Capital One is worried that AI expenditures are increasing through its cloud-computing relationship with Amazon and possibly through shopping around, according to an internal Nvidia document reviewed by Business Insider. In the document, an Nvidia employee wrote that the chip giant had held talks with Capital One about AI infrastructure options, including Amazon Web Services, because the bank was “looking to control costs.”
The Nvidia staffer was summarizing talks with Capital One in an internal email after sitting down with the bank at a tech conference earlier this year.
“They realize that their demand for GPUs and reasoning models is growing, but the cost in AWS is getting out of control very fast,” the Nvidia employee wrote, referring to Capital One.
Terms discussed between Nvidia and Capital One included “AI factory and neo-clouds,” the email read. An AI factory is an in-house data center that a company can build to train and run A.I. models instead of renting compute from a third party. Financial companies can tap into this infrastructure for jobs such as fraud detection, customer service, and algorithmic trading, Nvidia said.
Neoclouds are upstart cloud providers, typically running on Nvidia-based hardware, that focus on AI workloads, while AWS serves a much broader range of computing needs. Leading the race for neocloud are CoreWeave, Lambda, Crusoe, and Nebius. Nvidia has been working closely with some of these players to reduce its dependence on established cloud giants as customers.
The Capital One episode underscores a fundamental dynamic of the A.I. frenzy. Generative AI is a race companies must embrace, even as they grapple with escalating cloud expenses. Such new technology looks very promising, but the cost of building and running AI models, as well as applications , can add up.
Big companies also often calculate their cloud costs, a mix of configurations that now include multiple providers. 43% of businesses use more than two public cloud providers, according to a recent RBC Capital Markets report.
“AWS remains our largest strategic cloud provider,” a Capital One spokesman wrote in an emailed statement. NVIDIA declined to comment.
“We have a pricing philosophy that we are going to continue to work really hard to take costs out of our own cost structure and pass that back into our AWS customers in the form of lower prices,” said an AWS spokesperson. It’s easy to lower prices; it’s tough to drive the cost down and be able to afford to lower prices, and at AWS, we do that day in and day out.”
Capital One is not the only company concerned about AWS spending. AI startups are at least waiting to do the typical AWS kind of work in favor of building AI models and developer tools later, per internal Amazon documents provided to Business Insider in March and July by an Amazonian.
The documents stated that 90 percent of the startups in Radical Ventures’ portfolio were building first and foremost on competing clouds because of the cost of using AWS. Amazon also wrote in an internal document that there was growing customer demand for so-called neocloud providers, which allow customers to rent GPU power in tiny chunks and pay only for what they use.
When that story came out in October, an AWS representative told Business Insider: “You’re using old data to draw outdated conclusions” and “AWS is the top choice for startups building because we have far more core services than anyone else, offer the best performance, the most profitable partners, the biggest community of customers,” plus the most innovative and influential generative AI offerings.
