On October 8, Microsoft Azure announced on social media that it has acquired AI servers equipped with NVIDIA's GB200 superchip, making it the first cloud service provider globally to implement the Blackwell architecture.
The GB200 superchip and the new Blackwell computing platform were unveiled by NVIDIA in March of this year. Compared to the previous generation H100 GPU, the GB200 offers a remarkable 30-fold increase in performance for large language model inference workloads.
The GB200 chip is based on NVIDIA's next-generation AI graphics processing architecture, Blackwell, and is manufactured using TSMC's cutting-edge 4-nanometer (4NP) process. According to NVIDIA's official data, the chip delivers an impressive AI performance of 20 petaflops, significantly surpassing the H100's 4 petaflops. This enhanced processing capability will enable AI companies to train larger and more complex models.
Analysts estimate that the price of a single NVIDIA GB200 superchip, which includes both CPU and GPU, could reach $70,000 (approximately 494,000 Chinese Yuan).
Additionally, during the 2024 Hon Hai Technology Day held on October 8, Foxconn Senior Vice President Benjamin Ting announced plans to construct the world's largest NVIDIA GB200 chip manufacturing facility in Mexico, although specific location details were not disclosed.