Overview
Snowflake AI Data Cloud Features and Architecture is 31% of the SnowPro Core exam, the single heaviest domain, having grown in the COF-C03 version. It covers Snowflake's defining three-layer architecture: the decoupled storage layer, the compute layer of virtual warehouses, and the cloud services layer that coordinates everything. It covers editions and their features, micro-partitions and how Snowflake stores and prunes data, and core objects like databases, schemas and tables. It adds the newer AI and format features, Cortex AI, Apache Iceberg tables and Dynamic Tables, and the tooling for working with Snowflake: Snowsight, SnowSQL, Snowpark, Streamlit and Notebooks.
The decoupled architecture is Snowflake's most important concept and the foundation for performance, scaling and sharing. The exam rewards understanding the three layers and their roles, how micro-partitions enable pruning, the edition differences, and the new AI and format capabilities.
How it's tested
SnowPro Core questions are scenario and knowledge based, testing understanding of Snowflake's platform. In this domain they describe an architecture concept, storage behavior, edition or feature and reward understanding the three layers, micro-partitions and the platform capabilities. Read for the layer, object or feature the scenario points to.
-
Which layer
A capability is described, and you identify the storage, compute or cloud-services layer responsible.
-
Storage and pruning
A data-storage or query-efficiency need is described, and you apply micro-partitions and pruning.
-
Feature or edition
A feature or requirement is described, and you identify the edition, Cortex AI, Iceberg or Dynamic Table capability.
A team needs to run a heavy data science workload and a separate reporting workload at the same time without them competing for resources. Because Snowflake decouples compute from storage, they can create two independent virtual warehouses that each access the same data in the storage layer without contention. Recognizing that separate warehouses provide workload isolation thanks to the decoupled architecture is the reasoning the exam expects, rather than resizing one shared warehouse.
Domains 4
Official domains from the Snowflake outline, in outline order.
1.1 The three-layer architecture: storage, compute, and cloud servicesDecoupled storage, compute and cloud services.
- What it means
- This covers Snowflake's three-layer architecture: the storage layer that holds data in compressed, columnar micro-partitions in cloud object storage; the compute layer of virtual warehouses that process queries; and the cloud services layer that handles authentication, metadata, query optimization, access control and transactions.
- What to understand
- Snowflake decouples storage from compute, so they scale independently and multiple warehouses can access the same data concurrently without contention. The cloud services layer is the always-on brain coordinating the platform. This separation enables elastic scaling, workload isolation and zero-copy sharing. Understanding what each layer does, and that they are independent, is the most important concept on the exam.
- How it shows up
- A capability or scaling need is described, and you must identify the responsible layer or apply the decoupling.
- Common confusion
- Not understanding that storage and compute are decoupled, or confusing the roles of the three layers.
1.2 Editions, micro-partitions, and core objectsSnowflake editions, micro-partitions and object hierarchy.
- What it means
- This covers the Snowflake editions (Standard, Enterprise, Business Critical and Virtual Private Snowflake) and their differing features, micro-partitions as the units of storage with their metadata enabling pruning, and core objects such as databases, schemas, tables, views and stages in Snowflake's object hierarchy.
- What to understand
- Each edition adds capabilities (for example, longer Time Travel and materialized views in Enterprise, enhanced security in Business Critical). Micro-partitions are small, immutable, columnar storage units whose metadata lets Snowflake prune (skip) partitions not needed by a query, which drives performance. Core objects are organized in a hierarchy. Knowing edition differences, micro-partition behavior and the object model is tested.
- How it shows up
- A feature requirement or storage concept is described, and you must identify the edition, micro-partition behavior or object.
- Common confusion
- Confusing which features belong to which edition, or misunderstanding how micro-partition metadata enables pruning.
1.3 Cortex AI, Apache Iceberg, and Dynamic TablesNewer AI, open-format and declarative features.
- What it means
- This covers features emphasized in COF-C03: Cortex AI (Snowflake's built-in AI and machine learning functions and services, such as Cortex Analyst and Cortex Search), Apache Iceberg tables (open-format tables Snowflake can manage or reference), and Dynamic Tables (declarative, automatically maintained transformation tables).
- What to understand
- Cortex AI brings large language model and analytics capabilities into Snowflake through SQL and services. Apache Iceberg tables let Snowflake work with open table formats for interoperability. Dynamic Tables declaratively define a transformation and are refreshed automatically to a target freshness. These are newer, tested additions reflecting Snowflake as an AI data cloud. Recognizing what each does is expected.
- How it shows up
- An AI, open-format or declarative-transformation need is described, and you must identify Cortex AI, Iceberg or Dynamic Tables.
- Common confusion
- Confusing Dynamic Tables (declarative, auto-refreshed) with standard tables or streams and tasks, or overlooking Iceberg and Cortex as newer content.
