AM Intelligence (AMI) is moving to build one of Asia’s early frontier AI computing clusters with a binding order for 9,000 Nvidia Rubin GPUs, which it plans to deploy in a 30-MW AI factory in Hyderabad in the first quarter of 2027. The company said the facility would be among Asia’s first large-scale AI compute clusters based on Nvidia’s next-generation Vera Rubin architecture.
The move comes as demand for high-end AI computing infrastructure is running ahead of supply, with hyperscalers and artificial intelligence companies competing for advanced clusters to train and run increasingly large models. AMI’s initial Hyderabad capacity has already been purchased by a US customer, Mahesh Kolli, founder and president of Greenko Group, told Bloomberg, although he declined to identify the customer citing a non-disclosure agreement.
AMI’s Hyderabad facility will be the first tranche of a planned 1 GW compute-as-a-service platform that the company is developing across India, the US, Finland and Malaysia. It plans to bring 200 MW of capacity to market in the near term, involving more than $8 billion in capital expenditure. The company is targeting hyperscalers, AI labs, neo-cloud providers, sovereign AI initiatives and enterprises as customers for the computing capacity.
The company is positioning access to low-cost renewable power as a key advantage in competing for AI workloads, where electricity costs form a significant component of the economics of serving AI tokens. AMI is backed by Greenko’s renewable generation and energy-storage assets, allowing it to combine power infrastructure with high-density computing.
AMI is developing 5 GW of powered AI data centres across India, the US and Europe, with the facilities designed around high-density computing and liquid cooling to accommodate the substantial power and heat generated by advanced AI chips. The company plans to fund the buildout through a combination of debt and equity and is targeting 5 GW of data-centre capacity by 2030.
The Hyderabad cluster will use Nvidia Vera Rubin NVL72 rack-scale systems, high-throughput RDMA over converged ethernet networking and advanced storage infrastructure. AMI said the system is being designed to run trillion-parameter models and next-generation agentic AI applications. Nvidia’s Rubin architecture includes NVFP4, a low-precision computing format, and next-generation HBM4 memory, with Nvidia claiming significantly lower inference costs compared with its previous Grace Blackwell generation.
AMI expects the Hyderabad facility to deliver about 450 exaFLOPS of NVFP4 inference compute. The company said its broader strategy is to integrate renewable energy, data-centre infrastructure and next-generation AI hardware to offer compute capacity to global customers while also widening access to advanced AI infrastructure for Indian developers and sovereign AI initiatives.