The Faridabad-based company said the product is designed to address a core constraint in AI deployments — that GPU clusters are increasingly starved of data rather than compute power