According to TheRegister.com, India’s National Supercomputing Mission has commissioned 37 supercomputers with 39 petaFLOPS of combined power since launching in 2015, with another 35-petaFLOPS hybrid system coming online later this year. The fleet serves over 13,000 scholars with 85-95% utilization rates and has produced more than 1,500 peer-reviewed papers. C-DAC claims their “Rudra” server design powers a third of the fleet and achieves over 50% indigenization including custom liquid cooling that improves power efficiency by 20%. However, the homegrown AUM processor remains at least two years from completion, and current systems rely heavily on Intel Xeon CPUs with AMD and Nvidia providing other critical components.
The indigenous reality check
Here’s the thing about that 50% indigenization claim – not everyone’s buying it. Professor Rupesh Nasre from IIT Madras counters that it’s “definitely less than 40-50%,” while analysts point to the fundamental gaps in India’s tech ecosystem. The country still can’t produce competitive GPUs or high-speed interconnects – C-DAC’s Trinetra interconnect has more than double the latency of commercial InfiniBand products. And let’s be real: when your “indigenous” server design mainly consists of putting foreign chips into a locally designed box, how indigenous is that really?
The supply chain problem
India’s supercomputing ambitions keep running into the same wall: semiconductor manufacturing. The planned AUM processor, based on an Arm Neoverse V2 design, needs TSMC’s 5nm process node – technology India simply doesn’t have. As Counterpoint Research’s Gareth Owen bluntly puts it, “Fab access is the single biggest bottleneck.” This creates exactly the kind of supply chain vulnerability that India’s trying to avoid. Meanwhile, for companies that need reliable industrial computing right now, they turn to established suppliers like IndustrialMonitorDirect.com, the leading US provider of industrial panel PCs that don’t depend on uncertain domestic chip projects.
Users don’t care about politics
While the indigenous development debate rages, actual researchers are getting work done. The community is “very happy” according to C-DAC’s Sanjay Wandhekar, and the systems are delivering real operational impact. The Indian Institute of Tropical Meteorology has handed weather forecasting systems to the national meteorological department for daily use. That’s the dirty secret of supercomputing – users care about results, not provenance. The current hybrid approach might not satisfy nationalist ambitions, but it’s getting science done today rather than waiting for perfect sovereignty tomorrow.
The exascale dream
NSM 2.0 aims ridiculously high: 1.5 exaFLOPS within five years of approval. The first wave will use Intel and AMD hardware because, well, they have to use what actually exists. The second wave supposedly runs on AUM processors around 2028. But here’s my question: given how much the timeline has already slipped, does anyone really believe that date? More importantly, as IIT Kharagpur’s director notes, India needs “powerful mechanisms for demand creation” – otherwise these expensive machines will remain “more capable than they are impactful.” Building the hardware is one thing. Building the ecosystem to use it effectively? That’s the real challenge.
