JPMorgan Shifts From AI Experimentation to Budget Discipline in Singapore

Token Economics Force Strategic Shift

JPMorgan has moved to impose internal spending controls on its generative AI operations, ending a phase of open-ended testing as token consumption drives up technology budgets across the financial sector.

The bank now allocates computing resources based on expected financial return, according to Max Neukirchen, co-head of JPMorgan Global Payments, who spoke in Singapore in June. Simple tasks such as summarizing internal documents will run on less expensive models, while high-cost computational capacity is reserved for revenue-generating applications.

"The days where everybody could just do anything are probably over," Neukirchen said, noting that the bank now applies "a lot more discipline and thoughtfulness" to how it generates AI prompts.

The shift reflects a broader recalibration underway in enterprise AI adoption. While per-token pricing has declined industry-wide, the volume consumed by advanced reasoning systems and autonomous agents has made budgets harder to predict. Financial institutions deploying these tools are now weighing output against input cost at a granular level.

Revenue-Adjacent Functions Take Priority

JPMorgan is directing premium computing power toward applications tied to client service and operational efficiency. The bank is building real-time treasury agents that forecast cash positions, automate document workflows, and manage liquidity for corporate clients operating with lean finance teams.

Administrative functions that do not directly contribute to revenue or client outcomes will rely on standard language models with lower token costs.

This tiered approach marks a departure from the initial wave of AI deployment, when experimentation was encouraged across departments without strict budget oversight.

Fraud Defense Drives Hiring Push

The bank's efficiency drive on one side of its AI ledger is offset by rising costs in security and risk management. JPMorgan is deploying resources to counter AI-enabled fraud, which has grown more sophisticated as attackers gain access to the same generative tools.

The bank is a partner in Anthropic's Project Glasswing, which grants access to Claude Mythos 5, a restricted cybersecurity model available only to vetted organizations following recent U.S. national security reviews.

Building defenses against AI-driven threats requires specialized engineering talent. JPMorgan is hiring data architects and engineers to develop proprietary models and forecasting systems in-house, with Singapore serving as a key location for this expansion.

Singapore as Regional AI Hub

The bank's Asia-Pacific operations are concentrated in Singapore, where it is adding headcount to support both AI development and broader digital infrastructure. Neukirchen praised the Monetary Authority of Singapore for regulatory clarity and a pro-innovation stance that allows controlled experimentation within compliance boundaries.

The city-state's regulatory framework has made it a preferred location for financial institutions testing new technologies while managing risk. JPMorgan's decision to localize talent acquisition reflects confidence in Singapore's ability to supply the technical workforce needed for advanced AI and blockchain projects.

Beyond AI, the bank is scaling Kinexys, its blockchain and digital currency platform, from Singapore. The platform has processed over USD 4 trillion in transaction volume, with daily transactions exceeding USD 7 billion. Neukirchen expects that figure to grow as multinational corporations seek round-the-clock cross-border liquidity management.

Industry-Wide Cost Discipline Ahead

JPMorgan's internal budget controls may signal a broader trend across the financial sector. As banks move from proof-of-concept to production-scale AI, the economics of token consumption are forcing a more disciplined approach.

Institutions that initially treated AI as a low-cost productivity tool are now confronting the reality that advanced models carry significant ongoing expenses. Matching those costs to measurable business outcomes is becoming a priority for technology leaders and CFOs alike.

The bank's strategy in Singapore suggests that cost discipline and capability building are not mutually exclusive. By hiring locally and focusing on high-value use cases, JPMorgan is positioning itself to sustain AI investment over the long term without the budget volatility that has plagued early adopters.