]
• 40 systems / 68 PF (September 2026) [1]

4. Core Static Facts

• Units: FLOPS = floating point operations per second. 1 TF = 10¹² FLOPS; 1 PF = 10¹⁵ FLOPS (one quadrillion); 1 EF = 10¹⁸ FLOPS (one quintillion). [1]
• World's fastest supercomputer (2026): about 2.19 exaflops. [1]
• Parallel processing: many CPUs or nodes work on one problem at the same time, which is why supercomputers are also called "parallel computers". [1]
• No dedicated Act or constitutional provision governs NSM; it is an executive mission.

5. Multi-Dimensional Analysis

Scientific / Technological

• NSM covers the whole chain: design, manufacturing, system software and applications. India now builds its own servers (Rudra), interconnect (Trinetra), cooling and software. [1][4]
• Weather HPC cut model run time from about 12 hours to 3–6 hours, which made high-resolution forecasts usable in real time. [8]

Strategic / Geopolitical

• Indigenous capability reduces dependence on imported technology. Supercomputers are critical infrastructure for strategic applications. [1]
• AIRAWAT's TOP500 ranking puts India among AI-supercomputing nations. [2]
• The gap is still large: India's 68 PF in total is a small fraction of the world's fastest single system (about 2.19 EF). [1]

Economic

• India generates nearly 20% of the world's data, which drives demand for AI and HPC infrastructure. [1]
• Domestic manufacturing of servers and subsystems supports the electronics value chain.

Disaster Management / Environmental

• Arka and Arunika improve the accuracy and lead time of forecasts for cyclones, heavy rain, heatwaves and droughts. [3]
• HPC supports river-flood mapping and climate modelling. [1]

Social

• Uses include crop-yield prediction, tracking disease outbreaks and drug discovery. [1]
• NSM systems support more than 10,000 researchers. [5]

Administrative

• Two ministries steer the mission (DST and MeitY) and two agencies implement it (C-DAC and IISc). The MoES weather HPC is a separate track outside NSM. [1][3]
• Physical targets have fallen behind schedule: 50 systems were planned, 40 exist so far. [1]

6. Recent Developments (last 12–18 months)

• Late 2025: NSM reported 37 systems and 40 PF, supporting more than 10,000 researchers. [5]
• January 2026: 38 systems and 44 PF deployed. [4]
• 2026: PARAM Rudra inaugurated at IIT Bombay. [6]
• September 2026: 40 systems and 68 PF. [1]
• 4 October 2026: PIB explainer on India's supercomputing ecosystem released. [1]
• 27–29 October 2026: ESTIC-2026 at Bharat Mandapam, organised by CSIR. [1]

7. Prelims Hooks

• NSM was launched in April 2015 with an outlay of about ₹4,500 crore. [1]
• NSM is steered jointly by DST and MeitY, not by MoES. [1]
• It is implemented by C-DAC Pune and IISc Bengaluru. [1]
• PARAM 8000 (C-DAC, 1991) was India's first indigenous supercomputer. [1]
• NSM target: 50 supercomputers with more than 123 PF. [1]
• Status in September 2026: 40 systems, 68 PF. [1]
• "Rudra" is India's first indigenous HPC-class server. [2]
• "Trinetra" is C-DAC's indigenous high-speed interconnect. Trinetra-A runs at 100 Gbps and Trinetra-B at 200 Gbps. [4][11]
• AIRAWAT is an AI supercomputer that ranked 75th in TOP500 at ISC 2023, Germany. [2]
• Arka is at IITM Pune (11.77 PF). Arunika is at NCMRWF Noida (8.24 PF). Both are MoES systems; the project cost ₹850 crore. [3]
• Arka and Arunika raised MoES HPC capacity from 6.8 PF to 22 PF. [3]
• PARAM Ganga is at IIT Roorkee. [10]
• 1 petaFLOPS = 10¹⁵ calculations per second. [1]
• India generates nearly 20% of the world's data. [1]
• ESTIC-2026 is organised by CSIR. Its theme is "Creative Minds, Catalyzing Science, Empowering Technologies." [1]

8. How Much of the 'Indian Supercomputer' Is Really Indian

• The machines are built in India, but the chip at their centre is not yet Indian
• C-DAC designed the Rudra server and the Trinetra interconnect (the network that links the server parts). All NSM systems are made in India and run on Indian-built software [1][4].
• The processor (the main chip that does the calculating) is a separate matter. India's own HPC processor, AUM, is still being designed [12].

