31% of Remote Banks Complete Large Model Deployment – Is the Explosion of AI Banking Applications Beginning?

A representative from Bank of Shanghai told Caixin that its AI mobile banking now supports "conversation-as-a-service," allowing users to complete high-frequency transactions such as account management and financial consultations via voice or text commands. Previously, banks including ICBC and Bank of Beijing also disclosed using AI to enhance mobile banking, offering customized wealth management solutions for different users.

A source from a major state-owned bank revealed to Caixin that the bank has recently built an Agent platform and is advancing related projects. The insider noted, "Industry expectations for OpenAI were high, but foundational models like ChatGPT fell short in practical applications. Once these models mature, Agent applications may enter an explosive growth phase."

"Whether in front, middle, or back offices, banks are expanding AI application scenarios, but the gap between different banks is widening," said a senior executive from a leading financial cloud provider, adding that computing power and talent investments are key differentiators.

Many banks are ramping up investments in computing power and related infrastructure. For instance, SPD Bank placed a 100-million-yuan order for its "2025 Large Model Computing Expansion and Kunpeng Computing Management Project." Meanwhile, joint-stock and city commercial banks have also launched model-related procurement projects exceeding 5 million yuan, primarily focused on computing power and GPU servers.

However, challenges remain in AI innovation for banks. On one hand, technical teams still rely heavily on modifying open-source solutions, with a shortage of professionals who understand both business and implementation. On the other hand, complex internal processes—extensive reviews, technical audits, and security checks—slow innovation due to strict compliance requirements.

Zeng Gang, Chief Expert at the Shanghai Finance and Development Laboratory, told Caixin that banks’ traditional IT architectures, designed for stability, clash with AI’s agile iteration. Core system overhauls are high-risk and costly, while integrating legacy and new systems faces technical barriers.

"While large models excel in efficiency, deployment hurdles persist," said Wang Pengbo, Senior Analyst at Broadcom Analysis. "Their search accuracy, contextual understanding, and multi-step reasoning reliability still need improvement, posing risks for finance-grade precision. Additionally, the massive costs of hardware like GPUs create high entry barriers for smaller banks."