HiddenMerit Daily · Issue 49

# 📊 HiddenMerit Daily · Issue 49

> Focus on Database Frontiers, Practical Insights for DBAs

> June 24, 2026 ｜ 5 Selected Global Breaking News

## 01｜Snowflake Defines AI Data Operating System: Databases Evolve from Storage Tools to Agent Foundation

At the Snowflake Summit in June 2026, CEO Sridhar Ramaswamy made a statement that sparked deep industry reflection: “Among the four core components of an Agentic enterprise, the Agentic control plane is the most important.” The company, known for its data warehouse, is investing $6 billion to deepen its partnership with AWS and deeply integrate with Anthropic’s Claude, repositioning itself from a “data warehouse” to an “AI data operating system.”

Three Key Industry Judgements:

- The biggest bottleneck in AI is not models, but data infrastructure: Anish Sharma, Senior Managing Director at Accenture, shared that “85% of client problems are not AI problems – they are data problems.” Large models cannot magically access internal enterprise data such as order records and supply chain information – these data silos spread across dozens of systems are the real gap that AI cannot cross.

- AI consumes software, but data will not be consumed: As agents gradually take over the workflows of SaaS applications, the value of traditional software is being redefined. However, the strategic value of enterprise data is increasing rather than decreasing – whoever controls a unified, trustworthy, and governable data entry point holds the most core asset of the AI era. Snowflake’s CEO phrased it precisely: “Not the brain of AI, but the nervous system of AI.”

- Agent governance capability is more scarce than reasoning capability: Snowflake introduced the “Intelligence Control Plane” concept, aiming to become the central hub shared by all enterprise agents. Anthropic co‑founder Daniela Amodei emphasised: “Trust is the accelerator of innovation.”

The China Path: Snowflake’s path faces a fundamental difference in the Chinese market – data sovereignty. YashanDB’s layout addresses this need. As a fully self‑developed database with 100% independently controllable kernel code, it natively embeds AI capabilities from the kernel level rather than through external integration, achieving unified storage and cross‑modal fusion computing of structured, vector, JSON, and other heterogeneous data within a single engine, with vector retrieval accuracy nine times higher than open‑source vector databases. YashanClaw, YashanDB’s enterprise‑grade AI agent management platform, is highly aligned with Snowflake’s “Intelligence Control Plane” concept, innovatively introducing data sandbox technology to provide highly reliable support for AI agents’ high‑frequency read/write operations.

- DBA Perspective: Snowflake’s strategic pivot sends a clear signal – in the AI agent era, databases are evolving from passive data storage tools into the central foundation supporting agent operation and collaboration. For DBAs, this means the boundaries of database responsibilities will expand significantly: not only managing data storage and queries, but also providing unified context, permission control, and task orchestration for AI agents. YashanDB’s “data sandbox + four‑layer isolation” solution provides a reference technical path for DBAs in AI security and compliance scenarios.

- CTO Perspective: The “iron triangle” alliance of Snowflake, Anthropic, and AWS reveals a deeper trend – competition in the AI era has escalated from technology competition to ecosystem competition. For Chinese enterprises, the three‑layer architecture of data foundation + AI large models + agent management is becoming the standard paradigm for enterprise AI, but the core data layer must be independently controllable.

- Investor Perspective: Snowflake’s strategic upgrade from “data warehouse” to “AI data operating system” represents a shift in valuation logic for the database track – from storage tools to AI infrastructure, the market is willing to pay a higher premium for “data control points.” Chinese database vendors with independent controllability and an evolution toward AI‑native foundations will benefit from this valuation restructuring.

## 02｜Dameng Stock Drops 3.41% to RMB 29.3 Billion Market Cap, Net Outflow of Nearly RMB 50 Million in 5 Days

On June 23, Dameng (688692) fell 3.41%, with a turnover of RMB 515 million, turnover rate of 2.66%, and a total market capitalisation of RMB 29.329 billion.

Fund Flows: Main net outflow was RMB -23.35 million on the day, with a 5‑day main net outflow of RMB -49.48 million, and consecutive 2‑day reduction by major funds. Industry ranking 113/136, with the industry’s main net outflow at RMB -1.251 billion.

Distribution: Major funds do not have a controlling position, and the distribution is very dispersed. Major fund turnover was RMB 115 million, accounting for 5.71% of total turnover. The average trading cost was RMB 266.26 per share, and recent attention to the distribution has decreased.

