Artificial intelligence (AI) in banking has moved beyond experimentation. It is increasingly embedded in underwriting, payments processing, liquidity management, fraud detection and client advisory workflows. The question is no longer whether banks should deploy AI but how they govern it, scale it and attribute its economic impact. The transition underway is structural: operating models are being reorganised around enterprise decision infrastructure.

JPMorgan Chase (JPMorgan) illustrates the scale of this transition. The bank reported full-year 2025 net income of $57.5 billion with return on tangible common equity of 20%. Chief financial officer Jeremy Barnum said that technology spending is expected to reach approximately $19.8 billion in 2026, up 10% year-on-year. AI initiatives form part of a broader productivity and infrastructure modernisation programme rather than a standalone innovation agenda.

Chief executive Jamie Dimon has described AI as a technology likely to influence nearly every part of the bank’s operations, from software development and customer service to risk management and trading. He has argued that large institutions must invest heavily to ensure they remain competitive as AI becomes embedded across the financial industry.

Bank of New York Mellon (BNY) frames AI similarly. The custody and payments institution generated more than $20 billion in revenue last year with pre-tax margins of roughly 44%. On its fourth-quarter 2025 earnings call, chief executive Robin Vince described the year as one of record revenue and operating leverage. The emphasis was on platform execution and infrastructure simplification rather than technology experimentation.

BBVA too has articulated the same perspective in Europe. Chief executive Onur Genç has repeatedly described data and AI capabilities as central to pricing, risk management and client engagement. The technology is not peripheral to the bank’s strategy; it underpins profitability under increasingly competitive conditions.

In Latin America, Nu Holdings presents a different but complementary model. Founder and chief executive David Vélez emphasises that AI now sits at the centre of the company’s credit and service architecture. Proprietary models embedded in underwriting allow the bank to scale lending while maintaining portfolio discipline.

Across Asia and the Middle East, banks including DBS, Standard Chartered and Emirates NBD have expressed similar priorities in recent earnings calls. Executives increasingly attribute resilience to behavioural analytics, funding depth and disciplined execution rather than to interest rate cycles. AI supports these capabilities by embedding data-driven decisioning into core banking processes.

The common thread is not the scale of AI spending but its integration into accountable banking. The intelligent bank at scale is defined less by experimentation than by governance, economic attribution and the ability to embed machine-assisted decisions into institutional processes without eroding trust.

From digital bank to intelligent infrastructure

The distinction between digital capability and intelligent infrastructure lies in decision authority. Automation digitises workflows; intelligent systems shape judgement. The former improves efficiency, while the latter influences credit approval, fraud detection, liquidity allocation and client advice.

At JPMorgan, Barnum highlighted the expansion of AI use cases across customer service, fraud analytics and internal software development. Machine learning and analytical AI have improved revenue and expense outcomes for years, particularly in marketing and fraud detection. Generative AI applications are now expanding, with the number of use cases in production doubling over the past year.

BNY’s approach reflects the same logic. AI is embedded within treasury services and payments infrastructure rather than treated as a separate innovation initiative. Reconciliation, liquidity monitoring and client engagement workflows increasingly incorporate machine-assisted analytics to strengthen operational reliability across institutional mandates.

BBVA has similarly positioned AI as a core component of pricing discipline and risk monitoring. Data analytics and automation support cost efficiency and personalised client engagement while improving credit underwriting precision.

Nu Holdings illustrates how AI can be embedded from inception. As a digital-native bank, its infrastructure was designed around data and automation. Vélez described proprietary models integrated into underwriting and customer interaction systems that enable rapid scaling of small-ticket lending while maintaining portfolio discipline.

Across these institutions AI is not an add-on capability but part of the operating model. The intelligent bank at scale relies on centralised data architecture, disciplined model governance and feedback loops linking decision outcomes to capital allocation and risk oversight.

Jeremy Barnum, Chief Financial Officer, JPMorgan Chase

Jamie Dimon, Chief Executive Officer, JPMorgan Chase

Governance, explainability and institutional trust

Deploying AI at scale introduces regulatory and reputational constraints. Models that influence credit decisions, fraud detection or pricing must be explainable, auditable and consistent across jurisdictions.

JPMorgan’s scale makes governance unavoidable. Barnum insisted that AI initiatives are pursued alongside careful risk management even as use cases expand. Technology spending therefore sits within a framework of internal controls embedded across the bank’s three lines of defence.

