Segment 1: HiddenMerit Morning Post · Issue 103

Focus on Database Frontiers, Practical Insights for DBAs September 15, 2026 ｜ 5 Selected Global Breaking News

01｜IDC Latest Report: AI Data Infrastructure Market to Reach $73.8 Billion by 2030, 37.7% CAGR

On September 10, at the 2026 Inclusion · Bund Summit Insights Forum, Wang Nan, Research Manager for China Enterprise Software Market at IDC, shared the latest research on AI data infrastructure. The report predicts that from 2025 to 2030, China’s AI data infrastructure market will grow at a compound annual growth rate of 37.7%, reaching $73.8 billion by 2030; by 2035, the market is expected to further reach $154.8 billion (approximately RMB 1 trillion), more than a 10x increase from 2025.

IDC divides AI data infrastructure into three major areas: AI Data Platform, AI Data Intelligence, and AI Data Services. Among them, AI Data Intelligence will be the fastest‑growing sub‑market, with a CAGR of 48.2% from 2025 to 2030. Wang Nan noted that the key to future competition lies in who can first form a complete closed loop covering “data management — data intelligence — data services — agent consumption.”

DBA Perspective: A 37.7% CAGR means databases are moving from “supporting business” to “serving intelligence.” Traditional databases, data warehouses, and data lakes all need to adapt to the data requirements of AI applications and agents. For DBAs, this means skill development over the next 3‑5 years needs to expand from “SQL tuning” to “multi‑modal data management, AI data platform architecture, and agent data services.”

02｜MongoDB Shares Surge 5% in a Single Day: Atlas High Growth Plus MCP Features Ignite AI Theme

On September 15, cloud database provider MongoDB (MDB) saw its shares surge approximately 5.09% intraday to $380.64. Market buying focused on the company’s growth momentum in AI and data centre modernisation, as well as the high growth trend of its cloud product MongoDB Atlas.

According to MarketBeat, MongoDB’s latest fiscal quarter saw revenue growth exceeding 30% , while also improving operating margin and cash flow. CEO CJ Desai emphasised at the Citi TMT Conference that the company is benefiting from enterprise modernisation, AI‑native customer demand, and large enterprises migrating from traditional relational databases to MongoDB to support scalable, real‑time AI workloads.

Key Data: MongoDB added approximately 2,900 new customers this quarter, with about 99% choosing Atlas. The newly launched embeddings and fully managed MCP (Model Context Protocol) features have boosted developer usage and Atlas traffic, becoming the primary catalyst driving the stock price upward.

DBA Perspective: MongoDB’s AI theme is moving from “concept” to “revenue validation.” Of the 2,900 new customers, 99% chose Atlas, indicating that the cloud‑managed model has become the mainstream choice. The launch of MCP features is particularly noteworthy — it means MongoDB is positioning itself as the data access layer for AI agents, and DBAs need to pay attention to MCP protocol security configuration and permission management.

📌 HiddenMerit In‑Depth Preview: Today’s in‑depth article – “What is MCP? Why Are MongoDB, SQL Chat, and Postgres MCP All Betting on This Protocol?” – will explore the technical principles of MCP and DBA response strategies.

03｜Oracle AI Database 26ai Launches Rust Driver and LangChain.js Integration: AI Developer Toolchain Accelerates

In its September technical resource update, Oracle introduced several developer tools for AI Database 26ai.

Core New Tools:

rust-oracledb: A brand‑new high‑performance, lightweight pure Rust driver that requires no Oracle client libraries, allowing Rust developers to connect directly to Oracle databases.
langchain-oracledb: Brings Oracle AI Database 26ai capabilities into the LangChain.js ecosystem, enabling developers to build AI applications and workflows using TypeScript/JavaScript.
RAG Pipeline Tutorial: Combines Oracle AI Vector Search, @oracle/langchain-oracledb TypeScript integration, and OCI Generative AI to provide an end‑to‑end RAG building guide.

Additionally, Oracle API Access Control is now available for Autonomous AI Database Dedicated deployments, allowing enterprises to designate selected administrative operations as privileged operations requiring authorised, time‑limited approval.

DBA Perspective: Oracle is positioning “developer friendliness” of AI databases as a point of differentiated competition. The Rust driver’s launch means Rust ecosystem developers can seamlessly connect to Oracle, while LangChain.js integration directly targets the mainstream technology stack for AI application development. For DBAs, this means Oracle database’s AI capabilities are expanding from “kernel‑level features” to “developer toolchain‑level integration.”

04｜Yugabyte Agentic AI Product Line Fully Rolled Out: Meko Daily Database Creation Grows 10x

Yugabyte recently announced significant progress in its Agentic AI product line. The company’s distributed PostgreSQL database YugabyteDB has achieved over 5 million deployed clusters, covering 100+ countries, serving Fortune 100 customers in finance, retail, telecommunications, and other industries.

