Exam facts and blueprint
The official credential name is Microsoft Certified: Azure Data Fundamentals, and the exam code is DP-900. Microsoft positions it for candidates beginning to work with data in the cloud. Candidates should understand relational and non-relational data and transactional and analytical workloads. The credential can support preparation for later Azure data certifications, but Microsoft says it is not a prerequisite.
Microsoft's certification page states that candidates have 45 minutes to complete the assessment. The study guide says a score of 700 or greater is required and links official scoring details. It also explains that related topics may appear, most questions cover generally available features, and commonly used preview capabilities can appear. Localized updates can lag the English exam, so check the live page for the booked language.
The independent bank's 14/12/9/15 split equals 28%, 24%, 18%, and 30%, which stays within all official ranges. It is a study allocation, not a claim about the number or composition of official items.
Domain 1: describe core data concepts
Representations
Structured data has a consistent schema, commonly rows and columns. Semi-structured data uses keys, tags, nesting, or metadata while allowing records to vary. JSON and XML are familiar examples. Unstructured data lacks a predefined tabular model: images, audio, video, and free-form documents fit here. A file can carry metadata even when the content is unstructured.
Formats and stores
CSV is a simple interoperable format for flat tables, but it has limited type and nesting support. JSON handles hierarchical records and flexible properties. Parquet is columnar and schema-aware, making it useful for analytical compression and selective column reads. Learn the trade-off rather than memorizing that one format is universally best.
A file store organizes files; object storage organizes objects and metadata; a database adds a management system for structured access, control, updates, and queries. A data lake can preserve varied raw and curated content for several engines. A data warehouse optimizes integrated historical data for structured analytical querying.
Workloads and roles
Transactional processing records individual operations and emphasizes consistent writes, concurrency, and point access. Analytical processing scans and aggregates larger datasets, often across history. A retail solution may use an operational order database and a separate analytical star model because access and optimization patterns differ.
Database administrators focus on database availability, security, backup, recovery, and performance. Data engineers build ingestion, transformation, and serving pipelines. Data analysts build models, reports, visualizations, and interpretation. Real organizations can overlap roles, but the exam tests their primary responsibilities.
Domain 2: relational data on Azure
Relational data uses tables, rows, columns, keys, and relationships. A primary key uniquely identifies a row; a foreign key references another entity. Constraints protect valid state. Indexes can improve selected reads at the cost of storage and write maintenance. Views expose named queries. SQL statements include SELECT for reading, INSERT for adding, UPDATE for changing, and DELETE for removing rows.
Normalization separates entities and reduces redundant facts. If a customer address appears in every order, changing it requires many updates and creates inconsistency risk. Store the customer once and reference the customer key from orders. For analytics, a deliberately denormalized dimensional model may be appropriate; normalization is not a moral rule detached from workload.
Azure SQL family
Azure SQL Database is a managed database-oriented platform service. Choose it when the application needs SQL Server-compatible relational capability without host management. Azure SQL Managed Instance provides broad instance compatibility for workloads that depend on instance-scoped behavior but still want a managed service. SQL Server on Azure Virtual Machines provides operating-system and full instance control, which also leaves more patching, availability, backup, and security responsibility with the customer.
For open-source engines, identify Azure Database for PostgreSQL and Azure Database for MySQL. Engine compatibility, extensions, migration dependencies, scaling, availability, and operating model should drive selection.
Domain 3: non-relational data on Azure
Azure Blob Storage is object storage for large unstructured datasets such as images, backups, logs, and analytical files. Block blobs fit general objects, append blobs optimize append operations, and page blobs support random access patterns such as virtual disks.
Azure Files supplies managed file shares through supported SMB and NFS scenarios. It is useful when an application expects a file-system share. Azure Table storage is a schemaless key-attribute store centered on partition and row keys. It does not provide relational joins and constraints.
Azure Cosmos DB supports globally distributed operational applications with elastic scale, latency and consistency choices, and several APIs. Cosmos DB for NoSQL is the native JSON document API. Cosmos DB for MongoDB supports applicable MongoDB protocol and driver scenarios. Cosmos DB for Apache Cassandra supports Cassandra-compatible workloads, Cosmos DB for Apache Gremlin targets graph vertices and edges, and the Table API supports compatible key-value or table patterns. Choose from the application data model, protocol, partitioning, query, distribution, consistency, migration, and cost requirements—not merely the product name.
Domain 4: analytics workloads on Azure
Ingestion, processing, and stores
A modern analytics path begins with sources and ingestion. Data is copied or streamed into a landing area, validated and transformed, and served through a lakehouse, warehouse, semantic model, report, application, or model. Every stage needs ownership, identity, schema handling, quality, lineage, retention, monitoring, and cost controls.
A data lake supports varied raw and curated files. A warehouse emphasizes structured analytical serving. A lakehouse combines lake storage with table and management capabilities. The exam expects conceptual selection rather than deep platform administration.
Azure Databricks and Microsoft Fabric
Azure Databricks is a collaborative analytics platform using Apache Spark and lakehouse capabilities for data engineering, analytics, and machine learning. Microsoft Fabric is an end-to-end SaaS analytics platform spanning Data Factory, Data Engineering, Data Science, Data Warehouse, Real-Time Intelligence, Power BI, and related experiences. Both can contribute to large-scale analytics; organizational architecture, existing tools, governance, integration, skills, and operating model influence the decision.
Batch and streaming
Batch processes bounded groups, such as last night's orders. Streaming handles events continuously or with low latency, such as factory alerts. Azure Stream Analytics provides managed stream processing with a SQL-like language. Fabric eventstreams capture, transform, and route real-time events into Fabric Real-Time Intelligence scenarios. A solution can use streaming for detection and batch for history; avoid labeling a nightly dashboard real time.
