AM Intelligence Orders 20,000 More Nvidia Rubin GPUs
AM Intelligence has placed two binding orders for approximately 20,000 Nvidia Rubin GPUs for planned facilities in India and Malaysia, adding about 70MW to its committed computing programme. Delivery is scheduled for the second quarter of 2027. Mint, citing two people familiar with the development, valued the orders at around US$4 billion; the company’s announcement confirms the quantity and binding status but does not disclose a purchase price. The orders lift AMI’s commitments to roughly 29,000 Rubin GPUs and nearly 100MW. For Nvidia, this is a named customer expanding an actual procurement commitment outside North America, with the eventual revenue timing still dependent on delivery and the applicable sales contracts.
The additional systems follow AMI’s August commitment for 9,000 Rubin GPUs at its first AI factory in Hyderabad. The new sites are intended to use Vera Rubin NVL72 rack-scale systems and a common infrastructure design, including high-speed networking, storage and liquid cooling. Standardising the architecture can simplify the company’s operating model across locations, while the different sites still require their own power and construction execution. The announced 70MW relates to the additional capacity, not the whole programme. Maintaining that distinction is important when assessing the scale of the procurement: the total of approximately 100MW combines the original Hyderabad commitment with the two newly announced orders for further facilities.
AMI was established by the promoters of Greenko, bringing an energy-infrastructure background to a business selling computing capacity. Mint reported that renewable generation and storage associated with the group would support the GPU estate. One person familiar with the plans described the intention as providing “Indian electrons for Indian tokens.” That formulation captures the commercial proposition of linking power supply with AI services, although the expansion also includes Malaysia and a wider international pipeline. Ownership or access to energy assets can help coordinate development, but it does not automatically resolve every local grid, connection or permitting requirement. The advantage depends on delivering dependable electricity at the specific sites where the computing equipment will operate.
The company’s longer-term programme is much larger than the orders placed so far. AMI describes a pipeline of approximately 400MW of compute capacity across several regions and plans to add another 300MW over about 15 months, with capital expenditure exceeding US$20 billion. That is additional to the US$6 billion it says is committed to the initial 100MW. These are infrastructure investment figures, not disclosed Nvidia revenue. Mint separately reported that around US$500 million had already been deployed toward the initial investment. The distinction between capital planned, capital committed and capital spent shows how much of the programme still depends on financing and project execution rather than completed operating assets.
The intended customers include cloud providers, enterprises, frontier-model developers and sovereign-AI initiatives. The orders establish AMI’s hardware procurement direction, but the announcement does not identify a corresponding set of signed end-customer contracts or guaranteed utilisation for every rack. That leaves the operator responsible for matching the available capacity with paying demand as it arrives. Its energy background may help address one major input cost, while the service business must also deliver reliable scheduling, networking, storage and support. Nvidia supplies the accelerated platform within that offering; AMI must turn the platform and its infrastructure into a service customers prefer to competing cloud capacity available locally or from international providers.
Analysis
The additional orders are more than twice the original 9,000-GPU commitment, materially increasing AMI’s exposure to Rubin and Nvidia’s potential delivery opportunity. AMI’s own US$6 billion figure for the initial 100MW implies about US$60 million of capital per megawatt across the infrastructure programme, not a per-chip price. That scale makes access to power and financing as important as obtaining processors. A coordinated energy-and-compute developer could reduce costly mismatches between those inputs, but it still bears utilisation risk. Nvidia’s strongest outcome is a customer whose infrastructure and contracted service demand arrive in step with the hardware.