CMB's Daily Token Use Tops 80 Billion as Banks Scale Up LLM Operations

On November 4, China Merchants Bank told its Q3 2026 earnings briefing that large-model deployments had reached more than 1,500 business scenarios by end-September, with daily token consumption across business scenarios exceeding 80 billion, Cailian Press reported — the first Chinese bank to disclose an 80-billion daily token figure.

Four months earlier the number was less than half that. At a shareholders meeting in June, chief information officer Zhou Tianhong put daily token use at 33 billion as of end-May. Consumption grew to roughly 2.4 times that level in four months, reflecting heavier real-world calling across approvals, customer service and R&D workflows.

CMB is not alone. A review of Q3 filings and briefings by Cailian Press found banks' daily token consumption rising sharply quarter-on-quarter, with some banks nearly tripling; ICBC, Postal Savings Bank of China, Bank of Jiangsu, Ping An Bank and WeBank have each disclosed token figures or LLM progress in recent weeks.

The mix is shifting as well. Industry reports describe banks moving from Q&A-style assistance toward autonomous agents that execute workflows: ICBC has handed part of its customer-facing service to agents, while relationship-manager assistants cover marketing scripts and document drafting. Token counts have become a common yardstick for bank technology investment.

Scaling brings conversion pressure. Cailian Press noted that banking AI has entered large-scale operation but value conversion "will take time" — token growth is running ahead of measurable returns on revenue and cost lines.

When token counts appear in earnings briefings alongside transaction volumes, the question has moved from whether banks use LLMs to how much. Whether those tokens convert into lower risk and new revenue will be settled by the 2026 annual accounts.