What is Snowflake AI Data Cloud?
“What is Snowflake?” questions appear on most exams in formats that catch candidates who only know the marketing pitch. Today defines what Snowflake is and what it is not.
Snowflake overloads everyday English words with precise technical meanings. Three of these terms collide directly with intuitions from other platforms.
| Word you know | What it usually means | What it means in Snowflake |
|---|---|---|
| Warehouse | A building where Amazon stores boxes | Compute. A named cluster of servers that runs your queries. It holds no persistent data. The most overloaded word in Snowflake’s vocabulary. |
| Cloud | AWS / Azure / GCP themselves | Snowflake runs on a cloud. It is not one. “AI Data Cloud” is the platform brand. The exam writes distractors that flip this. |
| Account | Your personal login | An entire isolated tenant. One account hosts many users, roles, databases, and warehouses. Your login is a user inside an account. |
| Credit | Loyalty points | The unit Snowflake bills compute in. List rate sits roughly between $2 and $4 per credit depending on edition and cloud. Standard is cheapest, Business Critical highest. |
| Stage | A platform at a train station | A named location where data files live before being loaded into a table. Internal or external. We unpack this properly on Day 29. |
Rule of thumb for Week 1: if a Snowflake term feels like a generic English word, assume Snowflake has given it a specific technical meaning. By Day 7 the habit becomes automatic.
Today’s Concept
Micro-Concept 1: The One-Liner
Snowflake is a cloud-native AI Data Cloud platform delivered as a fully managed SaaS service.
Each modifier in that sentence carries weight. The exam writes wrong answers that fail one or two of them. The rest of today explains each modifier.
One core of data. Every workload. Any cloud.
One copy of your data feeds six workloads. The same platform runs on any major cloud.
Micro-Concept 2: Cloud-Native ≠ Cloud-Hosted
Cloud-native is the first distinction the exam tests. Many candidates get it half right. Older database vendors took their on-premises product, packaged it in VMs, and ran it on AWS or Azure. That is cloud-hosted. Same code, different hardware. Snowflake was designed for the cloud from the first commit. It was founded in 2012 and released in 2014. There has never been an on-premises build target. No “Snowflake Enterprise Server” exists to install in a data centre.
Being cloud-native is what lets the architecture do four things older databases physically cannot:
→ Zero hardware management. No servers, disks, OS patches, or engines to tune.
→ Storage and compute scale independently. They sit on different layers. Day 2 is built around this.
→ Pay-per-use billing. You pay for seconds compute is actually running, not provisioned capacity.
→ Storage that manages itself. No manual indexing, partitioning DDL, or statistics jobs. Micro-partitions handle it (Day 10).
Older databases tie compute to provisioned hardware that runs whether work is happening or not. Snowflake separates the two. Compute stops billing when queries stop running.
Micro-Concept 3: It Runs ON Cloud Providers, It Is NOT One
This is one of the most common Day 1 traps. Snowflake runs on exactly three cloud providers:
→ Amazon Web Services (AWS)
→ Microsoft Azure
→ Google Cloud Platform (GCP)
That is the full list. The exam slips Oracle Cloud or IBM Cloud into multi-select answer sets. A tired candidate at minute 90 may check four boxes because the fourth option also sounds like a real cloud.
Snowflake is not a cloud provider itself. It does not sell EC2-equivalent compute, S3-equivalent storage, or raw networking. It is a software platform that consumes those services from the cloud providers. It exposes a unified SQL surface on top.
Account creation locks in the cloud and region pair. You pick both at signup. That pair is fixed for the life of the account. There is no “switch cloud” button. No support ticket changes it. For a different cloud or region, you create a new account in the target and use replication to copy data over. Wrong answers will sound reasonable, such as “contact support to migrate.” That option does not exist.
Micro-Concept 4: SaaS Delivery Model
“SaaS” gets used loosely. Worth being precise. Nothing needs to be installed on your laptop except optional clients (SnowSQL, the new snow CLI, drivers). The platform runs in the browser. What you get on signup:
→ A web UI called Snowsight. Worksheets, dashboards, admin, and Query Profile (Day 4 takes a full tour).
→ Automatic upgrades. Every account in a region runs the same build. No “we’re on 8.4, they’re on 9.1” fragmentation.
