Breaking the AI‑Trusted Storage Barrier: Commercial Banks’ AI+Domestic‑Innovation Practices

1. Dual Strategic Demands: AI+Domestic‑Innovation

Rapid advances in AI, especially large‑model platforms such as DeepSeek and ChatGPT, are driving financial institutions to integrate AI into core services. At the same time, China’s “Digital China” policy and the State‑owned Assets Supervision and Administration Commission’s 2022 “79‑document” require state‑owned enterprises to achieve full replacement of operating systems, databases, and CRM tools with domestic (信创) solutions by the end of 2027. This policy push has accelerated the migration of banking IT infrastructure to domestic distributed databases.

Faced with these twin pressures, banks need an architecture that can both satisfy AI data‑governance requirements (high‑performance, secure data ingestion) and ensure the reliability of large‑scale domestic‑technology deployments.

2. Four Core Challenges of the Existing Architecture

The prevailing “server‑local disk multi‑replica + domestic distributed database” setup suffers from:

Data‑ingestion performance and security bottlenecks : Sensitive transaction data must be safely and efficiently streamed into AI data lakes; tight compute‑storage coupling and multi‑replica local disks hinder both speed and security.

Data consistency across distributed nodes : Network latency, node failures, and message loss cause state divergence, degrading model training accuracy and inference reliability.

Excessive data exposure surface : Co‑located compute and storage expose large volumes of sensitive data, violating the principle of minimal data exposure.

Low storage utilization and difficult scaling : Multi‑replica strategies waste space (effective capacity 1/2 or 1/3 of raw) and make capacity expansion complex, threatening business continuity.

Survey results from the TWT community show that 81 % of respondents consider the “professional SAN storage + domestic distributed database” architecture strategically necessary, with stability, performance, and high availability identified as the top selection criteria.

3. Data‑Productivity Boost via Compute‑Storage Separation

Integrating professional SAN storage (e.g., FlashNexus) with domestic distributed databases creates a true compute‑storage‑separated architecture that addresses the above challenges:

Data readiness : FlashNexus delivers sub‑millisecond latency, enabling “second‑level” synchronization of transaction data to AI data lakes, thus eliminating AI‑pipeline bottlenecks.

Strong consistency : Enterprise‑grade data protection and multi‑control mechanisms guarantee exact data replication, ensuring trustworthy AI decisions.

Architectural agility : Independent scaling of compute and storage lets banks provision hundreds of terabytes on demand, reducing AI model training turnaround time.

Ecosystem collaboration : FlashNexus has been jointly validated with leading domestic distributed databases (e.g., CirroData, GBase, PingKai), forming an “AI‑ready data platform” that meets financial‑core performance and stability requirements.

4. Advanced Storage Capabilities Supporting Financial Digital Transformation

FlashNexus, a centralized all‑flash SAN, demonstrates:

Extreme performance : Over 2 million IOPS in a three‑chassis 25‑disk configuration, satisfying most transaction‑processing workloads.

Active‑active disaster recovery : Gateway‑less A‑A dual‑site design provides seamless cross‑region failover and supports synchronous/asynchronous replication for multi‑data‑center deployments.

Ease of use : Single‑node configuration of disaster‑recovery features, visual management modules, and support for multiple protocols (TCP, RoCE, FC) simplify operations.

Benchmark tests with CirroData showed microsecond‑level response times and billions of IOPS; tests with GBase and PingKai demonstrated a 20 % reduction in transaction latency and a 30 % increase in batch processing efficiency.

These results illustrate a viable path for banks to achieve trustworthy AI storage, improve data productivity, and sustain high‑availability digital services.

5. Outlook

Adopting a true compute‑storage‑separated, trusted AI storage infrastructure is positioned as a critical step for Chinese financial institutions to meet both AI acceleration and domestic‑technology (信创) deepening goals. Continued collaboration between storage vendors and database providers is expected to further close performance gaps and solidify the domestic ecosystem.