BNY’s AI deployments follow similar principles. Machine learning models operate within existing monitoring frameworks rather than outside them. Anomaly detection, reconciliation automation and liquidity monitoring reinforce compliance processes while improving operational efficiency.

BBVA has also stressed that ethical AI frameworks and governance structures are integral to its deployment strategy. In markets subject to strict supervisory oversight, transparency and accountability remain prerequisites for technological adoption.

The implication is clear: the intelligent bank must remain auditable. Governance frameworks must track model inputs, outputs and drift, while boards and senior management review AI performance with the same discipline applied to credit risk and capital allocation.

Trust in this context is institutional rather than rhetorical. It reflects confidence that machine-assisted decisions remain fair, consistent and aligned with prudential standards.

Funding, balance sheet resilience and AI-enabled performance

Ultimately the intelligent bank is judged by financial performance rather than technological ambition. In 2025 declining benchmark rates and tightening liquidity conditions placed pressure on margins across several markets. Institutions that maintained profitability did so through disciplined funding strategies and increasingly precise analytics.

JPMorgan delivered full-year net income of $57.5 billion with ROTCE of 20%, while maintaining a standardised CET1 ratio of 14.5%. Barnum indicated that net interest income excluding Markets could remain around $95 billion in 2026 despite lower forward rates. Deposit growth, balance sheet mix and disciplined loan expansion are expected to offset part of the rate headwind.

DBS faced similar conditions. The bank reported net profit of SGD 11.0 billion ($8.1 billion) on total income of SGD 22.9 billion ($16.9 billion) even as benchmark rates declined. Loan growth reached approximately 6%, while deposits increased by SGD 64 billion ($47.3 billion). Chief executive Tan Su Shan described resilience as increasingly dependent on funding depth and behavioural analytics rather than cyclical repricing.

Standard Chartered reported operating income of $20.9 billion and underlying profit before tax of $7.9 billion, with return on tangible equity (RoTE) to 14.7%. Customer deposits increased to roughly $530 billion while loans grew more modestly, reflecting a shift towards wealth management, transaction banking and capital markets activity.

Emirates NBD provides another example. The bank’s balance sheet expanded beyond AED 1 trillion ($317 billion) in assets while lending and deposits grew strongly across its regional markets. Yet margin pressure persisted, reinforcing the importance of pricing analytics, customer segmentation and operational efficiency.

Nu Holdings operates a different but complementary model. Its small-ticket, high-frequency lending business depends on AI-driven underwriting and service automation. Proprietary models allow the bank to process large application volumes at low unit cost while maintaining risk discipline.

Across these institutions AI increasingly informs pricing elasticity, delinquency forecasting, funding behaviour modelling and liquidity optimisation. Technology strategy and capital strategy are becoming inseparable.

Cross-border execution and network complexity

As global trade corridors fragment and capital flows diversify, operational complexity in banking continues to increase. Payments infrastructure must process growing volumes across currencies, regulatory regimes and settlement systems.

BNY’s global payments network processes transactions in more than 120 currencies. AI-enabled workflows support reconciliation, exception management and liquidity monitoring across treasury services, improving reliability while remaining embedded within existing governance frameworks.

Standard Chartered’s franchise across Asia, Africa and the Middle East illustrates another dimension of cross-border complexity. The bank increasingly focuses on financial institution and corporate clients whose activities span multiple jurisdictions. AI-supported analytics help monitor trade flows, liquidity patterns and client exposures across these corridors.

Emirates NBD’s expansion into markets such as Saudi Arabia and India demonstrates how corridor-driven growth intersects with technology deployment. As the bank broadens its regional footprint, analytics increasingly support monitoring of funding composition, lending discipline and transaction flows.

In this environment AI becomes a critical tool for managing network complexity while preserving regulatory oversight.

Robin Vince, Chief Executive Officer, Bank of New York Mellon

Onur Genç, Chief Executive Officer, BBVA

Operational industrialisation and measurable AI value capture

For AI to qualify as enterprise infrastructure it must deliver measurable economic value. Productivity gains, revenue improvements and risk reduction must be quantified and assessed alongside traditional financial metrics.

JPMorgan’s technology spending illustrates this approach. Barnum indicated that approximately $19.8 billion will be allocated to technology in 2026, with around $600 million in efficiencies already identified across operations and support functions.

BNY’s institutional franchise demonstrates similar integration. AI embedded within reconciliation, liquidity monitoring and operational analytics improves straight-through processing rates and reduces manual intervention.

