"It's been less than a week since OpenAI completed its for-profit restructuring that saw Microsoft lose its ability to retain a stranglehold on the AI firm's computing resources, and Altman and company have already inked a $38 billion deal with AWS to start data crunching in another cloud. Prior to the restructuring that closed last Tuesday, Microsoft retained a right of first refusal if OpenAI were to take its cloud computing needs elsewhere,"
"Along with access to existing AWS resources, OpenAI will also be getting some custom infrastructure deployed for its use in the coming years that will involve some good old fashioned Nvidia GPUs. Per the press release, AWS is building "a sophisticated architectural design optimized for maximum AI efficiency and performance" using clusters of Nvidia Blackwell GB200 and GB300 chips. Those Nvidia GPUs will be clustered on the same network as EC2 UltraServers, AWS added."
OpenAI signed a seven-year, $38 billion agreement with Amazon Web Services to expand AI compute capacity beyond a single hyperscaler. The agreement grants immediate access to AWS EC2 UltraServers and will scale over seven years to tens of millions of CPUs for agentic workloads. AWS will deploy custom infrastructure incorporating clusters of Nvidia Blackwell GB200 and GB300 GPUs on the same network as EC2 UltraServers to optimize AI efficiency and performance. The deal follows OpenAI's for-profit restructuring that removed Microsoft's prior right of first refusal. AWS is also developing a rival platform to OpenAI's Stargate to build large-scale AI compute.
Read at Theregister
Unable to calculate read time
Collection
[
|
...
]