DGX servers equipped with GB200 will enter mass production in 2024H2

In March this year, Nvidia introduced its Blackwell architecture GPUs at the GTC 2024 conference held at the San Jose Convention Center in California. This lineup included the B200 GPU, designed to replace the H100/H200 models, and the GB200, which is integrated with the Grace CPU. Additionally, these new chips were utilized to develop the latest generation of the DGX series servers.

According to Wccftech, the DGX servers equipped with the GB200 chip are expected to enter mass production in the second half of 2024, potentially igniting a new fervor in the artificial intelligence (AI) market. With improvements in the new architecture and keen market interest, Nvidia is likely to see substantial profits. Reports suggest that the new generation AI servers based on the Blackwell architecture are more popular in the market than those based on the Hopper architecture, with anticipated purchases exceeding 40,000 units by 2025.

Nvidia plans to segment the AI servers equipped with the GB200 into three distinct markets, offering the DGX NVL72, NVL32, and HGX B200 models. The NVL72, previously introduced by Nvidia, is a full-rack solution comprising a cabinet with 18 1U servers, each incorporating two GB200 Grace Blackwell Superchips, totaling 36 Grace CPUs and 72 Blackwell GPUs. The NVL72 is rumored to be priced at around $3 million each, with an expected sale of 10,000 cabinets potentially generating $30 billion in revenue for Nvidia, not including ancillary products.

Foxconn and Quanta will handle the majority of the DGX series server supply. As the demand for AI computing power continues to rise, the future revenue prospects are looking even more promising, seemingly nothing can hinder Nvidia and its suppliers from experiencing explosive financial growth.