Nvidia ran into some resistance as one of the world’s biggest banks struggled to adopt its AI enterprise software, signaling how hard it can be for massive, highly regulated companies to put cutting-edge technology to use.
Nvidia sales executives recapped conversations with key customers — including Bank of America — following a conference late last year, according to an internal email thread from November viewed by Business Insider.
The chip giant has been selling its “AI Factory” — a full setup of chips and software designed to build, train, and run large-scale AI systems — to large businesses.
Bank of America told Nvidia it was struggling with deployment, according to the email thread. The exchange reveals that while companies rush to purchase AI infrastructure, operational and regulatory hurdles make deploying it far harder — a key challenge for Nvidia as it expands from selling chips into enterprise software. On the thread, Nvidia executives also discussed how they can better work with customers in using its AI products.
“You sold us a Formula 1 race car,” an Nvidia executive reported the bank said, comparing the AI Factory to the race car, “and now you have to help us as local car mechanics drive the race car!”
Bank of America declined to comment. Nvidia did not respond to a request for comment from Business Insider.
A second executive later responded that Nvidia “can’t just sell” AI Factory hardware but needs to provide a software solution to help business customers succeed.
The gap between buying infrastructure and actually deploying AI is common across industries, said Rumman Chowdhury, who advises companies on responsible AI.
“Buying GPUs or signing a cloud contract is a business decision; deploying AI is an institutional change,” she told Business Insider. “It’s much easier to approve a budget line item than to re‑architect workflows, retrain teams, and rewrite governance processes.”
Lacking ‘MLOps skills’
Recapping its meeting with Bank of America, the first Nvidia executive said the bank lacked “the MLOps skills in house.” MLOps refers to machine learning operations, or the processes for implementing AI models in real-world use cases.
That executive added that Bank of America did not think Nvidia’s AI enterprise software was “ready for their highly regulated banking industry.”
The executive also pointed to other concerns, including Bank of America’s security and governance requirements, such as documentation and support for air gapping — isolating systems from other networks to improve security. They noted the challenges the bank faced in supporting multiple AI models and software systems to meet different needs.
Nvidia vice president Ian Buck subsequently jumped into the thread, signalling how senior leaders at the chip giant can step in when customer concerns surface.
“Looks like they need help and/or our product is coming up short,” Buck wrote.
The struggles at Bank of America echo earlier issues with Nvidia’s enterprise software efforts, including a need to educate prospective clients on what it is and isn’t.
AI deployment obstacles aren’t exclusive to banking; they are prevalent across sectors. While banks have a long history of using AI for tasks like credit decisioning, they may be the first to run into issues because of the scale of their data and customers, said Tom Davenport, an information technology and management professor at Babson College.
“The technology’s out way ahead of what individual banks or most companies actually can implement quickly,” he said.
Have a tip? Contact this reporter via email at [email protected] or Signal at @geoffweiss.25. Use a personal email address, a nonwork WiFi network, and a nonwork device; here’s our guide to sharing information securely.
