JPMorgan tightens AI spending while expanding Singapore tech hiring

JPMorgan expects banks to impose stricter internal cost controls on generative AI as rising token usage makes technology budgets harder to predict. Max Neukirchen, co-head of JPMorgan Global Payments, said the bank is applying more discipline to AI prompt generation after a period when employees could experiment more freely.

The bank is linking AI computing costs more closely to each project’s expected financial upside. Routine work such as summarising daily news will use standard, lower-cost language models, while premium computational tokens will be reserved for more complex uses, including real-time corporate treasury agents that forecast cash flows, automate document exchanges and manage liquidity for corporate clients.

AI is also increasing spending in risk and security as JPMorgan responds to more sophisticated AI-driven fraud. The bank is a partner in Anthropic’s Project Glasswing, giving it access to Claude Mythos 5, a restricted cybersecurity AI model available to vetted organisations after recent US national security reviews.

Singapore is central to JPMorgan’s hiring push, with the bank recruiting engineers and data architects to develop proprietary AI capabilities and forecasting models. JPMorgan is also expanding Kinexys, its blockchain and digital currency platform, which has processed over US$4 trillion in transaction volumes and averages daily transactions exceeding US$7 billion as companies seek 24/7 cross-border liquidity management.