AI accelerators (GPU)
AI accelerators are the primary compute chips for model training and inference.
Relationships
- Depends onHigh-bandwidth memory (HBM)— AI accelerators require HBM for high-throughput model workloads.
- Depends onCoWoS advanced packaging— High-end accelerator supply is gated by advanced package integration capacity.
- Competes withGoogle TPU— TPUs are accelerator alternatives available through GCP instances on IndiaAI.
- Depended on byIndiaAI Compute Portal— PIB July 2025 cites 34,381 provisioned GPUs; later PIB updates cite more than 38,000 onboarded.
- Controlled byNVIDIA Corporation— OECD cites estimates above 80 percent share for GPU chips used in AI.