The Story
AM Intelligence (AMI), the AI infrastructure company backed by the founders of Greenko, has placed a binding order for around 9,000 NVIDIA Rubin GPUs, deployed as Vera Rubin NVL72 rack-scale systems, for its first AI factory in Hyderabad. The order was announced on 25 August 2026.
The systems are scheduled for delivery in the first quarter of 2027 and will provide 30 MW of capacity at the Hyderabad facility, which the company said would be one of the first frontier AI compute clusters in Asia based on the Vera Rubin platform. The factory is being engineered to deliver about 450 exaFLOPS of NVFP4 inference compute, using liquid cooling and high-throughput networking to run large models, including trillion-parameter and agentic AI systems.
The order is a capital commitment rather than an external funding round. AMI said the Hyderabad factory is the first tranche of a planned 1 GW of compute-as-a-service capacity across India, the US, Finland and Malaysia, with more than $8 billion in stated capital expenditure and an initial 200 MW expected in the near term. The company said its customers would include cloud-service providers, frontier AI labs, enterprises, sovereign AI initiatives and organisations building homegrown Indian AI models.
AMI is the AI infrastructure arm of AM Group, the venture controlled by Greenko co-founders Anil Chalamalasetty and Mahesh Kolli. Chalamalasetty, who chairs AMI, framed the strategy as an extension of the founders' energy business, describing an "electron-to-token" opportunity that converts power infrastructure into AI compute. AM Group, founded in 2023, holds around a quarter of Greenko and has a large pipeline of renewable generation and long-duration energy storage under construction.
The Hyderabad order sits within a wider set of announced ambitions. AMI has said it is developing about 5 GW of powered AI data centre capacity across India, the US and Europe, and AM Group has separately outlined a $25 billion, 1 GW green-powered AI and high-performance computing hub planned for Greater Noida. These are stated plans, and the figures have not been independently verified.
Why It Matters
AM Intelligence is built on a bet that the scarce resource in artificial intelligence is not chips or talent but power. Training and running large AI models consumes enormous, continuous electricity, and the cost and reliability of that power increasingly determines where compute can be built economically. AMI's founders come from Greenko, one of India's largest renewable energy and storage operators, and their proposition is to turn that energy capability directly into AI computing capacity.
The business model is compute-as-a-service. Rather than sell power or hardware, AMI intends to build and operate AI data centres and rent their capacity to customers, hyperscalers, AI labs, enterprises and government-backed AI programmes, that need large amounts of compute without building it themselves. The Hyderabad AI factory, running NVIDIA's Vera Rubin systems, is the first physical instance of that model, sized at 30 MW and designed for the most demanding trillion-parameter and agentic AI workloads.
The cost structure explains why energy ownership matters so much. An AI data centre's economics are dominated by two things: the capital cost of the GPUs and the operating cost of powering and cooling them around the clock. By controlling renewable generation and long-duration storage upstream, AMI is trying to compress the largest recurring cost in the business, and to offer power that is both cheaper and lower-carbon than drawing from the grid. The liquid-cooled, high-density design of the Hyderabad factory is aimed at the same goal of running dense compute efficiently.
However, an order is not an operating business. A binding commitment for 9,000 of NVIDIA's newest systems demonstrates intent and capital, not revenue, and the company has not disclosed the committed customer demand that would fill 30 MW, let alone the 1 GW it plans. Whether the energy-first thesis produces a genuine cost advantage will only be visible once the factory is running and its compute is sold, and that is at least a year away.
The Strategic Read
The assumption being tested here is that India can host frontier AI compute at globally competitive scale, and that the binding constraint on doing so is power, not chips. AMI's entire pitch, the "electron-to-token" framing, is that the hard part of an AI data centre is not buying GPUs but feeding and cooling them cheaply and reliably. A venture rooted in renewable generation and long-duration storage is making the case that energy is the moat, and compute is the product it converts that energy into.
That is a genuinely differentiated position, and it is also where the risk concentrates. Owning the power stack could give AMI a cost advantage over data-centre operators who buy grid electricity, particularly for the enormous, sustained loads that AI training and inference demand. But converting a renewables-and-storage business into a frontier compute operator is a leap across very different disciplines, and the order underlines how capital-heavy it is: 9,000 of NVIDIA's latest systems is one of the largest known Rubin commitments anywhere, and it is the first tranche of a buildout the company itself sizes at more than $8 billion, before a separate $25 billion Noida project. These are among the largest infrastructure numbers announced by any Indian company, and they are announced plans, not deployed capacity.
Execution and timing are the immediate tests. Rubin supplies are tight, with multiyear backlogs as global hyperscalers compete for the same hardware, and NVIDIA has reportedly been raising system prices on the back of surging memory costs. Bringing 30 MW online in Q1 2027 would be one of the fastest Rubin deployments at scale anywhere; any slippage would ripple through the larger $8 billion plan and the credibility of the Noida timeline. The demand side is also unproven: AMI names hyperscalers, sovereign AI and homegrown model builders as customers, but has not disclosed committed offtake, and building gigawatts of compute ahead of contracted demand is the central bet of the entire AI-infrastructure cycle.
The context is a market already asking whether this cycle is overbuilt. Zoho's Sridhar Vembu has called AI valuations an "insane bubble", and India has seen large compute-and-cloud commitments announced faster than they can be filled. AMI's advantage, if it holds, is that it is selling the one input everyone else has to buy, power, and that its backers have built and financed energy infrastructure at scale before. The largest risk is the mirror image: that gigawatt-scale compute capacity, funded against announced rather than contracted demand, arrives into a market whose appetite has been assumed rather than proven.
For daily, sharp analysis of the biggest moves in the Indian business and startup ecosystem, follow StartupFox.
