1. Direct Investment in Energy and Data Center Companies: Nvidia has invested in companies like Cloverleaf, SB Energy, and Lancium, aiming to secure new power and data center resources in advance and ultimately use this power for Nvidia GPU deployment.
2. Leasing Data Centers Initially: Nvidia has signed approximately $20 billion in data center leasing commitments, sometimes securing server rooms first and then gradually transferring them to cloud service providers and other customers.
3. Global Search for Idle Power: Nvidia is tracking available land, power, and data center resources globally, focusing particularly on regions with available power, such as Northern Europe and Texas, and helping customers match power suppliers with data center developers.
4. Shift to Smaller, Distributed Data Centers: Compared to large campuses with complex approval processes and long construction cycles, Nvidia and its customers are increasingly focusing on smaller projects to reduce the risk of large projects failing to reach power on time.
5. Developing Modular Data Centers: Breaking down traditional data centers into many standardized, prefabricated modules and utilizing existing power-connected sites such as factories and office buildings facilitates rapid deployment. For example, Crusoe launched a 1MW modular data center with a volume only the size of a shipping container.
6. MaxLPS Staggered GPU Power Consumption: NVIDIA launched the MaxLPS power management system, which uses software, sensors, and in-rack control devices to stagger server power consumption peaks. NVIDIA claims that up to 40% more GPUs can be deployed with the same power capacity.