• So until AUM is ready, the main chip inside each 'indigenous' server comes from outside India.

• AUM itself depends on foreign partners

• AUM is based on the Arm Neoverse V2 platform. Arm is a foreign company that licenses its chip designs to others [12].
• C-DAC is designing it with a consortium (a group of partner firms): MosChip (India) and Socionext (Japan) [12].

• So even the 'Indian chip' sits on a foreign design base, built with a foreign partner.

• Why this matters for the exam

• NSM's own three phases are assembly → manufacturing → design [1]. Self-reliance therefore comes in layers.
• India has reached the system layer (server, network, software) but not yet the chip layer. Say this clearly when a question asks you to 'assess self-reliance'.

9. Why the Mission Is Running Past Its Deadline

• The deadline has already passed, but the target has not been met
• NSM began in April 2015. It was extended to 31 December 2025 [11].
• In September 2026 it still stood at 40 of 50 systems and 68 of 123+ PF [1].

• That is about 55% of the computing-power target, more than eleven years after launch.

• Building at home is slower than buying, and that delay was built into the plan

• The 'build approach' moves step by step: first assemble imported parts, then manufacture them in India, then design them in India [1].
• Each step needs the skills from the step before it. A server or network card must be designed, tested and made before it can go into a system.

• Buying ready-made machines would have reached the numbers sooner, but would not have built the Rudra–Trinetra capability [2][4].

• Nobody has said when the remaining 10 systems will arrive

• None of the sources here gives a new end date after December 2025 [1][11].
• With no fixed date, it is hard to judge whether the mission is late or on track.

10. What the 68 PF Figure Hides

• Most of the systems are small
• 68 PF spread across 40 systems is about 1.7 PF per system on average [1].

• The world's fastest single machine runs at about 2.19 EF (2,190 PF) [1]. That is more than 1,000 times India's average NSM system, and about 30 times all NSM capacity put together.

• Capacity comes in sudden jumps, not steadily

• From January to September 2026, capacity rose from 44 PF to 68 PF while only two systems were added [4][1].
• So in eight months, 24 PF was added. That is more than half of the 44 PF built in the previous ten years.

• This means a few large systems decide the headline number. Trinetra-B, for example, was built for a single 20 PF PARAM Rudra at C-DAC Bengaluru [11].

• How to read it in an answer

• Do not read '40 systems' as 40 big machines. Most are small machines at institutes, built for training and research access. Only a handful are large.
• Also, AIRAWAT's 'AI petaflops' cannot be added to the 68 PF, because they use a different kind of calculation (mixed precision) [2].

11. Machines Running Almost Full: Success and Warning Sign

• High utilisation shows real demand
• Utilisation rate means the share of time a machine is actually busy with work.
• NSM systems run at over 85% utilisation, and many are above 95% [5].

• They serve 10,000+ researchers, including 1,700+ PhD scholars from 200+ institutions [5]. These are not idle prestige machines.

• The same number also signals a bottleneck

• A machine that is busy 95% of the time has almost no free time left. New jobs have to wait in a queue.
• The scholars who suffer most are those at smaller colleges with no computer of their own. They depend fully on shared NSM time.
• So the high usage figure is also an argument that the remaining 55 PF of the target is overdue [1].

12. The Case for Simply Buying Computers — and Where It Falls Short

• The strongest argument against the 'build' path
• Global chip and server technology moves fast. A home-built system can fall behind before it is even finished.
• The build path has cost time. NSM reached only about half its target in more than a decade [1][11].

• Researchers need computing power now. Buying ready-made systems, or renting cloud computing, would have been quicker.

• What the argument gets right

• Speed has been lost. India's total capacity is still a small fraction of a single top global machine [1].

• Full independence is not realistic either. Even AUM uses Arm's foreign design and a Japanese partner [12].

• Where it falls short

• Supercomputers run weather, disaster and strategic work. That makes them critical infrastructure (systems the country cannot afford to lose) [1]. A bought machine depends on the seller for spare parts, upgrades and permission to use it.
• The build path has already given India its own server, network and software [2][4]. A purchase does not leave that capability behind.
• Balanced answer: a mixed path. Buy or rent for urgent capacity, and keep building the layers India can actually master.

