AWS said that the transition from H100 to Blackwell is for the Ceiba project only, and will continue to provide cloud services based on H100 in the future.
Author of this article: Li Xiaoyin
Source: Hard AI
AWS responds to reports that “it has been completely updated to Blackwell GPUs”: untrue, only for the Ceiba project.
On Tuesday local time, Amazon Cloud Computing Services (AWS), the world's largest cloud service provider, publicly stated that the department has not cancelled any Nvidia chip orders and plans to purchase Nvidia's next-generation Blackwell chips.
The media previously reported that AWS had “completely transitioned” its order for the Nvidia H100 to a new generation of Blackwell GPUs.
An AWS spokesperson further stated to the media that the transition from H100 to Blackwell is for the Ceiba project only, and cloud services based on H100 will continue to be provided in the future.
The statement emphasized:
“To be clear, AWS is not cancelling any Nvidia orders. In the process of working closely with Nvidia, we decided together to move Project Ceiba from the Hopper architecture-based H100 chip to the Blackwell GPU, which made a leap forward in performance.”
At the end of 2023, AWS announced a partnership with Nvidia to jointly launch Project Ceiba, the world's largest cloud AI supercomputer. In March of this year, after Nvidia announced the next generation of Blackwell GPUs, the Project Ceiba project also officially announced chip iterations.
According to some opinions, as one of the major customers of Nvidia chips, AWS's clarification this time is very important for investors, because market demand for Nvidia chips is regarded as an important indicator for measuring the health of the entire AI and cloud computing industry.