DataQube vs Snowflake Cortex Analyst
Both answer questions about governed data in natural language. They differ on one thing that cannot be configured away: where the data and the model actually run.
Updated 25 July 2026
The short answer
Snowflake settles this question in its own documentation
We do not need to characterise their architecture. They describe it precisely, and for most of their customers it is the correct decision.
“You can't install and run Snowflake locally or on private cloud infrastructures, whether on-premises or hosted.”
Documentationdocs.snowflake.com — Key concepts & architectureaccessed 2026-07-25
“Snowflake is provided as a self-managed service that runs completely on cloud infrastructure. This means that all three layers of Snowflake's architecture (storage, compute, and cloud services) are deployed and managed entirely on a selected cloud platform.”
Documentationdocs.snowflake.com — Key concepts & architectureaccessed 2026-07-25
Virtual Private Snowflake is a “completely separate Snowflake environment, isolated from all other Snowflake accounts”. It remains public cloud, Snowflake-operated, and connected — and bills three times Standard per Platform Credit.
Contractdocs.snowflake.com — Snowflake editions; Service Consumption Tableaccessed 2026-07-25
Deployment
“On-premise” describes four different things
The term is used loosely enough to be meaningless in a procurement conversation. These are four distinct states, and the useful question in an evaluation is simple: if we cut the network cable, does it still answer questions?
SaaS
Vendor-operated, in the vendor's own cloud account.
Customer VPC
Runs in your cloud subscription, but a vendor-operated control plane is still required. Private networking removes the public internet, not the dependency.
True on-premise
Runs on hardware you own and operate.
Air-gapped
No outbound path exists at all — including for licensing, telemetry, updates and model inference.
Snowflake operates every layer in its own cloud account. Virtual Private Snowflake isolates the tenant; it does not move the deployment onto your hardware.
Pricing
The regulated edition is the expensive edition
The standalone Cortex Analyst rate is denominated in Platform Credits, which vary by edition. The same question therefore costs more on the security posture a regulated firm is required to buy.
Cortex Analyst bills 67 Platform Credits per 1,000 messages. That is $0.134 per message on Standard and $0.268 on Business Critical — the edition where Tri-Secret Secure and PrivateLink live.
ContractSnowflake Service Consumption Table, effective 2026-07-24accessed 2026-07-25
Data residency is a documented surcharge: $2.20 per AI Credit for regional routing against $2.00 for global — and Snowflake's documentation notes the cheaper global option offers “broader model availability”.
ContractSnowflake Service Consumption Table, effective 2026-07-24accessed 2026-07-25
Cortex Search bills continuously at 6.3 AI Credits per GB per month indexed, whether or not anyone asks a question.
ContractSnowflake Service Consumption Table, effective 2026-07-24accessed 2026-07-25
DataQube licenses by concurrent active session. The bill does not move when a team asks more questions, and it does not change with the security posture you deploy.
Provenance
What survives after the answer
Both products can show the SQL behind a figure. The difference is what remains months later, when a model validator asks what an analyst saw on a specific date.
DataQube
- The thread is an append-only record: the conversation, the generated query, the permission decisions and the artefacts, kept verbatim.
- Compaction updates the agent's working state and appends an event. It never rewrites history.
- A notebook is a projection of a thread, not a re-execution of it.
- Audit logging is part of the product, not a billable meter.
Snowflake
- Cortex Analyst is an API rather than a finished application — “Cortex Analyst takes an API-first approach, giving you full control over the end user experience.” The durable record is the one you build around it.
- Generated SQL is captured in agent observability. It is not captured in the billing usage view — a distinction repeated incorrectly even in Snowflake's own security blog.
- AI observability is opt-in, and retained events bill per LLM call, on top of the query cost of reading them back.
Fair comparison
Where Snowflake is the better choice
If the following describe you, Cortex Analyst is very likely the right answer. We would rather say so now than spend an evaluation cycle finding out.
Choose Snowflake when
- Your data already lives in Snowflake and there is no regulatory pressure to move analysis inside your own perimeter.
- You want an API to embed in an application you are building, rather than a finished analyst experience.
- Usage is wide and shallow — many occasional users asking a few questions each. Consumption pricing genuinely wins at that shape.
- You need the surrounding platform: warehousing, data sharing, the marketplace, Cortex Search.
Choose DataQube when
- Analysis has to run inside your perimeter — on-premise, private cloud, or fully air-gapped.
- Second-line review needs to reproduce what an analyst saw on a given date, not what the data says today.
- Usage is deep and concentrated, and a per-question meter makes the business case impossible to forecast.
- The sources span databases, files and internal tools rather than a single warehouse.
Questions we get in evaluations
Can Snowflake Cortex Analyst run in an air-gapped environment?
No. Snowflake's documentation states that you cannot install and run Snowflake locally or on private cloud infrastructure, whether on-premises or hosted. Virtual Private Snowflake provides tenant isolation inside Snowflake's cloud; it is not a deployment on hardware you own.
Does Cortex Analyst show the SQL it generates?
Yes. It returns a SQL statement alongside a natural-language interpretation, and the query is captured in agent observability. Note that it is not captured in the billing usage view — a claim that circulates widely and is incorrect.
Is DataQube cheaper than Cortex Analyst?
Not always, and it depends on usage shape. For a broad population asking a few questions each per month, consumption pricing is usually cheaper. Concurrent-session licensing wins where usage is concentrated and sustained, or where the cost has to be forecast a year ahead.
Can DataQube read data that lives in Snowflake?
Yes. DataQube connects to approved databases as a governed source. What changes is where the analysis and the model inference run, and what record is kept afterwards.