HiddenMerit Daily · Issue 44

# 📊 HiddenMerit Daily · Issue 44

> Focus on Database Frontiers, Practical Insights for DBAs

> June 16, 2026 ｜ 5 Selected Global Breaking News

## 01｜MongoDB CFO on AI Winners: Agentic AI Persistent Memory Layer as a Differentiated Core Competitiveness

On June 11, MongoDB (MDB) CFO Mike Berry attended D.A. Davidson’s 2nd Annual Technology & Consumer Conference and gave clear strategic responses to two core investor questions: “Is MongoDB an AI winner?” and “When will AI revenue explode?”

Core Views: Berry clearly stated that MongoDB’s document‑based database architecture is inherently aligned with the AI era’s need to handle unstructured data. Its design, which seemed unconventional twenty years ago, unexpectedly suits today’s AI needs – the document model naturally carries JSON‑format data, and the flexible schema perfectly adapts to unstructured data. Coupled with vector search, persistent memory, and deep LangChain integration, the company is becoming the data infrastructure layer for enterprise‑grade AI agents.

The “Three‑Step” Path to AI Revenue: Berry laid out a clear framework for AI revenue growth:

1. Phase One (current) : Enterprise customers are conducting initial experiments around vector search, but revenue contribution is not yet significant.

2. Phase Two: Some pioneering enterprises (e.g., AI‑native companies) have begun using MongoDB for production‑grade AI applications as the memory layer for AI agents.

3. Phase Three: When enterprise‑grade AI applications enter large‑scale deployment, the infrastructure value of MongoDB will truly explode.

Differentiated Competition: Management believes that at the “agentic AI application” level, MongoDB has built a significant differentiated advantage through its JSON document model, built‑in vector search, and native support for agent governance. The three major challenges enterprises face when deploying multi‑agent systems – persistent memory, vector search, and a real‑time data layer – all have native solutions in MongoDB. Customers such as Adobe and Zomato are already using the MongoDB platform as the memory layer for their real‑time AI agents, providing direct proof of product competitiveness.

Financial Data: FY2027 Q1 total revenue was $687.6 million, up 25% year‑on‑year, exceeding expectations by 3.5%; Atlas revenue grew 29% year‑on‑year; GAAP net profit was $4.4 million, compared to a net loss of $37.6 million in the same period last year, marking the company’s steady progress toward sustained profitability.

- DBA Perspective: The MongoDB CFO’s remarks provide DBAs with an important judgment: the persistent memory layer for AI agents is becoming a new core workload for databases. When leading enterprises like Adobe and Zomato use MongoDB as the memory layer for their AI agents, it means that DBAs’ database management scope will expand to an entirely new area – multi‑turn conversation context for AI agents, long‑term memory storage, vector search, and retrieval‑augmented generation. DBAs need to start learning about vector index maintenance strategies, agent memory data lifecycle management, and how to implement AI agent audit trails.

- CTO Perspective: The “three‑step” path to AI revenue growth revealed by MongoDB management provides CTOs with a reference framework for pacing AI infrastructure investments. From “experimental exploration” to “small‑scale production deployment” to “large‑scale expansion,” the focus of database selection changes at different stages. While Postgres may be acceptable in the early stages, performance and scalability bottlenecks at scale will force platform migration. The cases of Adobe and Zomato using MongoDB as an AI agent memory layer are strong references for assessing MongoDB’s maturity in agentic AI scenarios.

- Investor Perspective: MongoDB’s Q1 earnings beat (revenue +25%, EPS +11.5%) combined with the CFO’s clear AI strategy statement has prompted the market to reassess its value in the AI infrastructure layer. Atlas revenue growth of 29% and RPO up 88% year‑on‑year indicate strong cloud service and long‑term contract locking ability. However, management’s candid admission that AI revenue is still in the early stage means investors should watch for “Phase Two” signals in future quarterly reports.

## 02｜Google Releases Gemini-SQL2: Text-to-SQL Execution Accuracy Breaches 80% for the First Time, NL2SQL Commercialisation Inflection Point Arrives

On June 12, Google Research officially released the new model Gemini-SQL2, built on Gemini 3.1 Pro and specifically designed for “Text‑to‑SQL” tasks. According to the latest data from the industry benchmark BIRD, Gemini-SQL2 achieved 80.04% execution accuracy in the single‑model track, surpassing Google’s previous model versions.

