NVIDIA's new DGX GB200 AI server is expected to enter mass production in the second half of this year, with the volume expected to hit 40,000 units in 2025.
NVIDIA and Quanta are the two suppliers of NVL72 and NVL36 cabinets, respectively, with the NVL72 packing 72 GPUs and 36 Grace CPUs. Each of the AI server cabinets cost 96 million NTD (around $3 million USD or so).
The new NVIDIA DGX NVL72 is the AI server with the most computing power, and thus, the highest unit price. Inside, the DGX NVL72 features 72 built-in Blackwell-based B200 AI GPUs and 36 Grace CPUs (18 servers in total with dual Grace CPUs and 36 x B200 AI GPUs per server) with 9 switches. The entire cabinet is designed by NVIDIA in-house, and cannot be modified, it is 100% made, tested, and provided by NVIDIA.
The high unit price of the new DGX B200 AI servers means analysts expect around 10,000 units to be sold in 2025 of the most powerful DGX NVL72 AI server.
- Read more: NVIDIA's next-gen GB200 AI server chips go into mass production in September
- Read more: Quanta to make NVIDIA GB200-based AI servers for Google, Amazon, and Meta
- Read more: NVIDIA GB200 Grace Blackwell Superchip: 864GB HBM3E, 16TB/sec bandwidth
- Read more: NVIDIA's full-spec B200 AI GPU uses 1200W of power, Hopper H100 uses 700W
Next up we've got the NVL36 cabinet which is also designed by NVIDIA, packing 36 x B200 AI GPUs and 16 Grace CPUs (9 servers in total) with 9 switches. Open server ODM companies can design their own input and output interfaces and Ethernet connections.
NVL72 is a 1U design per server, while NVL36 is a taller 2U design, providing room for ODM operators to design their own cooling solutions that would fit in the larger area inside of the cabinet space. Most manufactuers are moving towards air-water cooling mixed-cooling designs, with Quanta to take the largest share of supply, with Wistron, Wiwynn, and Inventec with the rest of the NVL36 cabinet orders.
NVIDIA's new DGX B200 cabinet is the entry-level AI computing power solution, providing ODMs better design flexibility and space. UDN reports that because the DGX B200 AI server is not NVIDIA's leading server, NVL72 and NVL36 cabinets will be part of the first delivery waves.