Dell and Super Micro launched AI servers built on Nvidia’s Vera Rubin platform, scaling up to 1,152 GPUs per rack and introducing direct liquid cooling for high-density workloads. Reflection AI agreed to pay SpaceX $150 million per month from July 2026 through 2029 for Nvidia GB300 chips, totaling $6.3 billion.
Dell’s PowerEdge XE8812 and Super Micro’s liquid-cooled racks will deploy Nvidia’s Vera Rubin NVL4 architecture, supporting up to 144 GPUs per rack from Dell and 1,152 GPUs per rack from Super Micro, marking a major jump in compute density and memory capacity for AI and HPC workloads.
Reflection AI secured immediate access to Nvidia GB300 chips at SpaceX’s Colossus 2 data center under a $150 million per month contract from July 1, 2026, through 2029, totaling $6.3 billion, with a 90-day termination option after the first three months.
Apple’s WWDC outlined a hybrid AI architecture that routes demanding tasks to cloud models powered by Nvidia graphics processors, including a 20-billion-parameter sparse model in AFM 3, potentially increasing weighted cloud compute costs and GPU demand.
The combination of OEM server partnerships, large-scale compute contracts and integration into leading device ecosystems underscores Nvidia’s expanding role as a critical AI infrastructure provider, supporting its near-term revenue growth and long-term market leadership.