The BIRD evaluation set contains 95 databases from 37 professional domains, with over 12,000 questions, simulating real enterprise environments and incorporating dirty data and test cases requiring external knowledge assistance. It is one of the most rigorous Text‑to‑SQL evaluation benchmarks in the industry.

Google has not yet disclosed the specific model identifier, API interface details, or a detailed technical report for Gemini-SQL2, nor which products will first integrate this capability. However, the 80.04% execution accuracy is a milestone – Text‑to‑SQL model accuracy on BIRD had long hovered in the 70‑75% range. Breaking through 80% for the first time signals that the technical maturity of natural language database querying has entered a new phase.

- DBA Perspective: Gemini-SQL2’s breakthrough of the 80% accuracy threshold marks that “natural language to SQL” technology is moving from a “lab toy” to “production‑ready.” For DBAs, this means the barrier to self‑service data access for business users will be significantly lowered – non‑technical users can ask questions in natural language to generate SQL and retrieve data. However, this also brings new challenges: AI‑generated SQL may have performance pitfalls (missing indexes, full table scans) or logic errors (omitted multi‑table join conditions). DBAs need to establish quality review mechanisms for AI‑generated SQL, execution plan baseline management, and resource consumption monitoring. Moreover, the DBA role will evolve from “hand‑writing SQL” to “AI SQL quality controller.”

- CTO Perspective: Gemini-SQL2’s 80% accuracy is an important milestone for enterprise data democratisation. When natural language database querying accuracy reaches an acceptable level, enterprises can build “conversational data portals” – business users ask questions in natural language, the system automatically generates SQL, executes the query, and returns results. This will significantly compress the cycle from “business question” to “data answer.” When planning data platforms, CTOs should consider NL2SQL capability as a key component of the data service layer, but note that 80% accuracy means 20% of scenarios still require human intervention – a human‑AI collaboration mechanism needs to be designed.

- Investor Perspective: Google’s breakthrough in the Text‑to‑SQL field will accelerate commercialisation of the entire “conversational data analytics” track. Achieving 80% accuracy on the BIRD benchmark means NL2SQL technology has crossed the “usability threshold,” and application scenarios will rapidly expand. Pay attention to startups with technical accumulation in NL2SQL and enterprises integrating NL2SQL capabilities into data democratisation tools (BI, data portals).

## 03｜Mobile Cloud Database Wins Gold at Geneva International Exhibition of Inventions: Core Cloud‑Native Database Technologies Gain International Recognition

Recently, the 51st Geneva International Exhibition of Inventions concluded in Geneva, Switzerland. The achievement titled “Data and Knowledge Dual‑Driven Software‑Hardware Collaborative Optimisation for Cloud‑Native Databases” , jointly presented by Mobile Cloud and Guizhou University, won a gold award. The Geneva International Exhibition of Inventions, along with the Nuremberg International Invention Fair and the Pittsburgh Invention Fair, is recognised as one of the world’s three major invention exhibitions, often called the “Olympics of inventions.”

The project addressed four major pain points in the transformation from traditional databases to cloud‑native environments: difficulty in smoothly migrating heterogeneous architectures, performance bottlenecks with massive data, low efficiency in elastic resource scheduling, and difficulty in intelligent analysis and operations for complex faults. It achieved major breakthroughs in key directions such as dual‑mode parallel operation architecture, intelligent data placement, data‑compute integrated elastic scheduling, and software‑hardware aware intelligent operations. This resulted in an integrated cloud‑native database product combining elastic resource scheduling, cloud‑native storage, seamless migration, and full‑stack intelligent operations.

Leveraging these achievements, Mobile Cloud’s self‑developed DaYun HaiShan Database has been deployed in 31 provinces nationwide, achieving large‑scale application across nine core industries including government, finance, healthcare, education, and internet. In 2025, it passed the “Security and Reliability Level I” evaluation of the China Information Security Evaluation Center, fully adapting to mainstream domestic chips and operating systems.