→ Zero maintenance work. No VACUUM, REINDEX, statistics jobs, or version migrations.
The exam tests this through a “shared responsibility” framing. The line is clean. Snowflake owns infrastructure, security patches, version upgrades, query optimization, and storage compression. The customer owns the data, the queries, access control (roles and grants), and warehouse sizing. “Applying security patches” as customer responsibility is the trap. It is never the customer’s job.
Micro-Concept 5: “AI Data Cloud”, Not Just a Warehouse
The COF-C03 exam launched on February 16, 2026, replacing COF-C02 (which retires May 14, 2026). The shift from “data warehouse” to “AI Data Cloud” reflects a real product expansion. Snowflake today is a unified platform for six workload families, not one:
| Workload | What it means | Key feature |
|---|---|---|
| Data Warehousing | Traditional structured analytics | Virtual Warehouses, SQL |
| Data Engineering | ETL/ELT pipelines | Streams, Tasks, Dynamic Tables |
| Data Lake | Semi-structured and unstructured data | JSON/Parquet support, External Tables, Iceberg |
| Data Sharing | Live data across orgs. No copies | Secure Data Sharing, Marketplace |
| Data Applications | Build apps directly on the platform | Snowpark, Native Apps, Streamlit |
| AI & Machine Learning | LLMs and ML running next to your data | Cortex AI, Snowflake ML, Notebooks |
Why this matters for COF-C03: the older C02 exam barely touched AI/ML. C03 dedicates a sub-objective to Snowpark, Cortex, Notebooks, and Streamlit. Expect roughly 5 to 7 questions on these. Today just know these workloads exist under one roof. Week 3 (Days 15 to 17) goes into detail.
Micro-Concept 6: Consumption-Based Pricing (Two Meters)
Snowflake bills on two primary dimensions. A third meter (Cloud Services) usually nets to zero. It is explained below.
| Cost dimension | What’s measured | How it works |
|---|---|---|
| Compute | Processing power (virtual warehouses) | Measured in credits. Per-second billing with a 60-second minimum on every resume. Bigger warehouse = more credits per hour. |
| Storage | Data at rest | Measured in TB/month on compressed columnar data. Typically around 3× compression on raw load. Includes Time Travel and Fail-safe overhead. |
The third dimension is Cloud Services. This is the layer handling authentication, query compilation, metadata, and security. Per Snowflake’s docs, Cloud Services usage is billed only when daily Cloud Services credits exceed 10% of that day’s warehouse credit usage. Most workloads stay below the threshold and pay zero. Exam keyword: “10% daily.” Data egress (cross-region/cloud transfer) and serverless features (Snowpipe, Search Optimization, Dynamic Tables) are covered on Days 26 and 32. For Day 1: compute and storage are the two meters that matter.
The key economic effect: compute and storage are billed separately. You can park 100 TB of cold data and pay almost nothing if you never query it. This is mechanically impossible on most legacy databases where storage and compute are tied to the same provisioned server. Day 2 explains the architecture that enables this.
Cheat Sheet
| Concept | What to remember | Exam keyword |
|---|---|---|
| Architecture type | Cloud-native. Built for cloud from scratch, never on-premises | “Cloud-native” |
| Delivery model | Fully managed SaaS. Nothing to install on a server | “SaaS” |
| Cloud providers | Runs ON AWS, Azure, GCP. Is NOT one | “Multi-cloud” |
| Account creation | Pick cloud + region. Permanent per account, no swap | “Region selection” |
| Platform name | AI Data Cloud (not “data warehouse” on C03) | “AI Data Cloud” |
| Pricing | Credits (compute) + TB/month (storage) + Cloud Services | “Pay-per-use” |
| Compute billing | Per-second, 60-second minimum on each resume | “60-second minimum” |
| Cloud Services billing | Only billed above 10% of daily warehouse credits | “10% daily” |
| Maintenance | Zero. Snowflake handles upgrades, patches, tuning | “Fully managed” |
| Key differentiator | Storage and compute billed separately. Scale independently | “Decoupled” |
Exam Tip
Snowflake is a platform, not a database. On a recent SnowPro Advanced exam, a key question turned on picking cloud-native AI Data Cloud platform. The wrong options were “cloud database” and “cloud data warehouse.” Half the answer set will look reasonable.
