Key Takeaways:
AM Intelligence's binding order for 9,000 Nvidia Vera Rubin NVL72 rack-scale systems marks one of Asia's first large-scale deployments of Nvidia's newest AI computing platform, backed by an $8 billion plan to build 1 gigawatt of compute capacity on India's low-cost renewable energy advantage.
"Bringing the latest generation NVIDIA Vera Rubin to India positions AMI at the forefront of the emerging AI electron-to-token economy," Anil Chalamalasetty, chairman of AM Intelligence, said. The company's focus has evolved from transforming electrons into value to converting power infrastructure into frontier AI compute at scale, he added.
The systems are scheduled for delivery by Q1 2027 to an AI factory in Hyderabad, where the initial 30 MW phase will serve cloud-service providers, AI labs, and organizations building homegrown Indian AI models. The facility is engineered to deliver approximately 450 exaFLOPS of NVFP4 inference compute — enough to run trillion-parameter models and next-generation agentic AI applications. The Vera Rubin NVL72 architecture, which entered full production in May 2026, incorporates NVFP4 low-precision computing and next-generation HBM4 memory, and is expected to cut AI inference costs by up to ten times versus the prior Grace Blackwell generation, according to AMI.
The Hyderabad order is the first phase of a broader plan to bring 1 GW of compute-as-a-service capacity to market across India, the US, Finland, and Malaysia, with an initial 200 MW coming online in the near term. AM Group — founded by Chalamalasetty and Mahesh Kolli, who also built Greenko, India's clean-energy leader — is targeting 5 GW of data center capacity by 2030. Kolli said the group's core advantage is access to low-cost renewable power. "In the global token economy, energy prices are decisive," he said. "We are among the lowest-cost players in global AI compute infrastructure." Initial compute capacity at Hyderabad has already been pre-purchased by a US customer, and Kolli noted India's submarine cable infrastructure can deliver roughly 300 milliseconds of latency for US enterprises.
Beyond Hyderabad, AM Group has outlined a $25 billion, 1 GW green-powered AI and high-performance computing hub in Greater Noida, targeting 350 MW of capacity by 2028. The project pairs large-scale AI compute with renewable power generation, drawing on the founders' Greenko background. Together, the two sites represent the pillars of AM Group's AI strategy: rapid near-term hardware deployment in Hyderabad and longer-term green infrastructure at gigawatt scale in Greater Noida.
The order also carries strategic weight for India's sovereign AI ambitions. A domestic buildout of Nvidia Vera Rubin infrastructure, backed by billions in committed capital, gives Indian AI labs and enterprises a local alternative to running workloads through foreign-owned clouds — a shift that carries both cost and data-sovereignty implications. Japan is pursuing a similar path, planning to procure Nvidia's Rubin chips for a domestic foundation AI model targeting robotics, with its first data center expected operational by June 2028.
For Nvidia, the deal reinforces sustained demand for its newest platform despite ongoing supply constraints across the industry. Microsoft, Google, and Amazon have created multi-year backlogs for high-end AI compute clusters. AMI plans to fund expansion through a combination of debt and equity, with Kolli noting the business has a short cycle from capital expenditure to cash flow. The company's vertically integrated approach — pairing unique power solutions with data centers, hardware, and custom AI models — aims to deliver cost-competitive electron-to-token economics for hyperscalers, neo cloud providers, and sovereign AI initiatives.
The key risk is execution. The project is capital-intensive, relies heavily on a single hardware vendor, and the compute-as-a-service model's long-term profitability depends on securing large-scale clients at competitive pricing. If AMI brings its first 30 MW of Vera Rubin capacity online on schedule by Q1 2027, it would rank among the fastest deployments of this hardware anywhere at scale. Any slippage would test how much patience investors and government partners have for India's sovereign AI compute ambitions.
This article is for informational purposes only and does not constitute investment advice.