Core Products:

YugabyteDB AMP: A serverless, scale‑to‑zero, fully PostgreSQL‑compatible database layer using a “pay‑as‑you‑accelerate” consumption model. Users can launch hundreds of databases in seconds to meet multi‑agent system requirements.
Meko: An agent‑native context engine providing shared context, accumulated memory, and knowledge for multi‑agent AI systems. Since its launch in May 2026, the number of databases created daily by agents has grown 10x.

CEO Karthik Ranganathan stated: “We are at an inflection point where the market is shifting from cloud‑native applications to an agentic future. Internally, over 80% of our software and operations are already optimised through agentic workflows.”

DBA Perspective: The 10x growth in daily database creation via Meko is striking — it means agents are becoming the primary creators of databases, not just consumers. YugabyteDB AMP’s ability to “launch hundreds of databases in seconds” directly addresses the multi‑agent system’s need for elastic database creation. For DBAs, this means the objects of database management will expand from “fixed instances managed by humans” to “elastic instances dynamically created by agents.”

05｜Transwarp Passes HKEX Hearing: GPU‑Native Cognitive Database Moves from “Replacement” to “AI Infrastructure”

Transwarp passed the Hong Kong Stock Exchange main board listing hearing on August 27, taking a key step toward “A+H” dual listing. This AI infrastructure software vendor is shifting from a “domestic replacement” narrative to an “AI infrastructure platform company” positioning.

Technical Breakthrough: Transwarp’s GPU‑Native Cognitive Database migrates the database architecture from CPU to GPU, achieving full‑chain GPU execution. An article on NVIDIA’s Chinese developer website noted that the system achieved complete execution of all 99 TPC‑DS SF50 queries on GPU without CPU fallback, with overall performance improvement of approximately 10.5x and a peak single‑query speedup of 94x.

Multi‑Card Cluster Breakthrough: After multiple cards are interconnected into a small cluster, TPC‑DS 1TB benchmark performance reached 100x that of two CPUs, and the complete TPC‑DS test was over 70x faster.

A Guolian Minsheng Securities research report noted that during WAIC 2026, the company jointly released a next‑generation cognitive database solution based on Hygon DCU with Sugon Intelligent Computing, achieving 51.8x performance improvement over traditional CPU solutions in real‑world tests, with 7.97x scaling efficiency in 8‑card scenarios.

DBA Perspective: Transwarp’s GPU‑native database represents a more radical architectural path — not “using GPUs to accelerate databases,” but “moving the complete database onto GPUs.” For DBAs, this means tuning dimensions will expand from CPU cores, memory, and disk I/O to GPU video memory management, parallel stream scheduling, and data residency strategies. The H‑share listing will provide Transwarp with more funding support for commercialising GPU databases.

💡 SQL Little Knowledge Point

This Issue’s Knowledge Point: What is MCP (Model Context Protocol)?

MCP is a standard protocol open‑sourced by Anthropic at the end of 2025, designed to enable AI agents to securely and efficiently connect to various data sources and tools. MongoDB’s recently launched fully managed MCP feature is a reflection of this trend.

Core Value of MCP:

Dimension	Traditional Approach	MCP Approach
Connection Method	Each agent develops custom connectors for each data source	Data sources provide MCP Server, agents call uniformly
Permission Management	Scattered across agents	Centralised service accounts + auditing
Development Efficiency	High repetitive work	Configure once, reuse across agents

Significance for DBAs: MCP will become the mainstream protocol for AI agents to access databases. DBAs need to understand MCP Server deployment, permission models, and audit configuration. When MongoDB, SQL Chat, and Postgres MCP Server all begin supporting MCP, it means the “agent access layer” of databases is being standardised.

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

Segment 2: HiddenMerit In‑Depth · Issue 16

Focus on Database Frontiers, In‑Depth Analysis September 15, 2026 ｜ 1 In‑Depth Feature Article

What is MCP? Why Are MongoDB, SQL Chat, and Postgres MCP All Betting on This Protocol?

1. Behind MongoDB’s Share Surge, a Protocol Emerges

On September 15, MongoDB shares surged approximately 5% intraday. Market buying focused on the AI theme, but a closer look at the news details reveals an easily overlooked keyword: MCP (Model Context Protocol).

According to MarketBeat, MongoDB’s newly launched embeddings and fully managed MCP features have boosted developer usage and Atlas traffic, becoming one of the primary catalysts driving the stock price upward.

Meanwhile, looking back at news from the past few weeks: SQL Chat exposed an unauthenticated SQL injection vulnerability, where attackers could connect to arbitrary databases through unauthenticated API endpoints; Amazon awslabs postgres-mcp-server exposed an OS command injection vulnerability, where attackers could escape the read‑only sandbox via COPY ... TO PROGRAM.

