JPMorgan to rein in AI spend as it hires more Singapore tech talent

[SINGAPORE] Artificial intelligence has taken the world by storm in many sectors, but the costs are mounting.

At the heart of the squeeze are “tokens” – the fundamental units of computing power required to run AI models.

While the per-token price has dropped across the industry, the sheer volume consumed by advanced reasoning models and autonomous AI agents is making corporate technology budgets increasingly unpredictable.

The financial industry is no exception, with many players using AI agents.

JPMorgan expects the financial industry to adopt stricter internal cost controls for generative AI as global banks reckon with its mounting expenses.

The bank is now reining in its own costs, pivoting from an era of unchecked experimentation to strict internal budgeting as the technology’s computing expenses pile up. It also plans to spend more on hiring the right talent to build AI capabilities.

“The days where everybody could just do anything are probably over,” Max Neukirchen, co-head of JPMorgan Global Payments, said during a June interview in Singapore. He noted the bank is now managing the AI prompt generation with “a lot more discipline and thoughtfulness”.

Matching AI costs to projects’ upsides

To achieve this, the bank is evaluating how to match computing costs directly to a project’s financial upside.

Under the new framework, routine administrative tasks like summarising daily news articles will be restricted to standard, lower-cost language models. Expensive, premium computational tokens will be reserved for complex, revenue-adjacent functions instead.

This includes deploying real-time corporate treasury agents that forecast cash flows, automate document exchanges and manage global liquidity for thinly staffed corporate clients.

The push for efficiency comes as the enterprise integration of AI simultaneously alters internal banking workloads. While the technology reduces manual labour in treasury operations, it is quickly inflating workloads and expenses in risk and security.

To that end, JPMorgan has been deploying substantial resources to combat increasingly sophisticated, AI-driven fraud. The bank is also a key partner in Anthropic’s Project Glasswing.

The initiative grants JPMorgan access to Claude Mythos 5, a highly restricted, state-of-the-art cybersecurity AI model available only to a vetted group of trusted organisations after recent US national security reviews.

Expanding its Singapore hub

Preventing AI-driven fraud requires heavy investment in human capital, with JPMorgan hiring engineers and data architects to build out its proprietary AI capabilities and forecasting models in-house.

A large part of this talent acquisition is happening in Singapore, a primary hub for the bank’s Asia-Pacific operations, according to Neukirchen.

This localised hiring strategy is supported by the regulatory environment overseen by the Monetary Authority of Singapore.

Neukirchen praised the central bank for providing clear frameworks and “pro-innovation” oversight that allows the bank to experiment and scale new solutions within compliance parameters.

Beyond AI, JPMorgan is also scaling parallel digital infrastructure out of Singapore. Its blockchain and digital currency platform, Kinexys, has processed over US$4 trillion in transaction volumes, with average daily transactions exceeding US$7 billion.

This figure is expected to grow as multinational corporations seek 24/7 cross-border liquidity management.