1.4 Snowsight, SnowSQL, Snowpark, Streamlit, and connectivity toolingInterfaces and developer tooling.
- What it means
- This covers the ways to interact with Snowflake: Snowsight (the web UI), SnowSQL (the command-line client), Snowpark (the developer framework for Python, Java and Scala), Streamlit (for building data apps in Snowflake), Notebooks, and the drivers and connectors for programmatic access.
- What to understand
- Different tools suit different users: Snowsight is the primary web interface for queries, dashboards and administration; SnowSQL is the CLI; Snowpark lets developers process data with code in Snowflake's compute; and Streamlit builds interactive apps. Drivers and connectors enable application access. Knowing which tool fits a task is the expected knowledge.
- How it shows up
- An interaction or development need is described, and you must choose the Snowflake tool or interface.
- Common confusion
- Confusing Snowpark (code-based data processing) with Snowsight (the web UI), or the roles of the various tools.
High-yield concepts
The terms, rules and formulas this domain turns into questions most often.
-
Three-layer architecture
Storage, compute, cloud services.
- Remember
- Storage and compute are decoupled.
- Tested as
- Core architecture questions.
- Trap
- Confusing the layers' roles.
-
Decoupled storage and compute
They scale independently.
- Remember
- Enables workload isolation and sharing.
- Tested as
- Architecture questions.
- Trap
- Thinking they are tied together.
-
Cloud services layer
The always-on coordinating brain.
- Remember
- Handles metadata, optimization, security.
- Tested as
- Architecture questions.
- Trap
- Confusing with compute.
-
Virtual warehouses
Independent compute clusters.
- Remember
- Separate warehouses isolate workloads.
- Tested as
- Architecture questions.
- Trap
- Thinking one warehouse must serve all.
-
Micro-partitions
Small, immutable, columnar storage units.
- Remember
- Metadata enables pruning.
- Tested as
- Storage questions.
- Trap
- Confusing with manual partitioning.
-
Pruning
Skipping partitions not needed by a query.
- Remember
- Driven by micro-partition metadata.
- Tested as
- Performance questions.
- Trap
- Not linking pruning to metadata.
-
Editions
Standard, Enterprise, Business Critical, VPS.
- Remember
- Higher editions add features.
- Tested as
- Edition questions.
- Trap
- Confusing edition features.
-
Cortex AI
Built-in AI and ML functions and services.
- Remember
- New AI capabilities in COF-C03.
- Tested as
- Feature questions.
- Trap
- Overlooking it as newer content.
-
Apache Iceberg tables
Open-format tables Snowflake works with.
- Remember
- For interoperability.
- Tested as
- Feature questions.
- Trap
- Confusing with standard tables.
-
Dynamic Tables
Declarative, auto-refreshed transformation tables.
- Remember
- Maintained to a target freshness.
- Tested as
- Feature questions.
- Trap
- Confusing with streams and tasks.
-
Snowpark vs. Snowsight
Code framework versus web UI.
- Remember
- Snowpark processes data with code.
- Tested as
- Tooling questions.
- Trap
- Confusing the two.
Side by side
Concepts candidates mix up, compared on the point that decides the answer.
The three layers
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Storage layer | Holds data in micro-partitions | Decoupled from compute | Tying it to compute |
| Compute layer | Virtual warehouses process queries | Scales independently | Thinking one warehouse serves all |
| Cloud services | Coordinates the platform | Metadata, optimization, security | Confusing with compute |
Micro-partitions vs. manual partitioning
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Micro-partitions | Automatic columnar units | Metadata enables pruning | Expecting to manage them manually |
| Manual partitioning | User-defined partitions | Not how Snowflake works | Applying other systems' model |
Dynamic Tables vs. streams and tasks
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Dynamic Tables | Declarative, auto-refreshed | Define the result, refreshed to freshness | Confusing with streams/tasks |
| Streams and tasks | Imperative change tracking and scheduling | You build the pipeline | Confusing with Dynamic Tables |
Exam traps
Each trap is an answer that looks right. Learn the clue that should change your mind.
-
Resize instead of isolate
- What candidates pick
- Resize one shared warehouse to run two competing workloads.
- Why it looks right
- A bigger warehouse seems to help.
- Why it's wrong
- Because compute is decoupled, creating separate virtual warehouses isolates the workloads so they do not contend for resources.
Clue to watch for Two workloads must run at once without competing.
-
Manage micro-partitions manually
- What candidates pick
- Expect to define and manage micro-partitions by hand.
- Why it looks right
- Other databases use manual partitioning.
- Why it's wrong
- Snowflake creates and manages micro-partitions automatically; their metadata enables pruning without manual partitioning.
Clue to watch for The scenario asks how data is partitioned for pruning.
-
Cloud services is compute
- What candidates pick
- Attribute query execution to the cloud services layer.
- Why it looks right
- It coordinates queries.
- Why it's wrong
- Virtual warehouses in the compute layer execute queries; the cloud services layer handles metadata, optimization and coordination.
Clue to watch for The task is executing a query versus coordinating it.
-
Dynamic Table is a stream/task
- What candidates pick
- Build streams and tasks when a Dynamic Table fits.
