# hyperscaler capex — X 热门讨论 (2026-09-18 11:11 UTC)
## @mezza9_Equity (Mezzanine Equity Pvt Ltd) · 09-18 05:30 · ♥31 ↻5 💬3 Was going through the India AI/DC value chain over the last few days. Lot more companies involved than what we normally discuss.
US AI story started with chips. India looks different for now. Here I feel much bigger opportunity is around building the infra required to actually deploy those chips.
Netweb is probably the closest listed name on AI hardware side. GPU/HPC servers, storage, AI systems and now liquid cooled racks.
E2E is more on GPU cloud/compute rental side.
Then comes networking. This part I had underestimated earlier.
Sterlite Tech is important here. AI DC needs huge fibre density, both inside DC and DC to DC. STL is building products specifically for AI/data center connectivity and recently got the large multi year hyperscaler optical connectivity order.
Tejas also fits here with optical transport/DCI. HFCL on fibre/connectivity side.
After this comes actual DC capacity.
Anant Raj, Techno Electric, Sify. Then Nxtra under Bharti, AdaniConneX under Adani Ent and Reliance/Jio.
Black Box is interesting in a different way, more on designing/integrating the actual DC infra rather than owning DC capacity.
But bigger opportunity IMO could be below all this.
Every MW needs transformer, switchgear, substation, cables, UPS, backup power, cooling etc.
Hitachi Energy, GE Vernova T&D, ABB, Siemens Energy, CG Power, Schneider Infra, L&T, KEC.
Transformer side also TRIL, Voltamp etc.
Cables again a big requirement. Polycab, KEI, Apar, RR Kabel, Diamond Power, Finolex.
Backup power - Cummins, Kirloskar Oil Engines.
Cooling is another segment worth studying separately. AI rack densities are going up very fast, so normal DC cooling won't be enough everywhere.
Blue Star, Voltas, Thermax, Kirloskar Pneumatic on cooling/HVAC side.
Liquid cooling opens another small value chain altogether. Pipes, hoses, pumps, coolant distribution etc. Venus Pipes and Aeroflex are two names I came across here.
Then ultimately power.
No matter how many GPUs you import, without available MW nothing goes live.
NTPC, Tata Power, JSW Energy, Adani Power on generation.
Power Grid, Adani Energy Solutions and the entire T&D ecosystem for getting that power to the DC.
So basically India AI capex is spreading into sectors which otherwise don't even look like "AI companies".
My rough India AI/DC value chain list:
Power generation NTPC, Tata Power, JSW Energy, Adani Power Grid/T&D Power Grid, Adani Energy Solutions, Hitachi Energy, GE Vernova T&D, KEC
Electrical/transformers ABB, Siemens Energy, CG Power, Schneider Infra, TRIL, Voltamp EPC L&T, KEC, Techno Electric, NCC
Cables Polycab, KEI, Apar, RR Kabel, Diamond Power, Finolex Backup power Cummins India, Kirloskar Oil Engines
Cooling Blue Star, Voltas, Thermax, Kirloskar Pneumatic Liquid cooling related Venus Pipes, Aeroflex
Fibre/networking Sterlite Technologies, Tejas Networks, HFCL DC integration Black Box, Aurionpro
AI servers/hardware Netweb GPU cloud E2E Networks
DC operators/developers Anant Raj, Techno Electric, Sify Large DC platforms Bharti Airtel/Nxtra, Adani Ent/AdaniConneX, Reliance/Jio
Connectivity Tata Communications, Bharti Airtel, RailTel There will definitely be more names as we go deeper into each layer.
Main thing I am trying to understand now is simple - if India adds several GW of AI/DC capacity, where does every Rs 100 of capex actually go? That breakdown could be much more interesting than only tracking who is announcing how many MW.
#AI https://x.com/mezza9_Equity/status/2100819831314669688