Fundamentals: Dameng’s Q1 2026 revenue reached RMB 411 million, up 59.09% year‑on‑year; net profit attributable to shareholders was RMB 151 million, up 54.25% year‑on‑year. The company’s revenue composition is: software product usage authorisation 92.55%, O&M services 3.86%, data and industry solutions 1.78%, and database appliance sales 1.55%.

Concept Sectors: Xinchuang, data elements, AI corpora, Huawei concept, smart grid. The company stated on its investor interaction platform that it is actively exploring the use of its full‑stack database products for DB for AI business, including supporting and storing training corpora, RAG, knowledge graphs, and prompts, and steadily advancing the R&D of its full‑stack self‑developed multi‑modal intelligent computing products and platforms.

- DBA Perspective: Dameng’s short‑term stock volatility does not change its strong fundamentals – Q1 revenue growth of 59% and net profit growth of 54%, with software authorisation revenue accounting for 92.55% of the high‑margin structure, indicate that its core business is in a rapid expansion phase. For DBAs in the domestic database track, Dameng’s exploration in AI corpora, RAG, and knowledge graphs means that the technological layout of domestic databases in AI data infrastructure is accelerating.

- Investor Perspective: Dameng’s total market cap of RMB 29.3 billion, corresponding to Q1 revenue of RMB 411 million and net profit of RMB 151 million, places its valuation at a leading position in the domestic database track. Short‑term main fund outflows are related to sector rotation, but the company’s technological layout in AI (multi‑modal intelligent computing products, RAG, knowledge graphs) and IDC’s confirmation of an 11.3% on‑premises market share surpassing foreign vendors constitute medium‑ to long‑term valuation support.

## 03｜Kingware Community’s First Overseas Stop in Hong Kong: Multi‑Modal Convergence + AI Agent, Database Upgrades from Storage Tool to Agent Foundation

Recently, Kingware Community’s first overseas event, the “KING Big Shot Face‑to‑Face” offline technical salon, was successfully held at the Hong Kong Science Park. With the core theme “Beyond Compatibility, Converging Multi‑Modal to Build a Solid Digital Intelligence Foundation,” the event brought together industry experts from Hong Kong and the mainland to focus on multi‑modal database innovation and AI + database migration implementation.

Technical Sharing Highlights:

- One SQL Across Four Data Models: Independent data systems technical consultant Xiao Shaocong demonstrated – without splitting into multiple heterogeneous databases, a single SQL statement can uniformly operate relational data, vector data, JSON semi‑structured data, and graph topology data, showcasing the value of lightweight integrated data architecture.

- Making Databases the Infrastructure Architecture for AI Agents: Yin Haiwen, head of Kingware KVA, focused on the underlying infrastructure for large model deployment, explaining that AI agent operation relies on high‑performance database support, covering four core scenarios: vector retrieval storage, dialogue memory persistence, knowledge base mounting, and tool call data interaction.

- Multi‑Modal + AI Dual‑Drive: Industrial internet database expert Wang Dingding, based on industrial internet scenarios, noted that industrial scenarios involve混杂 device time‑series data, drawing JSONs, quality inspection vectors, and production line relational data. A single database cannot handle this complexity; a multi‑modal engine uniformly stores all data types, combined with AI for anomaly detection, predictive analysis, and intelligent scheduling, forming a closed loop of “unified data storage + intelligent analysis.”

Hong Kong Local Case: Matthew Wong, Delivery Manager at Automated System Ltd, shared real‑world implementation cases of Hong Kong local enterprises in finance, government, and commerce, covering the full lifecycle from project assessment, migration cutover, go‑live operations to long‑term optimisation, validating Kingbase’s competitiveness in stability, operations response, customised adaptation, and total cost.

Overseas Strategy: This salon was an important step in Kingware Community’s entry into overseas ecosystem layout. Fu Yi, Manager of the Presales Support Department at CETC Kingware, stated that compatibility is just the baseline for Kingbase; the core competitiveness lies in deeply integrating multi‑modal and AI capabilities to create integrated solutions, exporting highly reliable, scalable, and independently controllable digital intelligence foundation solutions to Hong Kong, Macau, and overseas markets.