- DBA Perspective: Mobile Cloud’s DaYun HaiShan Database winning a gold award at an international invention exhibition is a sign that domestic cloud‑native database core technologies have gained international recognition. For DBAs, this means that domestic databases are now technically competitive with mainstream foreign products in cutting‑edge areas such as cloud‑native architecture and intelligent operations. The database has been deployed at scale in 31 provinces and nine core industries, providing DBAs with a rich repository of domestic database operations cases.

- CTO Perspective: The breakthroughs made by Mobile Cloud’s database in smooth migration of heterogeneous architectures, elastic resource scheduling, and intelligent operations address CTOs’ core concerns when selecting cloud‑native databases. DaYun HaiShan Database has passed the “Security and Reliability Level I” evaluation and fully adapts to domestic chips and operating systems, making it a credible option for government and enterprise customers with Xinchuang compliance requirements.

- Investor Perspective: Mobile Cloud’s database winning a gold award at the Geneva International Exhibition of Inventions is a positive signal that Chinese foundational software core technology capabilities have gained international recognition. The large‑scale deployment of DaYun HaiShan Database across 31 provinces and nine core industries demonstrates that its product maturity has been validated through large‑scale commercial use. As a representative of a telecom cloud provider, Mobile Cloud’s technical accumulation in the database track is worth continuous attention.

## 04｜Mobile Cloud’s DaYun HaiShan Database Deployed in Core Systems of Three Major Telecom Operators: Operator Xinchuang Replacement Deepens

Following the award news, the latest deployment progress of Mobile Cloud’s DaYun HaiShan Database deserves separate tracking. According to disclosures, Mobile Cloud’s database has achieved large‑scale deployment in the core systems of the three major telecom operators, covering 31 provinces and nine core industries.

DaYun HaiShan Database is based on a cloud‑native architecture, supporting elastic resource scheduling, cloud‑native storage, seamless migration, and full‑stack intelligent operations. In terms of product capabilities, it has achieved key breakthroughs in intelligent data placement, data‑compute integrated elastic scheduling, and software‑hardware aware intelligent operations. The database has passed the “Security and Reliability Level I” evaluation of the China Information Security Evaluation Center, fully adapting to mainstream domestic chips (such as Kunpeng and Phytium) and operating systems (such as Kylin and UnionTech).

In Issue 42, we reported that OceanBase serves one‑third of China Mobile’s provincial companies and over 1,000 nodes at China Unicom. Now, Mobile Cloud’s DaYun HaiShan Database has also achieved large‑scale deployment in telecom operator core systems, marking a shift in operator Xinchuang replacement from “single‑vendor pilot” to a “multi‑vendor competitive landscape.”

- DBA Perspective: The large‑scale deployment of Mobile Cloud’s database in the core systems of the three major telecom operators provides DBAs with another option for Xinchuang selection in the telecom industry. Operator Xinchuang replacement has formed a multi‑vendor competitive landscape involving OceanBase, Mobile Cloud, Kingbase, and others. This means DBAs need to adopt a “cross‑vendor” perspective when building their skills – understanding the architectural differences and optimisation directions of different domestic databases in operator scenarios.

- CTO Perspective: The shift of operator Xinchuang replacement from “pilot” to “large‑scale deployment,” with the formation of a multi‑vendor competitive landscape, demonstrates that the technical maturity of domestic databases in operator core systems has been fully validated. When planning Xinchuang roadmaps in the telecom industry, CTOs can compare and select among multiple domestic database vendors, no longer limited to a single supplier.

- Investor Perspective: The large‑scale deployment of Mobile Cloud’s database in operator core systems validates the independent R&D capabilities of telecom cloud vendors in foundational software. The market space for operator Xinchuang replacement is huge, and database vendors with delivery cases in the telecom industry will continue to benefit.

## 05｜Vastbase Private Placement Accepted by SSE: HTAP + Multi‑Modal Time‑Series Dual‑Drive Adds New Variable to Xinchuang Market

On June 12, Vastbase ( 603138.SH ) disclosed that its application for a private placement of A‑shares for the 2026 fiscal year has been accepted by the Shanghai Stock Exchange. The company plans to raise up to RMB 702 million , with the entire amount to be invested in two core technology R&D projects: a next‑generation high‑performance hybrid transaction/analytical database (HTAP) project (RMB 489 million) and a multi‑modal time‑series database project (RMB 213 million).

