Sweat drips in a Pune data center as engineers tweak the latest Rudra server rack, fans whirring like a Mumbai monsoon.
India’s National Supercomputing Mission — launched back in 2015 — has cranked out 37 supercomputers, a fleet packing 39 petaFLOPS, with another 35-petaFLOPS hybrid slated for later this year. Impressive on paper. But here’s the kicker: while they’ve met basic computing needs for government and researchers, the big dream of semiconductor leadership? Still a mirage.
Look, I’ve covered tech sovereignty pushes from Japan in the ’90s to Europe’s squandered billions on homegrown chips. India’s NSM feels eerily similar — lots of assembly, scant invention at the silicon level. C-DAC, the outfit running the show, boasts over 50% indigenization in server nodes, interconnects, system software. Sounds good. Except critics — sharp ones from IIT Madras and Bombay — peg it closer to 40%, max.
“Our indigenization efforts have reached more than 50 percent, counting server nodes, interconnects, and system software,” C-DAC’s Sanjay Wandhekar told The Register.
“Our indigenization efforts have reached more than 50 percent, counting server nodes, interconnects, and system software.”
Nice quote. But Rudra servers? They’re the star, powering a third of the fleet. Latest versions slot in Nvidia and AMD GPUs, Intel and AMD CPUs. No big deal — except India’s own AUM processor lingers two years out, stuck in development hell.
Rudra: Homegrown Hero or Vendor Sidekick?
Rudra’s no slouch. C-DAC cooked it up under NSM, now every new install rocks direct liquid cooling — 10% power savings per node, site PUE down to 1.20-1.25. That’s 20% better than air-cooled junk from globals. Costs match OEMs, but in-house assembly shaves 15% off. Smart.
AMD’s Vinay Sinha brags about co-designing Rudra with EPYC CPUs and Instinct accelerators. Intel Xeons rule the roost now. Red Hat’s RHEL underpins some, but C-DAC swaps in custom open-source stacks — schedulers, monitors — dodging vendor lock-in. CentOS on others. It’s a hybrid mashup, not pure desi magic.
Trinetra interconnect? C-DAC’s pride, in three machines. Latency? Double InfiniBand’s. Ouch. No home GPU yet. IISc’s chasing IP cores, but that’s research, not racks.
And utilization? 85-95%. 13,000 scholars plugged in, 10 million workloads, 1,500+ papers. Needs met. Ambitions? Laughable.
## Is 50% Indigenous Just PR Spin?
Professor Rupesh Nasre at IIT Madras isn’t buying it.
“Definitely less than 40-50 percent,” he counters. “Reaching 50 percent would itself be a big achievement.”
IIT Bombay’s Madhav Desai nails the gap: no ecosystem to harden prototypes into products. Gartner analyst Sushovan Mukhopadhyay flags export controls, supply volatility. IDC’s Sharath Srinivasan: market scale yes, tech core no. Counterpoint’s Gareth Owen: fabs are the killer — AUM craves TSMC 5nm, India has zilch.
My hot take, absent from the original chatter: this mirrors India’s 1990s software outsourcing glory. We wrote code for the world, got rich — but owned zero IP. Supercomputing’s the same trap. C-DAC assembles, globals supply brains. Who profits? AMD, Intel, Nvidia cash in on sales. C-DAC gets prestige, government photo-ops. Real money — chip design royalties, fab margins — stays abroad.
C-DAC dreams exascale: chips, boards, photonics. Ambitious. But without nodes below 7nm domestically, it’s fantasy. India’s AI push wants 80 exaFLOPS. Plans for 10,000-GPU sovereign cluster. Recent flips: three homebrew machines online.
Short-term win. Long-term? Betting against full sovereignty by 2030, sans Tata or Micron fabs scaling fast.
## Why Can’t India Ditch Foreign Chips Yet?
Ecosystem void. Prototypes galore — AUM, Trinetra — but scaling? Fabs missing. TSMC won’t hand 5nm keys. Geopolitics bites: US export curbs could kneecap overnight.
Wandhekar claims Rudra pricing edges globals by 15%. Procurement smarts. Liquid cooling edge. But core IP? Rented.
Bold call: if AUM ships viable in two years — big if — it’ll run Arm cores, not x86 killers. GPUs? Five years off. Exascale by 2030? Only if China tensions force US alliances, flooding India with tech transfers.
Skeptical vet view: NSM’s a Band-Aid. Buys time, computes science. But ambitions scream ‘semiconductor power’ while pockets line foreigners’. Wake-up: invest fabs first, or stay assembler forever.
India’s churning supers — Param Siddhi, Michi — aiding weather models, drug discovery. Solid. 39 petaFLOPS ain’t sneeze. But leader? Nope. China laps at zetaFLOPS.
The Cash Question: Who’s Really Winning?
Follow money. C-DAC: contracts, jobs. Vendors: volume sales minus discounts. Scholars: free cycles. Taxpayers: Rs 4,700 crore spent wisely?
Unique angle — remember Japan’s Fujitsu saga? Poured billions into vector chips, lost to US GPUs. India risks same: Rudra relevant till AI shifts to inference clouds.
🧬 Related Insights
- Read more: Gemma 4’s Edge Push: NVIDIA’s Bid to Own Your Local AI
- Read more: Intel’s Arc Pro GPUs Claw Back Ground in MLPerf v6.0 — Still Nvidia’s Shadow?
Frequently Asked Questions
What is India’s National Supercomputing Mission?
Launched 2015, NSM built 37 machines for research compute, pushing local tech like Rudra servers — now at 39 petaFLOPS total.
Will India’s AUM processor power exascale supers?
Maybe in two years, but needs foreign fabs like TSMC 5nm; skeptics doubt full sovereignty without home nodes.
How indigenous are NSM supercomputers?
C-DAC says 50% (nodes, software); critics call BS, peg under 40% — heavy on Intel/AMD/Nvidia guts.
