India’s transition from AI policymaking to physical infrastructure deployment gained significant momentum today as UrsaCompute, the AI compute arm of Ursa Group, committed $300 million to build a national hub-and-spoke AI compute infrastructure, with Phase 1 anchored at ST Telemedia Global Data Centres in Greater Noida.
UrsaCompute received provisional empanelment under the Government of India’s Rs 10,372 crore IndiaAI Mission. The central government programme aims to establish domestic sovereign AI capabilities and reduce reliance on foreign hyperscaler cloud ecosystems for critical AI workloads.
UrsaCompute has committed to deploying 2,888 NVIDIA B200 and B300 GPUs. Based on NVIDIA’s latest Blackwell architecture, these chips represent the current frontier of AI accelerators. Securing these units is notable given the tight constraints on global Blackwell supply chains.
The infrastructure is designed to address the domestic deficit in high-density compute power required to train large foundation models, specifically those optimized for multilingual reasoning and Indic languages. Industry analysts have long warned that a lack of domestic infrastructure forces local developers to rely on overseas cloud regions, compromising data sovereignty.
“The compute layer is where AI sovereignty either happens or does not,” said Satish Abburi. “India’s AI models need to be trained on India’s infrastructure, governed under Indian law.”
In tandem with the physical hardware, the company is launching its Neo Cloud platform, tailored for GPU-as-a-Service (GPUaaS) and automated workload orchestration. The model reflects a global shift toward specialized AI-native cloud providers that bypass traditional enterprise cloud architecture to focus strictly on high-density compute.
While securing hardware is a critical milestone, industry experts emphasize that the operational execution of frontier-grade GPU clusters presents steep technical challenges in networking, liquid cooling, and distributed systems management. To manage this infrastructure, UrsaCompute has assembled a leadership team of former engineers from Oracle, NVIDIA, and Digital Realty, supported by ex-IndiaAI Mission leadership.
