Nvidia memo says Capital One discussed alternatives to AWS as AI costs could 'get out of hand'
Briefly

Nvidia memo says Capital One discussed alternatives to AWS as AI costs could 'get out of hand'
""They see their need for GPUs and reasoning models growing and the costs in AWS will soon get out of hand," the Nvidia employee wrote, referring to Capital One. Nvidia and Capital One discussed "AI factory and neo-clouds," according to the email. An AI factory is an in-house data center that a company can build to train and run AI models as an alternative to renting compute from a third party. Financial institutions can use this infrastructure for tasks such as fraud detection, customer support, and algorithmic trading, according to Nvidia."
"In the document, an Nvidia employee wrote that the chip giant talked with Capital One about AI infrastructure alternatives to Amazon Web Services, as the bank was "looking to control costs." Neoclouds are upstart cloud providers, often powered by Nvidia hardware, that focus on AI workloads, whereas AWS supports a much broader range of computing needs. Top neocloud players include CoreWeave, Lambda, Crusoe, and Nebius. Nvidia has been working closely with several of these players, in part to reduce its reliance on established cloud giants as customers."
Capital One is concerned about rising AI costs from its cloud relationship with AWS and is exploring alternatives to control spending. Nvidia representatives reported discussing AI infrastructure options with Capital One, including in-house AI factories and neocloud providers. An AI factory is an in-house data center to train and run models instead of renting third-party compute. Neoclouds are specialized, Nvidia-powered cloud providers focused on AI workloads, with players like CoreWeave, Lambda, Crusoe, and Nebius. Nvidia has collaborated with neoclouds to diversify customer channels. Companies adopting generative AI face growing compute demand and seek ways to mitigate escalating cloud expenses.
Read at Business Insider
Unable to calculate read time
[
|
]