[quads id="805"]

These three news items point to the same theme: MCP is becoming the standard protocol for AI agents to access databases, and security issues surrounding MCP are erupting in concentrated fashion.

2. What Problem Does MCP Solve?

Before MCP appeared, the way AI agents accessed databases could be summarised in four words: each doing its own thing.

Each agent framework (LangChain, LlamaIndex, AutoGPT) had its own database connection method; each database (MongoDB, PostgreSQL, MySQL) needed to provide customised connectors for each agent framework; each enterprise had to configure independent authentication and permission management for each data source.

This created problems:

Low development efficiency: Developing connectors for N agents × M data sources, with complexity of N×M.
Blurred security boundaries: Each agent had its own database credentials, making permission auditing impossible.
Heavy operational burden: DBAs had to configure connection pools, monitoring, and auditing separately for each agent.

MCP’s goal is to reduce N×M complexity to N+M: databases provide standard MCP Servers, and agents access them through a unified MCP protocol. Just as USB‑C lets different devices connect to a computer with the same cable, MCP lets different agents connect to databases with the same protocol.

3. MCP Technical Architecture

MCP uses a client‑server architecture:

Role	Function	Database Scenario
MCP Client	Protocol implementation on the AI agent side	Claude, Cursor, ChatGPT, and other agent tools
MCP Server	Protocol implementation on the data source side	MongoDB Atlas, PostgreSQL, MySQL, etc.
Transport Layer	Communication protocol	stdio (local) or HTTP/SSE (remote)

Capabilities exposed by MCP Server (using databases as an example):

Resources: Readable data objects, such as database tables and schema information.
Tools: Executable operations, such as executing SQL queries and listing database connections.
Prompts: Predefined prompt templates, such as “analyse slow queries.”

Security Model: MCP Server handles authentication and authorisation; MCP Client carries credentials. Ideally, the agent doesn’t need to know the database password — it only needs to know the MCP Server address and access token.

4. MCP Security Challenges: Lessons from Three Vulnerability Incidents

Over the past few weeks, three MCP‑related security incidents have revealed the core risks in the current MCP ecosystem:

Incident 1: SQL Chat Unauthenticated SQL Injection (CVSS 9.4)

SQL Chat’s four API endpoints accept client‑provided database connection parameters, allowing unauthenticated users to force SQL Chat to connect to arbitrary database hosts. Attackers can use the SQL Chat server as a springboard to access internal databases.

Lesson: If an MCP Server lacks authentication mechanisms, it becomes a “back door” for attackers to enter internal networks.

Incident 2: Postgres MCP Server OS Command Injection (CVE-2026-87911)

Amazon awslabs postgres-mcp-server’s read‑only enforcement mechanism has flaws. Attackers can execute operating system commands on the PostgreSQL server host without authentication via COPY ... TO PROGRAM statements.

Lesson: MCP Server’s “read‑only mode” does not mean it’s secure. Certain database features (such as COPY TO PROGRAM) can execute system commands under read‑only privileges, and MCP Servers need additional blocking for these features.

Incident 3: MongoDB C Driver Heap Overflow (CVE-2026-84969)

MongoDB C Driver’s BSON‑to‑JSON conversion helper function has a heap overflow vulnerability, affecting numerous MongoDB client applications.

Lesson: The underlying drivers and libraries that MCP Servers depend on may also have vulnerabilities. MCP ecosystem security needs to cover the entire dependency chain.

5. Practical Implications for DBAs

Implication 1: MCP Server is the new “database access layer” and must be included in the security asset inventory

When MongoDB Atlas, PostgreSQL, and MySQL all begin providing MCP Servers, DBAs need to treat these components as the “edge access layer” of databases, not “internal tools of the AI team.” They hold database connection credentials, execute SQL queries, and return data results — any security issue will directly propagate to the database.

Implication 2: MCP Server permission models require dedicated auditing

Traditional DBA permission management revolves around database accounts. MCP Server introduces a new permission layer: agents access databases through MCP Server, and MCP Server then connects to the database with its own credentials.

This means:

MCP Server credentials need least privilege — read‑only queries should not use writable accounts.
The tool exposure scope of MCP Server needs strict control — not all database operations should be exposed to agents.
MCP Server audit logs need to be correlated with database audit logs — to trace “which agent executed which SQL.”

Implication 3: DBAs need to understand the MCP protocol, just as they understood ODBC and JDBC back then

MCP’s adoption rate far exceeds expectations. MongoDB, SQL Chat, and Amazon aws are all rapidly following suit. For DBAs, understanding how the MCP protocol works, its security boundaries, and configuration methods may become a foundational skill for the next 3‑5 years — just as understanding JDBC 20 years ago and REST APIs 10 years ago.

MCP is becoming the “standard interface layer” between AI agents and databases. The security of this interface layer ultimately determines the security of the entire AI data architecture.