Two traps to watch on Day 1 material. First, “Snowflake is a cloud provider” is false. It runs on cloud providers. It does not sell raw infrastructure. Second, “you can change the cloud provider after account creation” is false. The cloud and region pair is permanent. The only path is a new account plus replication. Wrong answers will pretend a support ticket or a setting toggle exists. Neither does.
Hands-On Lab
Sign up for your Snowflake trial. Head to signup.snowflake.com. Pick Enterprise edition. Standard saves a few credits but locks you out of multi-cluster warehouses, 90-day Time Travel, materialized views, and search optimization. We use all four in later labs. Start on Standard and you will hit a wall by Day 3. Any cloud and region works for this series.
Log in to Snowsight and orient yourself. Once your account is active, log in and spend two minutes locating these in the left navigation:
→ Databases. Where your data objects live.
→ Warehouses. Your compute resources. This is where you spend money.
→ Worksheets. Where you write and run SQL.
→ Marketplace. Free and paid datasets shared by other accounts.
SNOWFLAKE_SAMPLE_DATA. You did not load it. Snowflake shared it in at account creation. This is Secure Data Sharing in production (Week 7). Live read-only data can appear in your account from someone else’s account with no file transfer involved. That capability is what the “Data Cloud” branding rests on.Run your first query. Open a new Worksheet and run this against the default warehouse:
SELECT
CURRENT_VERSION() AS snowflake_version,
CURRENT_ACCOUNT() AS account_name,
CURRENT_REGION() AS region,
CURRENT_CLIENT() AS client_type;region output returns something like AWS_US_WEST_2 or AZURE_WESTEUROPE. That string is the cloud-and-region pair you picked at signup. It is locked for the life of the account. CURRENT_VERSION() shows the build your account is on. Every account in your deployment runs that same build because Snowflake rolls upgrades centrally.Verify the shared sample data.
SHOW DATABASES LIKE 'SNOWFLAKE_SAMPLE_DATA';origin column points to a Snowflake-managed share, not your account. This database is read-only and shared in from outside. We lean on it heavily through the 50 days because every trial account gets it for free.Create your dedicated lab warehouse. This is lab_xs. We reuse it through all of Week 1 and beyond, so do not drop it at the end:
CREATE WAREHOUSE IF NOT EXISTS lab_xs
WAREHOUSE_SIZE = 'XSMALL'
AUTO_SUSPEND = 120
AUTO_RESUME = TRUE
COMMENT = 'SnowPro 50-Day lab warehouse';
-- Verify it was created
SHOW WAREHOUSES LIKE 'LAB_XS';AUTO_SUSPEND = 120 means it idles down after 120 seconds of inactivity. AUTO_RESUME = TRUE wakes it on the next query. The full pay-per-use model in five lines. Pay only when work is happening, with a 60-second minimum every spin-up.Run a real query against sample data.
USE WAREHOUSE lab_xs;
SELECT
COUNT(*) AS total_orders,
MIN(O_ORDERDATE) AS earliest_order,
MAX(O_ORDERDATE) AS latest_order
FROM SNOWFLAKE_SAMPLE_DATA.TPCH_SF1.ORDERS;MIN and MAX are technically answerable from metadata without scanning rows (Day 2 and Day 40). The COUNT(*) is what pulls the warehouse into work.Clean up. Manually suspend the warehouse instead of waiting for auto-suspend:
ALTER WAREHOUSE lab_xs SUSPEND;
-- Keep the warehouse itself — we reuse lab_xs all of Week 1.SUSPEND and AUTO_SUSPEND. Running both in this lab cements which one you control directly.Snowflake Documentation
Official docs for today’s topics. The exam pulls directly from these.
External References
Community resources to reinforce today’s concepts.
Practice Questions
Options:
A. An on-premises data warehouse that can be deployed in the cloud
B. A cloud-native AI Data Cloud platform delivered as SaaS
C. A cloud infrastructure provider similar to AWS
D. An open-source database management system
Why B: All three modifiers carry weight. Cloud-native rules out anything ported from on-premises. AI Data Cloud platform rules out the narrower “warehouse” framing C03 has moved away from. SaaS rules out anything you would install yourself. “Best describes” questions test whether you can pick the answer that hits all three.
