NVIDIA Rubin GPU order from AM Intelligence

For investors, the significance runs in two directions. For NVIDIA, the order reinforces the breadth of demand for its next AI generation even before the hardware is widely deployed, supporting the view that the company’s data-center franchise remains capacity constrained rather than demand constrained. NVIDIA has already disclosed supply and capacity commitments of $279 billion as of late July, underscoring how quickly the backlog for AI infrastructure is filling.

For AMI and its backers, the purchase is a high-stakes attempt to capture margin in the compute-as-a-service market before rivals build scale. The company says it is targeting a 400 MW compute pipeline across India, the US, Europe and Malaysia, with a longer-term goal of 1 GW of compute capacity and 5 GW of powered AI data centers. That is an aggressive buildout even by AI-infrastructure standards, requiring more than $20 billion of planned capital expenditure over the next 15 months on top of the $6 billion already committed for the first 100 MW.

The broader economic logic is straightforward: AI demand is increasingly bottlenecked by power, cooling and networking, not just by chips. AMI says its facilities will use liquid cooling, high rack-power density and high-speed RDMA over Converged Ethernet networking, the sort of infrastructure needed to run training and inference workloads for hyperscalers, AI labs, enterprises and sovereign AI programs. Backing from Greenko’s renewable power and storage assets gives the project an important advantage in markets where access to reliable electricity has become a strategic asset.

There are still material execution risks. The delivery schedule pushes the latest GPUs into 2027, which means the project must withstand changes in AI spending, competition from other cloud and GPU providers, and potential shifts in financing conditions before revenue is fully realized. The economics also depend on whether AMI can secure enough customers to monetize capacity at attractive utilization rates, a challenge that has already begun to separate winners from speculative build-outs in the AI infrastructure race.

Still, the transaction fits a wider pattern: AI infrastructure is moving from pilot deployments to industrial-scale capacity planning, with India and Southeast Asia emerging as important geographies. For NVIDIA, the order adds to a growing list of hyperscaler-style commitments that support its long-duration earnings case. For AMI, it is an attempt to position itself as a regional power-and-compute platform rather than just a data-center operator. Investors will now watch whether the company can convert a 400 MW pipeline into contracted revenue fast enough to justify the scale of the build.