Industrial-scale deployment also requires lifecycle governance. Models must be retrained, validated and monitored for drift while infrastructure continues to evolve through application and data modernisation.

The economic test of AI is therefore durability. Productivity improvements must persist across reporting periods, risk metrics must remain stable under stress and revenue contributions must withstand macroeconomic shifts.

Model governance, regulatory supervision and institutional accountability

As AI becomes embedded in banking infrastructure, supervisory expectations are evolving alongside technological capability. Models that influence credit decisions, liquidity forecasts or pricing algorithms increasingly fall within the scope of regulatory scrutiny traditionally applied to capital adequacy, market risk and operational resilience.

Large institutions therefore face a dual challenge. They must deploy AI systems capable of improving operational performance while ensuring that those systems remain explainable, auditable and consistent with prudential supervision. The complexity arises from the fact that machine learning models often operate as probabilistic systems whose outputs depend on dynamic data inputs rather than deterministic rules.

Banks are responding by expanding model governance frameworks originally developed for quantitative risk models. Model inventories are being extended to include machine learning systems used in underwriting, fraud detection and customer interaction. Each model typically requires documentation of its design, training data, validation procedures and performance monitoring metrics.

Institutions such as JPMorgan, BBVA and Standard Chartered have all emphasised the importance of integrating AI oversight into existing model risk management structures. Rather than establishing parallel governance regimes, most banks are adapting the frameworks already used for credit and market risk models. This ensures that machine learning systems remain subject to validation, stress testing and independent review.

Regulators are increasingly attentive to these developments. Supervisory authorities in Europe, the United States and Asia have begun issuing guidance on the responsible use of AI in financial services. These frameworks generally emphasise explainability, accountability and human oversight.

The challenge for banks lies in balancing model performance with transparency. Highly complex machine learning systems may generate strong predictive accuracy but provide limited interpretability. For regulated financial institutions, the ability to explain a credit decision or pricing outcome to supervisors and customers remains essential.

As a result, many banks are adopting hybrid approaches that combine machine learning with rule-based systems or simplified explanatory layers. These approaches allow institutions to retain the predictive power of advanced models while maintaining regulatory compliance.

Board oversight is also expanding. Senior management committees increasingly review AI deployments alongside traditional technology investments. Model risk committees assess potential bias, data quality and operational resilience. Internal audit functions evaluate whether governance processes remain effective as AI deployments expand.

In this environment AI becomes subject to the same institutional discipline as other core banking activities. Innovation is permitted, but only within clearly defined governance frameworks that ensure accountability.

Data infrastructure, ecosystem platforms and competitive advantage

AI in banking ultimately depends on data architecture. Without consistent data pipelines, integrated systems and scalable infrastructure, even the most sophisticated algorithms cannot operate effectively.

For this reason, many banks now describe AI deployment as inseparable from broader data modernisation programmes. Data lakes, cloud-based infrastructure and application programming interfaces (APIs) are becoming essential components of modern banking architecture.

JPMorgan’s technology investment strategy illustrates this dynamic. Much of the firm’s technology spending is directed not only towards AI models but towards infrastructure capable of supporting large-scale data processing. Modernised application layers and integrated data platforms enable machine learning systems to operate across trading, risk management and client services.

BNY faces similar requirements in its custody and payments businesses. The bank processes enormous volumes of financial transactions daily across multiple currencies and jurisdictions. AI-driven analytics can only function effectively when underlying data flows are reliable and standardised. As a result, infrastructure modernisation remains central to the institution’s strategy.

Digital-native banks demonstrate the advantages of building such infrastructure from the outset. Nu Holdings designed its architecture around unified data pipelines and automated decision systems, allowing the bank to scale rapidly without the constraints of legacy systems.

Traditional banks must therefore bridge the gap between historical infrastructure and modern data environments. This process often involves extensive system integration projects, migration of legacy databases and development of real-time analytics capabilities.

The competitive implications are significant. Institutions with integrated data platforms can deploy AI across multiple business lines simultaneously. Customer behaviour data collected in retail banking may improve risk assessment in lending or enhance liquidity forecasting in treasury operations.

Data ecosystems also extend beyond individual institutions. Partnerships with payment networks, e-commerce platforms and financial technology companies increasingly provide additional data sources. These ecosystems allow banks to develop more granular risk models and offer personalised financial services.

