Nvidia currently ships over 1,000 MGX modular server racks every week. This hardware is produced through a network of 350 factory sites located across 30 countries. The MGX architecture supports power densities exceeding 120 kW per rack and is designed to accommodate future computing platforms, including the upcoming Vera Rubin generation.
Cloud hyperscalers are deploying approximately 1,000 NVL72 racks weekly, which incorporates 72,000 Blackwell GPUs per week. This deployment pace follows the resolution of previous manufacturing and integration challenges. The hardware requires specialized facilities capable of managing the high power and cooling demands associated with the 120 kW load per rack.
In October 2026, Nvidia introduced DSX Pod reference designs to further accelerate infrastructure rollouts. Each pod contains seven NVL72 racks and is engineered to reduce deployment timelines from six months to 14 days. The company expects the first deliveries of these container-sized units to occur in early 2027.
The company's ability to maintain or increase this weekly shipping volume remains a key performance indicator for the AI infrastructure market. Future success depends on both the efficiency of the global manufacturing partner network and the ability of data center facilities to scale their power and cooling capacities to meet the requirements of the new hardware.
