Ping An Scales AI Operations With 300 Billion Daily Token Consumption

The Group’s transition relies on a dual-pronged "integrated finance and health" strategy, supported by a massive infrastructure expansion. By nearly doubling its computing capacity from 800 PF to 1,500 PF, Ping An achieved a tenfold increase in token output. This technical scaling was paired with a 368% boost in token output per unit of power and a 76% reduction in per-token costs, proving that enterprise AI can remain economically viable at scale.

Practical application remains the primary metric for the firm. The company’s Express Service, which allows users to execute complex financial tasks through natural language prompts, has processed over 100 million interactions in four months. The system boasts a 92% end-to-end resolution rate, effectively halving the time required for traditional service models. Similarly, the AI Doctor platform is now the first point of contact for a quarter of in-hospital services, with diagnostic assessments matching expert conclusions in 91% of breast and gastric cancer cases.

Internal productivity has seen a comparable shift, with AI now accounting for approximately 80% of the firm's total software code output. Beyond development, autonomous systems handled 81% of customer service volume in the first half of the year, while fraud detection measures saved the company RMB7.11 billion in claims. Wang emphasized that while general-purpose models are evolving, the future of corporate value creation lies in domain-specific intelligence—an integrated framework of proprietary data and rigorous production-grade quality controls.