However, ecosystem integration also introduces new governance challenges. Data privacy regulations, cybersecurity risks and cross-border data transfer restrictions require careful management. Institutions must ensure that data sharing arrangements comply with regulatory standards while preserving customer trust.

As AI adoption expands, the ability to manage large-scale data ecosystems will become a critical competitive differentiator. Banks that can integrate data across platforms, jurisdictions and business lines will possess stronger analytical capabilities and greater operational flexibility.

In contrast, institutions constrained by fragmented legacy systems may struggle to realise the full potential of AI.

David Vélez, Founder and Chief Executive Officer, Nu Holdings

Tan Su Shan, Chief Executive Officer, DBS

Agentic AI and the emerging structure of banking competition

As AI becomes increasingly embedded across banking infrastructure, a further development is beginning to shape the next stage of technological adoption: the emergence of agentic AI. While earlier waves of AI focused primarily on prediction or content generation, agentic systems are designed to perform sequences of tasks autonomously within defined operational boundaries.

In banking environments this distinction is increasingly significant. Predictive AI models estimate credit risk or detect fraud. Generative AI systems assist with writing code, summarising documents or responding to customer enquiries. Agentic AI extends this capability by coordinating multiple steps within a workflow, retrieving information, executing tasks and adapting actions as conditions change.

Several financial institutions are beginning to experiment with such models. Instead of using AI solely as an analytical tool for employees, banks are deploying digital agents capable of performing routine operational tasks across technology, compliance and service environments. These agents can retrieve internal knowledge, review policy rules, generate responses and trigger subsequent processes within the same operational chain.

In software development, AI agents may retrieve code libraries, generate draft modules, conduct testing and identify potential errors before handing results to engineers. Within customer service functions, similar systems can gather account information, interpret client requests and propose solutions before escalating complex cases to human staff. In risk operations, AI agents may assemble relevant documentation, evaluate data inputs and prepare decision recommendations for approval.

The productivity implications are considerable. Many banking activities involve sequences of repetitive analytical tasks and information retrieval. Agentic systems allow institutions to automate significant portions of these workflows, reducing coordination friction and accelerating operational throughput.

However, the emergence of agentic AI also reinforces a broader shift within the industry. As banks move from isolated AI experiments to industrial-scale deployment, competitive divergence is becoming more visible between institutions able to embed AI deeply into operating models and those still adopting incremental tools.

Large universal banks possess structural advantages in this transition. Their scale allows them to invest heavily in engineering capabilities, data infrastructure and model governance. They can deploy AI across multiple operating layers simultaneously, from internal productivity and risk analytics to client engagement and transaction processing.

This scale matters because AI increasingly benefits from network effects. Machine learning systems trained on large and diverse datasets generally produce more reliable results than models operating within narrow information environments. Banks processing vast volumes of transactions and client activity therefore possess powerful data advantages when training and refining AI systems.

Digital-native institutions demonstrate a different but equally influential model. Firms such as Nu Holdings were built around unified data architectures and automated decision systems from inception. Without the constraints of legacy infrastructure, these organisations can deploy AI rapidly across underwriting, customer interaction and operational workflows.

The result may be a gradual restructuring of the banking landscape. On one side will be large institutions capable of integrating AI into complex global franchises spanning payments, capital markets, lending and wealth management. On the other will be digital institutions optimised for high-frequency decisioning and highly automated customer interaction.

Agentic AI may accelerate this divergence. Institutions capable of deploying autonomous operational systems across multiple business lines can achieve substantial productivity gains and faster decision cycles. Banks constrained by fragmented infrastructure or limited engineering capacity may struggle to capture the same benefits.

Governance therefore becomes even more critical. When AI systems begin executing operational tasks rather than simply generating recommendations, institutions must establish clear boundaries around autonomy. Rules governing which actions can be performed independently, which require human approval and which remain restricted must be embedded into system design.

Most banks therefore emphasise controlled autonomy. Agentic systems may operate independently within predefined parameters but remain subject to monitoring, audit trails and override mechanisms. Human oversight continues to play a central role, particularly in decisions involving credit approval, financial advice or regulatory compliance.

The emergence of agentic AI therefore illustrates a broader transition in banking technology. AI is moving from analytical assistance toward operational execution. As this shift unfolds, technological capability and institutional governance must evolve together.

For banking leaders, the challenge is no longer simply adopting AI but integrating increasingly autonomous systems into accountable operating models that preserve financial discipline, regulatory trust and client confidence.

