📊 HiddenMerit Morning Post · Issue 70

Focus on Database Frontiers, Practical Insights for DBAs August 7, 2026 ｜ 5 Selected Global Breaking News

01｜Kingware Assesses Database Architecture Over the Next Three Years: Vector Capability “Internalisation,” AI‑Native Convergence Replaces “Database + Vector Library” Dual‑Track

CETC Kingware recently published a technical article titled “When AI Meets Databases: Architecture Evolution Directions for the Next 3 Years,” systematically assessing the future evolution path of database architecture. The core judgment is that in the coming years, databases will no longer merely be “warehouses” for storing structured data, but will become “brains” capable of carrying AI reasoning and knowledge retrieval.

Trend Assessment: The traditional “database + vector library” separated architecture has shown architectural limitations in data synchronisation latency and transaction isolation when handling scenarios with concurrent high‑frequency writes and real‑time queries. The proportion of unstructured data such as text, images, and audio is surging and is often tightly coupled with core business data. Simple JSON storage or external indexes can no longer meet the need for deep semantic mining of data.

Evolution Path: The core direction proposed by Kingware is to build a “storage‑compute integrated” and “semantic‑native” converged data platform. KingbaseES V9 introduces a native vector data model, allowing direct storage, indexing, and retrieval of high‑dimensional vector data, operating independently of external vector engines. At the same time, it achieves deep integration of SQL and vector queries – architects can write a single statement to simultaneously complete structured data filtering and vector similarity search, effectively solving data silo issues. Coupled with a data integration platform, it ensures low latency and high consistency across the full chain from data generation to AI invocation.

• DBA Perspective: Vector capability “internalisation” means DBAs no longer need to maintain data synchronisation and consistency between relational databases and independent vector engines. The architectural evolution direction proposed by Kingware – a paradigm shift from “relational storage” to “semantic computing” – requires DBAs in Xinchuang projects to re‑examine data models, shifting from storage format optimisation to cross‑domain data view orchestration and governance. KingbaseES V9 supports hybrid queries via standard SQL, allowing DBAs to handle both structured and vector data within a single transaction. DBAs are advised to deploy KingbaseES V9 in a development environment to test its vector capabilities and accumulate practical experience in hybrid query tuning.

02｜Google Launches Two AI Database Agents: Natural Language Takes Over Full Database Lifecycle Management

On August 4, Google officially launched two AI database agents – Database Onboarding Agent and Database Observability Agent – as the latest implementation of its Agentic Data Cloud strategy.

Onboarding Agent: Users simply describe workload requirements in natural language (performance needs, data scale, data types, reliability requirements, etc.), and the agent recommends the best‑matched managed database service (AlloyDB, Cloud SQL, Spanner, Bigtable, etc.) and generates complete configuration and deployment commands. The agent also understands technical metrics such as IOPS, latency thresholds, and replication lag, and is designed to “put database guidance in chat, console, and developer interfaces.”

Observability Agent: Focused on daily operations, integrating telemetry data from Database Insights, Cloud Monitoring, Cloud Logging, and Cloud Trace. Users can ask natural language questions like “Which databases had the highest CPU usage in the last hour?” or “Why is there lock contention?” The system automatically identifies root causes and provides remediation recommendations, with automatic execution after approval in some scenarios. The observability agent covers Google’s full range of managed database products, including AlloyDB, Bigtable, Cloud SQL, Spanner, Firestore, and Memorystore.

• DBA Perspective: Google’s AI database agents are another milestone in “intelligent operations” for cloud databases. Unlike Tencent Cloud’s DatabaseClaw approach of “distilling hundreds of thousands of DBA work orders into Skills,” Google has chosen a technological path of “natural language + full‑stack telemetry fusion.” For DBAs, routine monitoring, slow query analysis, capacity assessment, and other repetitive tasks will gradually be replaced by AI agents. The role is evolving from “manual operations” to “AI agent policy manager” – defining AI operational boundaries, auditing execution traces, and triggering circuit breakers during anomalies. The observability agent’s root cause analysis capability is the core indicator of whether AI truly “understands” the operational state of databases.

03｜ClickHouse Labs Established: CMU Professor Andy Pavlo Appointed VP of Research, Focusing on AI‑Era Database Architecture

On August 4, open‑source real‑time analytics database ClickHouse announced the establishment of ClickHouse Labs and appointed Andy Pavlo, Associate Professor of Computer Science at Carnegie Mellon University (CMU) and head of the CMU Database Research Group, as Vice President of Research.

Pavlo is a landmark figure in the database academic community, having received the IEEE TCDE Database Education Award, the VLDB Early Career Award, the NSF CAREER Award, the Sloan Research Fellowship, and the ACM SIGMOD Jim Gray Best Paper Award. ClickHouse Labs will focus on foundational research in database architecture, query processing, system performance, and infrastructure, with research outcomes benefiting both the ClickHouse and PostgreSQL open‑source projects.

Pavlo stated: “The most exciting database research doesn’t just put new ideas into papers, but proves their feasibility by building and testing real systems.” He explicitly plans to research the integration of AI and databases – “how DBMS adapt to AI and agent technologies,” including how databases can better support agents, and how agents can automate and improve DBMS development itself.

ClickHouse users include Sony, Tesla, Anthropic, Lyft, and Instacart.