- DBA Perspective: Kingware Community’s first overseas stop in Hong Kong signals the clear “technology export” intention of domestic databases. The capability of “one SQL across four data models” means DBAs can perform unified operations on relational, vector, JSON, and graph data within a single Kingbase database, eliminating data movement between multiple database systems. The real‑world implementation cases of Hong Kong local enterprises also provide reusable practical references for DBAs in cross‑border deployment and Xinchuang migration scenarios.

- CTO Perspective: Kingware’s strategic positioning of “beyond compatibility, converging multi‑modal” echoes the technical paths of OceanBase’s “lakehouse integration” and YashanDB’s “multi‑modal convergence.” The competition among domestic databases has upgraded from “compatibility” to a comprehensive contest of “multi‑modal convergence + AI native.”

- Investor Perspective: Kingware Community’s first overseas stop in Hong Kong is a landmark event for domestic databases moving from “domestic replacement” to “international competition.” Hong Kong, as an international financial centre, validates the competitiveness of domestic databases in high‑end scenarios such as finance and government through its local enterprise implementation cases, laying the foundation for subsequent expansion into Southeast Asia and other overseas markets.

## 04｜AI Replaces Human Labour: Oracle Lays Off Over 20,000 Employees, Compute Budgets Accelerate Replacing Human Resources

According to latest regulatory filings, Oracle Corporation ( ORCL.US ) reduced its workforce by 21,000 employees over the past 12 months, from approximately 162,000 to 141,000 – a 13% reduction – including positions eliminated due to the accelerating penetration of AI agents into enterprise business models.

Layoff Background: Oracle is raising $45‑50 billion for AI cloud infrastructure expansion, advancing large‑scale AI data centre contracts for customers such as OpenAI and Meta. The current fiscal year’s capital expenditure plan is approximately $70 billion. This large‑scale workforce reduction is not ordinary cost cutting, but a typical shift from “opex to capex” – fewer people, with cash flow and financing resources concentrated on GPU clusters, data centre power, network interconnection, and AI customer contract delivery.

Industry Trend: Oracle explicitly acknowledged in regulatory filings: “The large‑scale adoption and deployment of cutting‑edge artificial intelligence technologies across our businesses has resulted in, and may continue to result in, a significant reduction in our workforce.” Wall Street expects the AI compute arms race to enter a systemic expansion phase. Morgan Stanley has significantly revised its 2026 US large‑cap tech capital expenditure forecast upward from $433 billion to $805 billion, with 2027 projected to reach $1.1 trillion.

- DBA Perspective: Oracle’s layoff of 21,000 employees is a signal worth attention for Oracle DBAs. Although Oracle databases still hold an important position in the enterprise market, Oracle itself is transforming from a traditional database/enterprise software company into an AI cloud infrastructure contractor – AI compute budgets are replacing human budgets at scale. For DBAs, this means that relying solely on Oracle database operations skills is a narrowing career moat. Skill boundaries need to be expanded toward AI data infrastructure, multi‑cloud management, and vector search.

- CTO Perspective: Oracle’s layoffs reveal the fundamental change in enterprise operating logic in the AI era – a shift from human resource expansion to compute capital expansion. Tech companies are reallocating capital from traditional human resources to AI compute infrastructure, cutting away the human redundancy in the old production function and adding compute capital, data assets, and engineering delivery capabilities in the new production function. This trend carries a cautionary message for CTOs across all industries.

- Investor Perspective: Barclays maintained an “overweight” rating on Oracle, viewing the layoffs as a move to free up cash flow and reallocate resources from low‑return to high‑growth segments. Oracle is being repriced by the market as an AI cloud infrastructure contractor rather than a traditional database software company. The global AI capex benchmark model projects growth from $765 billion annually in 2026 to $1.6 trillion annually by 2031. As a key participant in AI cloud infrastructure, Oracle’s valuation logic is being rewritten.

## 05｜Weekly Security Vulnerabilities Focus: pgAdmin AI Assistant Transaction Bypass RCE, IBM Db2 DoS, PostgreSQL Multiple CVE Fixes

pgAdmin 4 AI Assistant Critical Vulnerability (CVE-2026-12045) : Affects pgAdmin 4 versions 9.13 through before 9.16. The AI assistant’s execute_sql_query tool places LLM‑generated SQL inside a BEGIN TRANSACTION READ ONLY wrapper to prevent data modification. However, when the LLM‑provided query is passed to the database driver, it is not restricted to a single statement or read‑only verbs – multi‑statement payloads containing COMMIT, END, ROLLBACK, or ABORT can terminate the read‑only transaction, with subsequent statements executed in autocommit mode. Attackers can use prompt injection to influence any database content read by the AI assistant, causing the LLM to issue malicious multi‑statement payloads.