China’s large-scale AI laboratories in banking

While many global banks are still expanding AI deployment incrementally, institutions in China have begun operating AI systems at large industrial scale. Both large state-owned banks and digital-native institutions are embedding AI across customer service, risk analytics, software development and internal operations.

Among traditional banks, Industrial and Commercial Bank of China (ICBC) is a useful example of how large incumbent institutions are incorporating AI into complex universal banking structures. With a vast retail and corporate franchise, ICBC has been developing enterprise AI platforms designed to support risk management, operational automation and customer service at scale. These systems integrate machine learning models, internal knowledge bases and governance frameworks that allow AI applications to be deployed across multiple business units. In this sense, ICBC represents the trajectory of the traditional large-bank model: AI layered into an existing financial system with significant scale, institutional complexity and governance requirements.

Among digital banks, WeBank illustrates a different path. As the online bank backed by Tencent, it was built around cloud infrastructure, data analytics and automated decisioning from the outset. Its operating architecture relies heavily on AI-driven credit assessment, automated service platforms and machine learning models capable of processing extremely high volumes of transactions and customer interactions. WeBank therefore represents the digital-native model, in which AI is not added onto legacy infrastructure but embedded into the institution’s design from inception.

Digital banking platforms serving hundreds of millions of users require systems capable of operating reliably under extreme transaction volumes. In response, several Chinese institutions have begun redesigning workflows around AI-native operating models in which machine learning systems participate directly in development processes, knowledge retrieval and operational decision support.

More recently, some banks and financial technology firms have begun experimenting with agentic AI architectures, where specialised digital agents coordinate sequences of operational tasks. Instead of merely generating recommendations or responses, these systems can retrieve information, interpret internal policies and execute predefined actions within controlled parameters.

In practice, this may involve AI agents assembling customer information, retrieving risk policies, preparing documentation or triggering operational processes before escalating final decisions to human staff. Within technology teams, similar agents can retrieve code libraries, generate draft software modules and perform automated testing as part of development workflows.

Credit underwriting for digital enterprises illustrates another dimension of this shift. Rather than relying solely on traditional financial statements, lenders increasingly analyse operational data such as inventory movements, sales flows and payment activity across digital platforms. These datasets allow institutions to construct dynamic risk profiles reflecting real-time business activity.

AI is also reshaping banking software development. Development teams increasingly use AI-assisted tools for coding, testing and deployment, allowing projects to progress more rapidly while maintaining governance standards required in regulated financial environments.

Despite the scale of experimentation, governance remains central. AI systems typically operate within defined operational boundaries, with human oversight required for sensitive decisions such as credit approval or financial advice. Institutions emphasise controlled autonomy, ensuring that automated systems remain subject to monitoring, audit trails and override mechanisms.

The scale of China’s digital ecosystems accelerates learning cycles. Large datasets and integrated platforms spanning payments, e-commerce and logistics provide extensive operational information for risk modelling and service innovation.

Taken together, institutions such as ICBC and WeBank illustrate two complementary paths through which AI is reshaping banking in China: one led by traditional banks integrating AI into existing financial infrastructure, and the other by digital banks building AI-driven operating models from the ground up.

These developments demonstrate how AI in finance is evolving from isolated analytical tools into integrated operating systems capable of supporting banking services at massive scale.

The accountable intelligent bank

AI is now embedded across underwriting engines, liquidity models, fraud analytics and client interaction systems. Its strategic significance lies not in algorithmic sophistication but in whether institutions can integrate machine-assisted decisioning into capital discipline, regulatory alignment and competitive strategy.

Executives across global banking increasingly emphasise that innovation must coexist with prudential safeguards. Infrastructure becomes strategic only when it improves measurable outcomes such as operating leverage, risk-adjusted returns and balance sheet resilience.

Institutions that industrialise AI effectively may achieve structurally lower operating costs, more precise risk pricing and faster operational throughput. Those that rely on fragmented legacy systems may struggle to match these efficiencies.

The intelligent bank at scale is therefore defined less by technology than by leadership discipline. It requires explicit return thresholds for technology investment, strong governance frameworks and board-level literacy in model risk.

Scale amplifies both advantage and fragility. Institutions that align AI deployment with capital discipline and supervisory expectations will strengthen resilience. Those that pursue experimentation without governance risk eroding trust.

In that balance between capability and control lies the architecture of the accountable intelligent bank.