Vastbase believes that the traditional “OLTP database + OLAP data warehouse” separated architecture can no longer meet the real‑time, integrated, and high‑concurrency requirements of core scenarios such as finance, retail, telecommunications, and manufacturing. The HTAP integrated architecture can deliver both high‑concurrency transaction performance and batch analytical efficiency. The multi‑modal time‑series database project targets core needs in emerging scenarios such as industrial internet and energy dispatch.

The company’s Vastbase database is a purely domestic relational database product with complete independent intellectual property rights. It has passed national security and reliability assessments and has achieved large‑scale deployment in key industries such as government, finance, telecommunications, manufacturing, energy, defence, and transportation. It is worth noting that the company has recorded four consecutive years of losses, with its Q1 2026 net loss attributable to shareholders widening to RMB 41.77 million, and operating cash flow plunging 214% year‑on‑year.

- DBA Perspective: Vastbase’s RMB 700 million private placement doubling down on HTAP and multi‑modal time‑series is another indication of “technology intensification” in the domestic database track. For DBAs, HTAP integrated architecture means saying goodbye to ETL pipelines and dual‑system maintenance, but it also requires DBAs to master hybrid workload tuning – coordinating transaction response and batch analytics within a single system, with refined resource scheduling and SLA guarantees. Time‑series data processing capability will also become a core competency for DBAs in IIoT scenarios. However, Vastbase’s four consecutive years of losses present a financial risk that cannot be ignored; DBAs need to assess the long‑term viability of vendors when choosing technology stacks.

- CTO Perspective: Vastbase’s customers are primarily state‑owned enterprises. This private placement targeting HTAP and multi‑modal time‑series, if successfully delivered, will add a more complete integrated domestic database option for telecom and energy scenarios. However, management’s history of repeated project delays and unimproved profitability are risks that cannot be overlooked. Technology decision‑makers should carefully assess vendor financial health when evaluating HTAP product delivery capability.

- Investor Perspective: The dual tracks of HTAP and multi‑modal time‑series have clear market demand – strongly driven by scenarios such as financial real‑time risk control, industrial internet, and energy dispatch. However, Vastbase’s historical baggage of “four consecutive years of losses + controlling shareholder penalty for illegal shareholding reduction” cannot be ignored. There is significant uncertainty whether the large offering will be approved. The subscription enthusiasm and multiples from institutional investors will be key indicators of short‑term market confidence. Subsequent focus should be on the order conversion efficiency of the investment projects and the pace of accounts receivable collection.

## 📚 SQL Little Knowledge Point

This Issue’s Knowledge Point: What is Text‑to‑SQL Execution Accuracy?

Text‑to‑SQL execution accuracy is a core metric for measuring the performance of natural language to SQL models. It refers to whether the result set returned by executing the SQL statement generated by the model on a real database matches the standard answer.

Difference from “Matching Accuracy” :

- Matching Accuracy: Compares whether the generated SQL string exactly matches the standard SQL string. This method is overly strict – SQL statements that are semantically equivalent but written differently (e.g., SELECT a,b FROM t vs SELECT b,a FROM t) would be marked as incorrect.

- Execution Accuracy: Actually executes both the generated SQL and the standard SQL on a database and compares the result sets. As long as the returned data is consistent, even if the SQL is written differently, it is considered correct. This approach better reflects real‑world application scenarios.

The BIRD Benchmark: BIRD (Big Bench for Large‑scale Database Grounded Text‑to‑SQL Evaluation) is currently the most rigorous Text‑to‑SQL evaluation benchmark in the industry. It contains 95 databases, over 12,000 questions, covering 37 professional domains. Its characteristics include simulating real enterprise environments, incorporating dirty data, and test cases that require external knowledge assistance.

Significance of Gemini-SQL2’s Breakthrough: 80.04% execution accuracy is an important milestone in the Text‑to‑SQL field – model accuracy on BIRD had long hovered in the 70‑75% range. Breaking through 80% means that the technical maturity of natural language database querying has entered a new phase, clearing a key obstacle to the commercialisation of “conversational data analytics.”

> HiddenMerit Team Production

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