Power BI models and visualizations
Power BI connects to data, transforms it, builds semantic models, creates interactive reports, and distributes governed insights. A semantic model contains tables, relationships, measures, and business logic. In a star schema, a fact table declares measurable grain—such as one row per order line—and dimension tables describe date, product, customer, or store. Clear grain prevents joins from multiplying values.
Visual choice follows the question. A line chart shows an ordered time trend. Bar or column charts compare discrete categories. A scatter plot examines relationships between numeric variables. A table provides exact detail. A card communicates a small number of headline measures. Slicers support interactive filtering. Validate axes, sorting, units, blank states, accessibility, refresh, and authorization.
| Requirement | Likely concept or service | Key caveat |
|---|---|---|
| Managed SQL database, low host administration | Azure SQL Database | Check database-scoped compatibility. |
| Managed SQL Server with broad instance compatibility | Azure SQL Managed Instance | Validate exact migration dependencies. |
| Full OS and SQL instance control | SQL Server on Azure VM | Customer manages more operations. |
| Images and logs | Blob Storage | Choose blob type, tier, redundancy, and access. |
| Managed shared file path | Azure Files | Confirm protocol and performance needs. |
| Global operational JSON data | Azure Cosmos DB | API, partition, consistency, and cost matter. |
| Seconds-level event detection | Stream Analytics or Fabric real-time path | Handle late, duplicate, and failed events. |
| Interactive business reporting | Power BI | Model grain, security, and freshness first. |
A five-week preparation plan
Week 1: Classify structured, semi-structured, and unstructured samples. Convert a small invented table among CSV, JSON, and Parquet. Compare files, object stores, databases, lakes, and warehouses. Map roles and transactional versus analytical workloads.
Week 2: Build a normalized relational schema with keys, constraints, indexes, views, and basic SQL. Compare Azure SQL Database, Managed Instance, SQL Server on Azure VMs, PostgreSQL, and MySQL from actual requirements.
Week 3: Prototype Blob, Files, and Table patterns. Model a small Cosmos DB document workload and explain API, partition, and consistency choices.
Week 4: Draw batch and streaming architectures. Compare lake and warehouse stores, Azure Databricks, Microsoft Fabric, Azure Stream Analytics, and Fabric eventstreams.
Week 5: Build a Power BI star model and question-led report. Complete all three projects, review 40 cards, answer 50 questions by the exact 14/12/9/15 distribution, use Microsoft's Practice Assessment and sandbox, and revisit weak objective bullets.
Three projects that make fundamentals concrete
The Azure data store decision lab classifies invented data, compares formats, creates relational and non-relational prototypes, tests permissions and recovery, and removes all resources.
The transactional-to-analytical project moves synthetic orders from an operational schema into a governed lake and gold star model, compares Azure Databricks and Fabric, and adds a small streaming threshold path.
The Power BI project creates explicit metrics and grain, a semantic model, appropriate visuals, slicers, row-level access tests, refresh and stale-data evidence, and clean retirement.
Question strategy and common traps
Underline the requirement: structure, protocol, engine compatibility, management level, distribution, latency, access pattern, or visualization question. “Shared SMB path” points to Azure Files; “object media” points to Blob; “global document data” can point to Cosmos DB; “instance compatibility without VM management” points to Managed Instance; “host control” points to SQL Server on Azure VMs.
Reject false absolutes. No store is best for all data. Semi-structured does not mean structure-free. A bigger database tier does not fix an analytical grain error. A nightly dashboard is not real time. A hidden report field is not authorization. Parquet does not create database transactions. Cosmos DB API choice does not eliminate partition design.
Readiness checklist
Readiness means explaining each official objective in plain language and applying it to a new scenario. You should classify representations, compare formats and stores, separate workloads and roles, identify relational objects and SQL statements, choose among Azure relational services, distinguish Blob, Files, Table, and Cosmos DB APIs, trace large-scale analytics, separate batch and streaming, describe Databricks and Fabric, and design a Power BI model and visualization set.
Use the five-phase roadmap, the 50 original questions, the 40 flashcards, and the three projects. Then use Microsoft's official Practice Assessment and exam sandbox. No independent practice result guarantees the certification result.
Official references
- Microsoft Certified: Azure Data Fundamentals
- DP-900 study guide — skills measured July 21, 2026
- Introduction to Microsoft Azure Data course
- Azure SQL documentation
- Azure Storage documentation
- Azure Cosmos DB documentation
- Azure Databricks documentation
- Microsoft Fabric documentation
- Power BI documentation
Continue preparing
- DP-900 five-phase roadmap
- 50 original practice questions
- 40 DP-900 flashcards
- Three Azure data projects
- Azure data engineer roadmap
- PrepKloud editorial policy
Frequently asked questions
Is DP-900 active in August 2026?
Yes. Microsoft's credential page is active and lists a July 21, 2026 update.
Which blueprint does this guide use?
The English skills measured as of July 21, 2026.
How long is the assessment?
Microsoft states that candidates have 45 minutes to complete it.
How is the independent bank allocated?
Exactly 14 Core, 12 Relational, 9 Non-relational, and 15 Analytics questions.
What official resources should be used?
Use Microsoft's DP-900 course and learning paths, official study guide, Practice Assessment, exam sandbox, and product documentation.
Are these official items or a pass guarantee?
No. They are original independent scenarios based on public first-party sources, with no official, live, recalled, copied, leaked, or dump content. No practice score guarantees a pass.