AI fills India’s data centers as inference moves to production

NEW DELHI: An increase in artificial intelligence (AI) consumption is rapidly filling up capacity in Indian data centers, even as the share of model training remains minuscule compared to global hubs, industry executives said.

AI workloads fall into two categories, but their infrastructure requirements differ sharply. Inference needs ultra-low-latency bandwidth and proximity to end users and data sources, while training requires massive parallel compute and high-throughput interconnects. Globally, these resource-intensive training workloads are heavily driven by foundational labs such as OpenAI, Anthropic, Google DeepMind, and leading Chinese technology firms.