Why not A: Snowflake has never had an on-premises build. Cloud-hosted is what you get when a legacy vendor ports its binary onto VMs. Snowflake was designed for the cloud from inception.
Why not C: This is the most common Day 1 trap. Snowflake runs on AWS, Azure, and GCP. It does not sell raw compute, storage, or networking. “Cloud provider” as a Snowflake descriptor is almost always a distractor.
Why not D: Snowflake is proprietary, closed-source commercial software. Do not confuse it with PostgreSQL forks or other OSS engines.
Options:
A. Change the cloud provider in Account Settings
B. Contact Snowflake Support to migrate the account
C. Create a new account on Azure and use replication to move data
D. Snowflake automatically migrates accounts between clouds
Why C: The (cloud, region) pair is fixed at account creation and is not mutable. The supported path is always the same. Provision a new target account, configure account-level replication, fail over. The same logic applies for region moves within the same cloud.
Why not A: No such setting exists. The exam writes this distractor because it sounds reasonable. Most SaaS products let you change region. Snowflake does not.
Why not B: Support can help you plan replication, but they have no internal tool to in-place migrate an account between providers. This catches candidates who default to “ask support” for hard scenarios.
Why not D: No automatic migration exists. Customer-initiated replication is the only mechanism.
Options:
A. Applying security patches to the platform
B. Configuring access control (roles and privileges)
C. Managing hardware capacity and upgrades
D. Choosing virtual warehouse sizes for workloads
E. Upgrading to the latest Snowflake version
Why B: Snowflake ships the RBAC framework. The customer designs the role hierarchy, creates roles, grants privileges, and assigns users. Day 18 covers this in depth. Getting it wrong is one of the most common production governance failures.
Why D: Warehouse sizing is the highest-leverage cost decision the customer owns. Snowflake offers the menu (XS through 6XL). You pick which to spin up and when. Day 8 unpacks the sizing trade-offs.
Why not A, C, E: All three are Snowflake’s responsibility in the SaaS model. Patches, hardware, and version upgrades happen invisibly. You never tune an engine, schedule an outage window, or run an installer. That is the whole pitch. If any of these show up as customer responsibility, they are distractors.
Options:
A. Amazon Web Services (AWS)
B. Oracle Cloud Infrastructure (OCI)
C. Microsoft Azure
D. Google Cloud Platform (GCP)
E. IBM Cloud
Why A, C, D: AWS, Azure, and GCP make up the full list of cloud providers Snowflake supports. Treat these as a fixed trio. A classic exam pattern is three correct options plus one well-known cloud distractor (usually OCI). A tired candidate at minute 90 will check four boxes because “OCI is a real cloud, isn’t it?”
Why not B: Oracle Cloud is a real cloud provider, but Snowflake does not run on it. This is the most common multi-select trap on Day 1 material.
Why not E: IBM Cloud is not a Snowflake target. Pure distractor.
Options:
A. Number of users in the account
B. Compute, measured in credits
C. Number of databases created
D. Storage, measured in TB per month
E. Number of queries executed
Why B: Compute is metered in credits. Credits accrue per second while a virtual warehouse runs, with a 60-second minimum on each resume. Bigger warehouse = more credits per hour. This is the meter most production teams optimise hardest.
Why D: Storage is billed per TB per month based on daily-average compressed bytes. This includes Time Travel and Fail-safe overhead. Day 11 (table types) and Day 44 (Time Travel) make the overhead concrete.
Why not A, C, E: No per-user, per-database, or per-query fee. Also no per-table or per-role charge. The model is intentionally usage-based, not seat-based or object-count-based. This is one of the key differences from legacy database licensing.
Today you learned: Snowflake is a cloud-native AI Data Cloud platform. It is not a database and not a cloud provider. It runs on AWS, Azure, or GCP as fully managed SaaS. Pricing has two main meters: credits for compute, TB per month for storage. Cloud Services is a third meter that only bills above 10% of daily warehouse credits.
Key takeaway: Cloud-native means storage and compute are billed separately at the architecture level and scale independently. This design choice underlies the next 49 days of content.
Tomorrow (Day 2): Snowflake’s three-layer architecture (Storage, Compute, Cloud Services). This is the most heavily tested single concept on the exam. Once the layers are clear, the rest of Domain 1 follows.