- Why it looks right
- Both support pipelines.
- Why it's wrong
- A Dynamic Table declaratively defines the transformation and is auto-refreshed to a target freshness; streams and tasks are an imperative pipeline you build.
Clue to watch for The need is a declarative, automatically maintained transformation.
-
Wrong edition feature
- What candidates pick
- Expect an Enterprise feature to be available in Standard.
- Why it looks right
- Features blur together.
- Why it's wrong
- Higher editions add features; for example, extended Time Travel and materialized views are Enterprise-level capabilities.
Clue to watch for A feature requirement implies a particular edition.
-
Snowpark is the UI
- What candidates pick
- Describe Snowpark as the web interface.
- Why it looks right
- Both are ways in.
- Why it's wrong
- Snowsight is the web UI; Snowpark is a developer framework for processing data with code in Snowflake's compute.
Clue to watch for The need is code-based data processing versus a UI.
Answer framework
For any Domain 1 architecture scenario, use these steps to find the answer the exam expects.
- Identify which of the three layers the question concerns: storage (micro-partitions), compute (virtual warehouses) or cloud services (coordination).
- Apply decoupling: storage and compute scale independently, and separate warehouses isolate workloads over the same data.
- For storage and performance, remember micro-partitions are automatic and their metadata enables pruning.
- Match features to editions, and recognize the newer Cortex AI, Apache Iceberg and Dynamic Tables capabilities.
- Choose the right tool: Snowsight for the UI, SnowSQL for the CLI, Snowpark for code, and Streamlit for data apps. Where two options seem to fit, choose the one that most precisely matches the scenario and current Snowflake behavior.
Practice question
Want more like this? The SnowPro Core practice test has questions for every domain.
Checklist
0 of 10
Tick each one you can do without notes. Your ticks stay in this browser.
Study strategy
Give this domain the most study time, since at 31% it is the largest and, with performance, over half the exam. The single most important concept is the three-layer architecture: internalize the decoupled storage, compute and cloud-services design until you can reason about scaling, workload isolation and sharing from it, since much of the exam flows from this.
Then master micro-partitions and how their metadata enables pruning, the edition differences (what Enterprise and Business Critical add), and the core object hierarchy. Make sure to study the newer COF-C03 content that many older guides miss: Cortex AI, Apache Iceberg tables and Dynamic Tables, keeping Dynamic Tables distinct from streams and tasks. Learn the tooling, especially Snowpark versus Snowsight.
Because the exam rewards platform familiarity, use a Snowflake free trial to create warehouses, inspect micro-partition behavior and try the tools. Afterwards, move to Domain 2, Performance Optimization, Querying, and Transformation. Because this exam rewards platform familiarity, use a Snowflake free trial to practice warehouses, loading, governance and sharing rather than only reading, and review missed practice questions by domain to find weak spots.
FAQ
How much of SnowPro Core is Domain 1?
Snowflake AI Data Cloud Features and Architecture is 31% of the exam, the single largest domain, and it grew in the COF-C03 version. With performance, it accounts for more than half the exam.
What is Snowflake's three-layer architecture?
A decoupled design with a storage layer (data in micro-partitions), a compute layer (virtual warehouses that process queries), and a cloud services layer (the always-on brain handling metadata, optimization, security and coordination). Storage and compute scale independently.
Why does decoupling storage and compute matter?
Because compute and storage scale independently, you can run multiple virtual warehouses against the same data without contention, isolating workloads, and scale each as needed. This also underpins features like zero-copy sharing.
What are micro-partitions?
Small, immutable, columnar units of storage that Snowflake creates and manages automatically. Their metadata lets Snowflake prune (skip) partitions a query does not need, which drives query performance without manual partitioning.
What is new in COF-C03?
COF-C03 (released February 2026) increased the weight of architecture and added Cortex AI, Apache Iceberg tables and Dynamic Tables, reflecting Snowflake's evolution into an AI data cloud. It collapsed the former six domains into five.
What is the difference between a Dynamic Table and streams and tasks?
A Dynamic Table declaratively defines a transformation and is automatically refreshed to a target freshness. Streams and tasks are an imperative approach where you track changes with streams and schedule processing with tasks to build the pipeline yourself.
How does Domain 1 connect to the rest of the exam?
The architecture underpins performance (Domain 2), is governed in account management (Domain 3), is the target of data loading (Domain 4), and enables data collaboration (Domain 5), so it is the foundation for everything else.
Test your Domain 1 knowledge
You have covered the three-layer architecture, micro-partitions, editions, Cortex AI, Iceberg and Dynamic Tables. Test whether you can apply Snowflake's architecture under exam conditions.
Keep going
- Training Training Built on your Weak areas
- Practice test Questions by domain, with explanations
- Mock exam Full length, timed like the real exam
- Readiness test Find your weakest domain in minutes
- Cheat sheet Key terms and formulas on one page
- Ask the SnowPro Core AI tutor Get any concept explained your way