In a significant development highlighting the financial pressures of artificial intelligence adoption, US banking giant Capital One is actively exploring options beyond Amazon Web Services (AWS) for its AI workloads. This strategic reconsideration, driven by concerns over escalating costs, was revealed in an internal Nvidia document obtained by Business Insider.
AI Cost Concerns Prompt Cloud Rethink
The internal communication from an Nvidia employee detailed discussions held with Capital One representatives at a recent technology conference. The memo explicitly stated that the financial institution foresees its demand for GPUs and reasoning models growing substantially, and is worried that the associated expenses on AWS will soon become unmanageable. This search for cost control underscores a growing tension between ambitious AI projects and the cloud budgets of major corporations.
According to the report, the talks focused on two primary alternatives: building an "AI factory" or turning to "neocloud" providers. An AI factory refers to a company constructing its own in-house data centre specifically designed to train and operate AI models, thereby reducing reliance on external cloud rentals. Neoclouds are a new breed of cloud service providers, such as CoreWeave, Lambda, and Crusoe, which specialise in AI and machine learning tasks using powerful Nvidia hardware.
Industry-Wide Scrutiny of Cloud Bills
Capital One's deliberations are not an isolated case. They reflect a broader industry trend where companies are diversifying their cloud partnerships to optimise costs. A report from RBC Capital indicates that 43% of enterprises now use multiple public cloud providers. Furthermore, Business Insider had earlier reported that numerous AI startups are postponing their AWS expenditures, choosing competitor platforms instead. Internal Amazon documents suggested that a staggering 90% of startups in the Radical Ventures portfolio are primarily building on rival clouds due to financial considerations.
This exploration is particularly noteworthy because of Capital One's historically deep ties with AWS. The bank was a pioneer in the financial sector, becoming the first major institution to completely migrate to the public cloud, a process that involved shutting down its data centres over a two-year period, as reported by CIO.com in July. Despite Capital One's AI chief, Prem Natarajan, having highlighted dramatic reductions in inference costs, the current search for alternatives suggests that the overall expenditure on AI infrastructure remains a pressing challenge.
Commitment to AWS Amidst Exploration
In response to the report, a Capital One spokesperson reaffirmed the bank's relationship with Amazon, stating, "We continue to be committed to AWS as our predominant strategic cloud partner." However, the confirmed discussions with Nvidia about other options reveal a pragmatic approach to infrastructure strategy. The bank appears to be balancing its long-term partnership with the immediate need to manage the ballooning costs associated with cutting-edge AI development and deployment.
The situation presents a critical moment for cloud providers. As AI becomes central to business operations across sectors, from finance to startups, the traditional cloud pricing and service models are facing intense scrutiny. Companies are now more willing to consider hybrid approaches, specialised providers, or even in-house solutions to keep their transformative AI ambitions financially sustainable.
