How ICBC’s AI‑Native Data Center Is Redefining Cloud Computing for Finance

In the era of large‑model AI reshaping industries, China’s computing power industry is rapidly scaling to a trillion‑level magnitude. Leveraging this AI‑driven intelligence leap, the Industrial and Commercial Bank of China (ICBC) Data Center has broken the conventional data‑center framework with an “AI‑native” mindset, upgrading its financial computing infrastructure. The intelligent computing center now transcends mere compute provision, becoming the core hub that supports a smart‑finance ecosystem and propels the financial sector toward a cognitive‑intelligence stage.

1. Architecture Elevation: Building an AI‑Native Computing Base

Driven by large‑model AI, traditional data centers are shifting from “general compute” to “AI‑native intelligent computing.” This transformation requires not only hardware redesign but also a systematic upgrade of the compute foundation to meet the intensive training and inference demands of massive models, enabling efficient scheduling and intelligent management of resources. ICBC’s data center is advancing AI chips, servers, storage, networking, and facility upgrades to construct a full‑stack AI‑native compute base capable of supporting trillion‑parameter models, markedly improving training efficiency and resource utilization for intelligent financial services.

Innovative breakthroughs have enabled autonomous, controllable infrastructure. As model compute needs grow exponentially, CPU‑centric architectures fall short. The new center adopts a heterogeneous system centered on AI chips, combining CPUs with AI accelerators to boost compute density. ICBC has achieved full‑stack autonomy across AI chips, high‑performance networks, storage, servers, and training frameworks, establishing a large‑scale distributed AI compute cluster that supports trillion‑parameter training.

Key technologies unify heterogeneous resources for intelligent scheduling. By integrating compute‑network fusion and cloud‑native techniques, the center builds a cross‑regional compute network for dynamic resource allocation. Advances in lossless high‑performance networking, deadlock avoidance, high‑speed interconnects, dynamic load balancing, compute‑network collaboration, and tiered storage enable zero‑loss, low‑latency, high‑throughput AI workloads, with RoCE networks managed under a unified cloud‑network paradigm.

The architectural upgrade hinges on “hardware innovation + software definition + ecosystem collaboration,” paving the way for future compute‑in‑memory and photonic communication technologies, and fostering an open, autonomous ecosystem.

2. Green Leap: Decoding Low‑Carbon Practices for Financial Computing

In the digital economy, compute infrastructure is the engine driving financial transformation. Aligning with national “dual‑carbon” goals, ICBC’s data center embeds green principles throughout planning, construction, and operation, establishing a three‑dimensional carbon‑neutral evolution system covering technology, management, and ecosystem.

Breaking the traditional “build‑then‑retrofit” model, the center incorporates green design from the outset, employing liquid cooling, natural cooling, UPS sleep modes, and intelligent lighting, alongside comprehensive environmental monitoring and a decreasing PUE metric. These measures earned the Jia‑ding campus the first national “green data center” certification, setting an industry benchmark.

Strategically, the center adopts a multi‑site approach aligned with the “East‑Data‑West‑Compute” initiative, integrating green technologies with high availability and flexible scaling. This creates a cross‑regional, multi‑layered green compute network that synergizes with regional energy structures and technological trends, offering a replicable transformation pathway for the industry.

3. Digital Twin: Reconstructing Financial‑Grade Operations with AIOps

As financial digital transformation accelerates and compute demand explodes, traditional, human‑driven operations can no longer sustain ultra‑large intelligent computing centers. ICBC’s data center leverages intelligent operations, combining big data and AI to build an integrated intelligent operation system that aligns with distributed IT architecture and long‑term data‑center planning.

Key initiatives include:

Upgrading monitoring systems for comprehensive, business‑oriented, intelligent analysis, enabling global observability and efficient change‑risk management.

Developing a self‑built chaos engineering platform to proactively inject faults, identify and remediate hidden issues, enhancing system stability.

Implementing FinOps for full‑lifecycle digital operation of resources, achieving fine‑grained compute resource management and rapid scaling.

Continuous technology innovation drives the transition toward AI‑augmented services. The center has built an operation data analysis engine and dynamic operation map, applying AI algorithms for fault detection, diagnosis, and performance trend analysis in cloud environments. Projects such as the “Cloud‑Native Intelligent Operations” have earned Level‑3 AIOps maturity certification from the China Academy of Information and Communications Technology.

Explorations of large‑model AI (e.g., DeepSeek) for operation knowledge bases, intelligent operation agents, and tool integration are underway, further advancing automation. After two decades of development, the center has reached L3 (condition‑based automation) and is progressing toward L4 (highly automated) operation stages.

Positioned at the intersection of large‑model AI revolution and digital China, ICBC’s data center aims to become the leading “new‑type compute” provider for the financial industry, integrating compute, algorithms, and data to build a trillion‑level intelligent compute platform that fuels high‑quality economic development and the construction of a financial powerhouse.

Published in "China Financial Computer" 2025, Issue 6.