Wall Street Banks Face an AI Dilemma: Big Opportunity, New Risks

Artificial intelligence is giving Wall Street banks two major opportunities at the same time.

They can finance the enormous amount of infrastructure required to support AI, from data centers to computing capacity.

And they can deploy AI inside their own organizations to automate work, develop products faster and reduce costs.

The problem is that both opportunities create new risks.

Executives at Goldman Sachs, Citigroup, JPMorgan Chase and other major U.S. banks are increasingly asking whether AI investment will generate sufficient long-term returns—and how financial institutions can safely use autonomous AI agents without exposing customers or internal systems to new threats.

Banks Are Financing an AI Infrastructure Boom

The rapid expansion of generative AI requires much more than software.

It depends on large amounts of physical infrastructure, including servers, advanced semiconductors, data centers, networking systems and electricity.

Building that capacity requires enormous amounts of capital.

For Wall Street banks, this creates opportunities in lending, debt markets, equity financing, advisory work and structured transactions.

But it also puts banks directly in the path of one of the largest and least predictable capital-spending cycles in technology.

Goldman Sachs Sees Opportunity—and Uncertainty

Goldman Sachs CEO David Solomon said the bank occupies a privileged position in the AI investment cycle.

Current demand for computing capacity remains strong, with supply yet to fully catch up.

More infrastructure is therefore likely to be built and more financing will be required.

At the same time, Solomon stressed that no one knows how demand for capital and compute will evolve as the broader AI ecosystem matures.

That uncertainty matters for lenders.

If a data-center project is built on assumptions of permanently explosive compute demand, weaker-than-expected utilization could eventually pressure returns and credit quality.

The AI boom is therefore a financing opportunity, but not every AI-linked asset will necessarily be a good loan.

Citi Does Not See AI Infrastructure Spending Slowing Yet

Citigroup CEO Jane Fraser said higher interest rates have not halted investment across the broader AI ecosystem.

One of the immediate constraints remains computing capacity.

Demand for compute remains strong as companies need increasing amounts of processing power to train and operate AI systems.

That imbalance continues to support infrastructure spending.

Still, strong demand today cannot eliminate uncertainty about how much infrastructure the market will ultimately need once the technology matures.

Banks Are Also Becoming Major AI Users

Banks are not merely financing other companies' AI spending.

They are deploying the technology internally.

The potential benefits include faster software development, fewer manual processes, lower operating costs and shorter customer-service response times.

Citigroup says it is already seeing measurable financial benefits.

Citi Says Product Launches Are 30%–40% Faster

During its third-quarter 2026 earnings call, Citi said AI had accelerated its product-development cycle.

The bank said products were typically being launched 30% to 40% faster, while also being developed at lower cost.

This figure needs to be interpreted carefully.

It does not mean productivity across Citigroup's entire workforce increased by 40%.

It specifically refers to the speed of product launches.

That distinction matters when assessing the economic impact of corporate AI adoption.

14,000 Citi Engineers Use an Agentic Coding Tool

Citi also said approximately 14,000 engineers use its main agentic coding tool, Devin.

The bank said the technology has made it easier to bring more work in-house.

Technology vendor spending has consequently declined by around 30% year over year, according to Fraser.

Around 100 highly manual processes are also being redesigned using AI and automation.

Fraser said financial benefits from AI and automation have significantly exceeded the bank's investments in the technology so far this year.

That provides one of the clearer examples of a large financial institution reporting tangible returns from AI deployment.

JPMorgan Says Benefits Can Take Time to Appear

Not every AI benefit is immediately visible in a bank's earnings.

JPMorgan CFO Jeremy Barnum said transformative technology can take considerable time before its effects become obvious at an institution of JPMorgan's scale.

That makes it unsurprising that some AI benefits are not yet clearly visible in quarterly results.

The challenge illustrates the difference between improving individual workflows and materially changing the profitability of an entire bank.

Then Comes Agentic AI

The next phase of AI adoption may be more difficult to govern.

Agentic AI systems can go beyond producing answers or recommendations.

They can potentially execute sequences of actions on behalf of users.

In financial services, that could eventually include gathering information, interacting with accounts or initiating tasks.

That capability makes agents potentially useful.

It also introduces an entirely different level of security and liability risk.

JPMorgan Is Worried About Rogue Agents

JPMorgan CEO Jamie Dimon said the bank is already using AI agents internally and expects to continue developing them.

But he also warned about the risk of “rogue agents.”

The bank is building tools designed to protect customers from agentic AI risks and does not intend to allow arbitrary external AI agents into its systems.

Dimon also highlighted the issue of liability.

If a customer gives credentials or access to a third-party agent, questions arise over responsibility if information is misused or the agent performs an unintended action.

That makes agentic AI not only a cybersecurity challenge, but also a governance and consumer-protection issue.

AI Could Make Cyberattacks More Powerful

Banks already operate in one of the most heavily targeted sectors for cybercrime.

AI potentially gives attackers new tools.

Generative systems can increase the scale and sophistication of phishing, social engineering and other forms of digital fraud.

The concern now extends beyond individual banks.

On October 16, G7 finance ministers and central-bank officials agreed to hold tabletop exercises designed to prepare financial systems for AI-enabled cyberattacks.

That development shows that policymakers increasingly view AI-driven cyber risk as a financial-stability concern.

Banks Are Exposed on Two Fronts

The industry's AI dilemma can be divided into two broad categories.

The first is financing risk.

Banks are helping provide capital to data centers and other infrastructure whose long-term economics remain uncertain.

The second is operational risk.

As banks themselves deploy AI, they must control cybersecurity, model errors, data privacy, autonomous agents and third-party access.

The same technology that creates a new investment-banking revenue opportunity can therefore introduce new risks inside the institution itself.

Doing Nothing Carries a Cost Too

Avoiding AI entirely is not a simple solution.

If Citi and other institutions can launch products faster, lower vendor costs and automate manual work, banks that fail to adopt similar tools may become less efficient.

Customers may also begin to expect faster and more intelligent digital services.

That means financial institutions face a balancing problem.

Move too slowly, and they risk losing competitiveness.

Move too quickly, and they may expose customers and internal systems to risks that are still poorly understood.

The Real Question Is How Fast Banks Should Move

The debate is therefore no longer primarily about whether banks will use AI.

They already are.

The more important questions concern governance.

Which tasks should an AI agent be allowed to perform?

When should a human remain responsible?

What happens when an external agent seeks access to financial data?

Who carries liability when an autonomous system makes a mistake?

And how much capital should banks commit to infrastructure whose long-term returns remain uncertain?

Those questions are becoming part of everyday banking risk management.

The Infrastructure Cycle Will Eventually Recalibrate

Current demand for computing power is exceptionally strong.

But Solomon has acknowledged that the investment cycle will eventually recalibrate, even if the timing is impossible to predict.

That is why credit discipline matters.

An AI label does not automatically make a project economically sound.

Banks still need to assess customers, cash flows, leverage, contracts and long-term demand just as they would in any other capital-intensive industry.

Bottom Line

Artificial intelligence has placed Wall Street banks in a powerful but complicated position.

They can earn fees and financing revenue from the massive buildout of AI infrastructure while using the same technology to make their own operations more efficient.

Citigroup says AI has helped increase product-launch speed by 30%–40%, while technology vendor spending has declined about 30% year over year.

Yet Goldman Sachs is warning about uncertainty over future capital and computing demand, while JPMorgan is building defenses against the risks of autonomous AI agents.

For banks, the central challenge is therefore not whether to participate in the AI boom.

It is how to capture the economic benefits without allowing credit, cybersecurity and operational risks to grow just as quickly.