Chip-design giant Nvidia is seeing high demand for its graphics processing units amid rising adoption of agentic AI. 

The demand is adding to Nvidia’s data center revenue, which increased 56% year over year to $46 billion, according to Nvidia’s fiscal second-quarter 2026 earnings results, announced Aug. 27.

Financial institutions are prioritizing deployment of agentic AI tools within their operations to boost accuracy and reduce hallucinations, Kevin Levitt, global business development lead for financial services at Nvidia, told Bank Automation News.  

Unlike general LLMs, Agentic AI models require  more computing power than general LLMs to think, reason and conduct research, Levitt said. 

Compared with traditional AI, agentic AI models could use 100 times — or even 1,000 times — more computing power, depending on the reading, research and comprehension required, Nvidia Chief Executive Jensen Huang said during the earnings call for fiscal Q2, ending June 30. 

“The amount of computation that has resulted in agentic AI has grown tremendously and, of course, the effectiveness has also grown tremendously,” Huang said. “Because of agentic AI, the amount of hallucination has dropped significantly.” 

However, the computing power required for agentic AI is expensive and that, combined with the demand for it, is creating a bottleneck in the tech’s adoption, Levitt said. 

The cost 

Agentic AI models often cost 10 times or more than per query compared to a single query from a general LLM, Jonathan Mitchell, risk management expert and financial industry lead at startup consultancy Founder Shield, told BAN. 

“This higher cost is driven by [agentic AI] architecture, which involves multiple iterative calls to the underlying LLM, as well as searches and tool-use to complete a single task,” Mitchell said. “While a general LLM provides a one-shot response, agentic models chain together many such actions, leading to a much higher cumulative computational and financial expense.” 

Nvidia’s graphic processing units (GPU), needed for the training and deployment of AI models, are expensive, and many smaller FIs might find their cost prohibitive, he said.  

On the other hand, larger FIs like American Express and BNY have the funding to invest in agentic AI. The $30 billion BNY, for one, is on track to deploy multiple AI factories to aid its AI tech roadmap, according to BAN’s prior reporting. 

GPU-as-a-Service 

Although smaller FIs don’t have the same resources, they can still play on the agentic AI field. However, their access to GPUs looks different. It is becoming common for them to use cloud-based services instead of investing in their own GPU infrastructures, Founder Shield’s Mitchell said. 

“Tech providers like CoreWeave, specializing in providing GPU-as-a-Service, would allow smaller FIs to rent the necessary compute power on a pay-as-you-go basis,” Mitchell said. “This approach would eliminate the need for a large upfront capital investment, making the adoption of agentic AI feasible for these institutions.” 

CoreWeave for instance allows companies to rent out GPUs’ compute for as low as $2.23 per hour, according to the company. Nvidia is one of the biggest investors in CoreWeave and hold nearly 7% of its equity, according to the CoreWeave’s prospectus filings with the SEC. 

The need for powerful GPUs will create an environment similar to the cloud market, Mitchell said. 

“Given the high cost and specialized nature of AI GPUs, it is highly likely that they will continue to be offered as a service, much like traditional cloud infrastructure.”