Nvidia is shifting its strategy to sell fully assembled AI servers, starting with the Vera Rubin platform. According to JP Morgan, this move involves providing Level-10 (L10) VR200 compute trays that come with all necessary hardware, including CPUs, GPUs, and cooling systems, pre-installed. By doing this, Nvidia aims to streamline the production process for its partners, which will reduce their design and integration workload. However, this consolidation may also cut into the profit margins of major Original Design Manufacturers (ODMs), as they will primarily focus on rack-level integration rather than full server design. The transition to pre-built L10 trays means Nvidia will standardize its compute systems, making them easier to produce and likely lowering costs through economies of scale. The Vera Rubin Superchip board features a complex design, suggesting that outsourcing production to specific Electronics Manufacturing Services (EMS) could be more efficient. J.P. Morgan also noted increases in power consumption for the new GPUs, which further highlights the necessity for enhanced cooling solutions. As a result of this integration, Nvidia’s partners will pivot from system designers to system integrators, maintaining some enterprise features and support roles. The heart of the server—the compute engine—will be entirely Nvidia’s responsibility. This shift raises questions about Nvidia's plans for its Kyber NVL576 rack-scale solution and whether it might further extend its influence in the supply chain.