Nvidia Invests in Cloverleaf to Solve Datacenter Power Bottlenecks
AI-generated image · US National Wire
The chip giant is leveraging its capital to ensure the physical infrastructure can support the power and cooling demands of next-generation GPUs.
As first reported by The Register, Nvidia has become a minority investor in Cloverleaf, a company founded in 2024 that specializes in securing the land and power required for datacenter construction. The move is part of a broader effort by Nvidia to remove roadblocks that could potentially pop the AI bubble, as the company can only sell as many GPUs as there are powered facilities to house them.
The shift is driven by the evolving hardware requirements of Nvidia's latest chips. While previous systems were largely air-cooled, the Blackwell generation introduced the NVL72 systems, which require liquid cooling. The Register reports that liquid-cooled facilities are more power-hungry, heavier, and require coolant distribution units (CDUs). This challenge intensifies with the upcoming Rubin generation, where Nvidia will offer chips exclusively in liquid-cooled varieties for HGX and NVL-style form factors, with rack power transitioning from roughly 250kW to 600kW starting next year.
To streamline deployment, Nvidia launched the DSX platform, which provides blueprints for power, cooling, and space. As part of the investment deal, Cloverleaf will adopt the DSX spec, including DSX Flex, which allows facilities to scale back workloads during grid demand spikes or offload demand to batteries and onsite power generation.
The Register notes that Nvidia has previously invested in model developers and neocloud partners—including OpenAI, CoreWeave, and Nebius—to help them secure financing. While the financial terms of the Cloverleaf deal were not disclosed, the investment aims to ensure that the physical layer of the AI economy can keep pace with Nvidia's hardware roadmap.