PNC Financial Services Group plans to develop most of its artificial intelligence technology independently rather than relying on outside vendors, a strategy the Pittsburgh-based bank's chief executive has branded an “AI factory.”

CEO Bill Demchak described the project at investor conferences, and Bloomberg highlighted the approach in an October 15 report. Where most banks treat AI like a streaming subscription—choosing a vendor, paying the monthly bill and hoping prices do not jump—PNC intends to own the underlying infrastructure itself.

The plan involves acquiring data centers and purchasing GPUs from Nvidia, the specialized chips that power modern AI systems. PNC intends to use that hardware to build its own large and small language models tuned specifically for banking.

What PNC Is Building

The bank's near-term use cases are narrow and practical: fraud detection and call-center automation. These are high-volume, repetitive tasks where a model trained on banking data could, in theory, outperform a general-purpose tool.

PNC is not cutting itself off from the outside world entirely. The bank still maintains selective external partnerships, but the dominant intent is self-sufficiency.

The Numbers the Bet

PNC holds $603 billion in total assets, placing it among the largest banks in the United States and giving it the balance sheet to fund its own infrastructure—a luxury smaller regional lenders may not share.

The bank has identified 171 potential automation opportunities, which together represent $1.4 billion in addressable spend over the 2025 to 2030 window. Addressable spend refers to the pool of existing costs that automation could potentially shrink or replace.

PNC expects these AI initiatives to deliver an additional 40 points of operating leverage over the same period. Operating leverage measures how much faster revenue grows than expenses; when a bank adds points of it, more of each new dollar earned falls to the bottom line rather than being absorbed by overhead.

There is a track record behind the projection. Earlier automation work produced 30 points of operating leverage in retail and care-center operations between 2022 and 2025.

Why Skip the Vendors

Cost control sits at the core of the strategy. Many external AI providers charge on a usage basis, meaning costs rise in step with consumption. For a bank seeking to automate across 171 use cases, usage-based pricing amounts to a meter that never stops running. Owning the infrastructure outright is designed to sidestep those future costs.

The trade-off runs the other way, too: ownership converts metered vendor fees into large upfront spending on data centers and chips, along with the ongoing work of running and refreshing that hardware.

A second concern is vendor lock-in. Once a bank builds its workflows around one provider's models, switching becomes expensive and slow. If that vendor raises prices, changes terms or falls behind competitors, the bank is left with little room to move.

The Execution Question

For PNC shareholders, the central question is execution. Building data centers and training proprietary models requires capital, talent and time, and any payoff would surface in efficiency metrics over several years. The results could also serve as a reference point for other banks weighing whether to rent AI capability from vendors or build it themselves.

Nvidia sits on the other side of the equation. Whether banks rent AI through vendors or build it themselves, the chips have to come from somewhere. PNC's plan means the bank is buying the hardware directly rather than paying for it indirectly through someone else's invoice.

The measurable markers are straightforward: progress on the 171 identified opportunities, any disclosed savings against the $1.4 billion in addressable spend, and whether operating leverage gains continue at the pace set during the 2022 to 2025 period.