• DBA Perspective: Andy Pavlo’s appointment is another significant milestone following his co‑founding of OtterTune (applying machine learning to database optimisation). For DBAs, this event means: the boundaries between academia and industry are further dissolving – CMU’s academic expertise will directly translate into production‑grade optimisations for ClickHouse; open‑source databases are building trust through “open research” – research outcomes will be publicly released, allowing external developers to review, test, and extend them. Pavlo’s research direction – “how DBMS adapt to AI and agent technologies” – is a core proposition for the database industry over the next 3‑5 years. DBAs should pay attention to ClickHouse Labs’ subsequent research publications – they are important references for assessing ClickHouse’s long‑term technology direction.

04｜Dameng Shifts Focus to Next‑Generation All‑in‑One: Hardware‑Software Integration for Financial and Energy Core Scenarios

On August 5, Dameng stated during an investor research visit that after achieving breakthroughs in key core technologies such as shared storage clusters, its strategic focus is gradually shifting toward the next‑generation database all‑in‑one business. Through deep integration and collaborative optimisation of hardware and software, the product aims to deliver higher performance, easier deployment, and higher availability data infrastructure, precisely matching the dual demands of autonomous control and extreme performance in key industries such as finance, energy, and telecommunications.

Dameng stated it will focus on building a data foundation for the AI era, continuously expanding the application boundaries and market space of database products, driving domestic databases from “usable” to “good‑to‑use” through technological innovation, and comprehensively empowering digital transformation across industries.

• DBA Perspective: Dameng’s strategic shift toward all‑in‑one products confirms the trend of domestic databases moving from “software replacement” to “hardware‑software synergy.” For DBAs, the all‑in‑one delivery model means the barrier to database deployment and tuning will be lowered – pre‑integrated and pre‑tuned hardware and software reduce the DBA’s workload in hardware selection and parameter configuration. However, it also requires DBAs to have a holistic view of hardware‑software synergy, understanding the integrated tuning logic of storage, networking, and compute. The performance data of Dameng’s previously released DAMENG PAI V2.0 all‑in‑one – I/O latency reduced from 400μs to 80μs, with IOPS starting at 12 million – provides a quantifiable reference for selection in core scenarios such as finance and energy.

05｜Domestic Database Panorama: Kingware Leads Xinchuang Landscape, Healthcare and Transportation Sectors Lead in Sales Volume

CETC Kingware recently published a series of articles including “Domestic Database Panorama: Kingware Database Leads the Xinchuang Landscape” and “Xinchuang Database Panorama,” systematically outlining the market landscape and selection framework for domestic databases.

Market Landscape: Domestic databases have moved from “single‑point breakthroughs” to “full‑stack collaboration.” The market is no longer simply about competition between individual products, but a contest of ecosystem and industrial synergy capabilities. According to CCID Consulting data, CETC Kingware leads domestic vendors in sales volume in the healthcare and transportation sectors, with cumulative deployments reaching the million‑set level, and continues to lead in sales share in critical domestic database application areas.

Selection Framework: Kingware recommends establishing a selection evaluation framework from four dimensions: “architectural capability,” “business scenarios,” “security compliance,” and “ecosystem compatibility,” rather than focusing solely on vendor branding. In terms of security compliance, Kingbase has passed the IT Product Information Security Certification EAL4+ enhanced level, and is one of the first domestic databases to pass the National Information Security Center’s security and reliability assessment.

Industry Penetration: In a tertiary hospital Xinchuang transformation project, Kingbase collaborated with Hygon domestic chips and Kylin operating systems to complete adaptation for dozens of core business systems including HIS, EMR, PACS, and LIS, achieving full‑chain domestic replacement of business systems. In the government sector, a provincial market supervision project achieved “zero interruption” of business and “zero loss” of data.

• DBA Perspective: Kingware’s leading sales volume in healthcare and transportation provides DBAs with a quantitative reference for selection in cross‑industry Xinchuang projects. The “four dimensions” recommended in the selection framework – architectural capability, business scenarios, security compliance, and ecosystem compatibility – are also the evaluation logic DBAs should follow when assessing domestic databases. The full‑stack adaptation cases of healthcare core systems such as HIS, EMR, and PACS provide a reference paradigm for healthcare DBAs in domestic replacement. Kingware’s cumulative million‑set deployment scale also means that Kingware skills have high reusability across cross‑industry Xinchuang projects.

📚 SQL Little Knowledge Point

This Issue’s Knowledge Point: What is Database “Vector Internalisation”?

“Vector internalisation” is the database architecture evolution direction proposed by Kingware – integrating vector data storage, indexing, and retrieval capabilities directly into the database kernel, rather than providing support through external vector engines or plugins.

“External Vector Library” vs. “Vector Internalisation” :

Dimension	External Vector Library	Vector Internalisation
Architecture	Relational DB + independent vector library (dual‑track)	Single converged database
Data Consistency	Cross‑library sync with latency	Native ACID guarantee
Query Capability	Requires stitching results across systems	Single SQL hybrid query
Operations Complexity	Two systems independently maintained	Unified management
Typical Scenarios	Early‑stage RAG applications	AI production‑grade applications

Trend for the Next 3 Years: Kingware assesses that as AI applications demand higher real‑time performance and consistency, converged databases capable of handling both OLTP and vector retrieval will become mainstream. Architects will increasingly prefer “store once, reuse multiple times” solutions rather than maintaining multiple heterogeneous systems. KingbaseES V9 has natively integrated the vector data model, supporting vector retrieval without third‑party plugins, with ACID support – vector data and structured data are safely persisted within the same transaction.

HiddenMerit Team Production Slogan: 绩优隐于内，金石启新程 | Hidden deep. Merit bold. Forge ahead.