13. What DST, MeitY and C-DAC Should Do Next

• DST and MeitY should announce a new deadline and a plan for the rest
• The last stated deadline was 31 December 2025 [11]. 10 systems and about 55 PF are still missing [1].

• Publishing a new date with yearly targets would let Parliament and the public track progress.

• C-DAC should set a date to move from AUM design to actual servers

• AUM is the missing chip layer [12]. Until Rudra servers carry AUM, the 'indigenous' claim stops at the system level.

• This links to the India Semiconductor Mission, which deals with making chips in India.

• DST should fund exascale research now, not later

• NSM already invites proposals for 'R&D-Exascale' projects [13]. Exascale means a machine doing 10¹⁸ calculations per second.

• The world's top machine is already beyond 2 EF [1]. Research on next-generation design takes years, so it has to start well before India can build such a machine.

• Add capacity where the queues are longest

• Systems busy more than 95% of the time [5] are the clearest sign of where the next machines are needed.

14. Anchors for Answers

• Data: 40 systems / 68 PF against a target of 50 systems / 123+ PF, which is about 55% of the computing-power goal (Sept 2026) [1]
• Data: NSM utilisation above 85%, many systems above 95%; 10,000+ researchers, 1,700+ PhD scholars, 200+ institutions [5]
• Data: Mission extended to 31 Dec 2025 [11]
• Scheme: AUM, the indigenous HPC processor C-DAC is designing on Arm Neoverse V2 with MosChip and Socionext. It is the missing chip layer of self-reliance [12]
• Scheme: NSM 'R&D-Exascale' call for proposals (DST), the base for India's next-generation machines [13]

15. Mains Relevance

• GS-III: S&T developments and their applications in everyday life; indigenisation of technology; awareness of IT and computers; disaster management.
• GS-II: government policies and interventions (mission-mode governance).
• Possible question stems: 1. Assess the progress of the National Supercomputing Mission in achieving self-reliance in high-performance computing. What challenges remain? 2. High-performance computing is critical infrastructure for India's AI ambitions and climate resilience. Discuss. 3. How have indigenous HPC systems improved India's weather forecasting and disaster preparedness? Illustrate with examples.

16. Related Topics to Study Next

• IndiaAI Mission (GPU compute): AI compute builds on AIRAWAT and the HPC base.
• National Quantum Mission: the next frontier in computing, also under DST.
• Semiconductor Mission (ISM): hardware self-reliance underpins indigenous servers.
• Bharat Forecast System / Mission Mausam: the main user of the MoES HPC systems.
• National Knowledge Network (NKN): connects NSM and AIRAWAT systems to users. [2]
• C-DAC and the PARAM series history: frequently asked in Prelims.
• Exascale computing and the TOP500 list: global benchmarks.

17. NCERT Concepts to Revise

• India Semiconductor Mission (Geography): The note says India's own AUM chip is still missing; chip-making under this mission decides true hardware self-reliance.
• Self-reliance (Economy): NSM's build approach aims at Aatmanirbhar Bharat; the note asks you to assess self-reliance layer by layer.
• Bharat Forecast System (Geography): Weather supercomputers like Arka and Arunika run this indigenous high-resolution forecast model, a key application of HPC.
• Mission Mausam (Geography): This MoES mission is the main user of the weather HPC systems and is a separate track from NSM.
• Weather forecasting (Geography): Supercomputers run weather models, cutting run time from 12 hours to 3–6 hours and improving disaster warnings.
• Infrastructure (Economy): The note calls supercomputers critical infrastructure for weather, disaster and strategic work, which is why buying instead of building is risky.

18. Common Errors / Trap Areas

• Ministry confusion: NSM belongs to DST + MeitY. Arka and Arunika belong to MoES and are not NSM systems. [1][3]
• Location mix-ups: Arka is at IITM Pune and Arunika at NCMRWF Noida. Do not swap them. [3]
• Capacity figures change with each release (40 PF → 44 PF → 68 PF). Always check the "as of" date. [1][4][5]
• Rudra vs Trinetra: Rudra is the server; Trinetra is the interconnect (network). [4]
• Units: peta = 10¹⁵ and exa = 10¹⁸. AIRAWAT's "AI petaflops" (mixed precision) are not comparable with standard PF. [1][2]