Home/The Atlas/Pillars/AI Infrastructure
Pillar map

AI Infrastructure

India is building data centres at scale, yet the compute, memory and silicon inside them are almost entirely imported.

65 players104 relationships30 chokepoints63% India-based
The stack

Where the value sits

Each layer and the chokepoints inside it, coloured by where India stands.

Indian capability Foreign-controlled Global / neutral
Who holds the veto

Chokepoints

Who controls each layer, and how much — ranked by concentration.

ControllerLayer / inputShareNote
IndiaAI Compute Portaldepends on AI accelerators (GPU)34381 gpu countPIB July 2025 cites 34,381 provisioned GPUs; later PIB updates cite more than 38,000 onboarded.
NVIDIA Corporationcontrols AI accelerators (GPU)80%OECD cites estimates above 80 percent share for GPU chips used in AI.
SK Hynixcontrols High-bandwidth memory (HBM)58%Counterpoint Q1 2026 HBM revenue share.
Samsung Electronicscontrols High-bandwidth memory (HBM)21%Counterpoint Q1 2026 HBM revenue share.
Micron Technologycontrols High-bandwidth memory (HBM)21%Counterpoint Q1 2026 HBM revenue share.
AI accelerators (GPU)depends on High-bandwidth memory (HBM)AI accelerators require HBM for high-throughput model workloads.
AI accelerators (GPU)depends on CoWoS advanced packagingHigh-end accelerator supply is gated by advanced package integration capacity.
AI cluster networking fabricdepends on Optical transceiversAI clusters require high-speed switching and optical interconnect.
TSMCcontrols CoWoS advanced packagingTSMC identifies CoWoS and other advanced packaging technologies as part of its technology platform.
The ecosystem

Players by role

All players →
The moat

Where this pillar connects

Cross-pillar links only Techadyant can show — one graph, not five silos.