Fix Strategy: Version 9.16 fixes the issue by requiring that LLM‑provided queries parse to exactly one non‑empty/non‑comment statement, with the first non‑whitespace token being one of SELECT, WITH, EXPLAIN, SHOW, VALUES, or TABLE.

IBM Db2 DoS Vulnerability (CVE-2026-11906) : Affects IBM Db2 versions 11.5.0‑11.5.9 and 12.1.0‑12.1.4. An authenticated user can exploit improper sanitisation of special elements in XMLTable derived column data query logic to cause denial of service, CVSS 6.5.

PostgreSQL Multiple CVE Fixes (AlmaLinux Security Update ALSA-2026:27743) : Fixes CVE-2026-6475 (pg_basebackup and pg_rewind symbolic link path traversal), CVE-2026-6477 (libpq buffer overflow, server superuser can overwrite client stack memory), CVE-2026-6478 (MD5 password comparison timing side‑channel), and CVE-2026-6473 (integer overflow leading to out‑of‑bounds write).

- DBA Perspective: The pgAdmin 4 AI assistant vulnerability is a “security warning sign” for the AI‑ification of database tools – when AI assistants are embedded in database management tools, prompt injection can become a new attack vector. DBAs using pgAdmin 4 should upgrade immediately to version 9.16. The IBM Db2 DoS vulnerability and four PostgreSQL CVE fixes (including symbolic link path traversal, buffer overflow, and timing side‑channel) constitute a high‑intensity security patch window this week. DBAs should identify pgAdmin versions, Db2 versions (11.5/12.1), and PostgreSQL versions, and schedule relevant patches into this week’s operations calendar.

- CTO Perspective: The pgAdmin AI assistant vulnerability exposes new security challenges in “AI + database tools” – prompt injection in AI assistants can be used to bypass read‑only transaction restrictions. This reminds CTOs that when introducing AI‑enhanced database management tools, AI‑layer security validation must be incorporated into the overall security architecture. PostgreSQL’s MD5 timing side‑channel vulnerability (CVE-2026-6478) also reinforces the need to migrate authentication from MD5 to SCRAM-SHA-256 as soon as possible.

- Investor Perspective: New security vulnerabilities arising from the AI‑ification of database tools will drive new demand for AI security auditing and prompt injection protection tools. pgAdmin, as one of the most popular PostgreSQL management tools, has a wide impact from its AI assistant vulnerability, creating market opportunities for companies providing AI application security testing and database tool security scanning.

## 📚 SQL Little Knowledge Point

This Issue’s Knowledge Point: What is an AI Agent’s “Control Plane”?

The “Intelligence Control Plane” concept introduced by Snowflake at Summit 2026 represents a core architectural innovation for data infrastructure in the AI agent era.

Traditional Architecture vs. Agent‑Era Architecture:

| Dimension | Traditional Architecture | Agent‑Era Architecture |

|-----------|-------------------------|------------------------|

| Data Access | Application → Database | Agent → Control Plane → Data Foundation |

| Permission Control | User‑level permissions | Agent‑level dynamic permissions |

| Context Management | Stateless sessions | Cross‑agent context sharing |

| Task Orchestration | Application‑level hard‑coding | Control plane unified scheduling |

| Observability | Database logs | Full‑chain agent behaviour tracing |

The Four Core Responsibilities of the Control Plane:

1. Unified Context: Provides shared session history, user preferences, and business rules for all agents.

2. Unified Permissions: Agent‑level fine‑grained access control, rather than traditional user‑level permissions.

3. Unified Data Access: Abstracts underlying heterogeneous data sources, providing agents with a unified interface.

4. Unified Governance: Full‑chain observability, auditability, and traceability.

Significance for DBAs: The emergence of the control plane means that the DBA’s management boundaries will extend from “database instances” to the “agent data access layer” – requiring the design of agent permission models, auditing of agent query behaviour, and optimisation of agent data access paths. This is a new responsibility for DBAs in the AI era, and a new direction for career development.

> HiddenMerit Team Production

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