Revolgy blog

Migrating to BigQuery from AWS or Azure: How to decide and where to start

Written by Muskan Goel | July 7, 2026

BigQuery can be the cheapest data warehouse a team runs or one of the most expensive, and which one you get has little to do with how it stacks up against Snowflake or Redshift. It comes down to whether your workload fits the way BigQuery charges, and whether your queries are written to respect it.

The default on-demand model bills $6.25 for every terabyte a query scans, with the first 1 TiB each month free, as of June 2026.

  • A team with spiky pipelines that sit idle for hours can pay far less than it would for a Redshift cluster running around the clock.
  • A team scanning wide, unpartitioned tables on a schedule can watch a single SELECT* on 10 TiB cost $62.50 every run, several times a day.

The bill follows the shape of the workload and the discipline of the queries.

For mid-market teams on AWS or Azure, this becomes a live question at a predictable moment: a warehouse that was fine at an earlier stage starts to strain. Queries slow down, the monthly bill climbs, and an engineer is spending real hours keeping it running. The proposal to migrate to BigQuery feels right at that moment.

Most of the advice on this decision comes from people who want the migration to happen.

Jack Davies, a Data and AI Cloud Engineer at Revolgy who runs these assessments, starts from the opposite position:

“It’s not guaranteed that moving will necessarily be the best choice for your business.”

That is an unusual thing for a Google Cloud partner to say, and it is the right starting point.

“It comes down to what your current setup costs you and how your workloads actually run, and sometimes the answer is to stay put.”

The question worth answering is narrower than a feature comparison: whether your specific workload is one of the cases where moving pays back. The only way to know is to profile it, and that has to happen before you commit.

This guide, based on a conversation with Jack Davies, Data and AI Cloud Engineer at Revolgy, and Miro Vlasák, CEO of Revolgy, covers when migrating to BigQuery pays back, the questions that decide it, and how to test the value on your own data before committing to a migration.

 

 

 

What migrating to BigQuery actually changes, and when it makes sense to migrate

Migrating to BigQuery is a change in architecture before it is a change in vendor: serverless compute, storage decoupled from that compute, and billing tied to the data each query reads. Each of those shifts maps to a situation where the move earns its cost.

 

Image source: Google

 

BigQuery pays off when operational overhead is your real cost

BigQuery is fully serverless, with storage and compute scaled by Google rather than provisioned by you. There are no clusters to size, no tables to vacuum, no sort or distribution keys, and no capacity planning to get wrong. If your current warehouse costs an engineer hours every week in tuning and upkeep, migrating moves that time back to data work, and removes the failure mode where a mis-sized cluster throttles queries or burns budget.

BigQuery fits spiky and analytical workloads

On-demand billing charges only for the data your queries scan, so idle hours cost nothing. A provisioned Redshift cluster bills around the clock whether it serves a query or not, which makes bursty, scheduled, or unpredictable analytical workloads a natural fit for BigQuery’s model. The engine is built for large parallel scans, so the heavier and wider the analytical query, the more that fit holds.

BigQuery removes pipeline work when your data already lives in Google’s products

GA4, Google Ads, and Search Console export into BigQuery natively, with no third-party connector or ETL tool in between. For a team whose data already sits in Google’s analytics surfaces, that strips out a layer of pipeline engineering the same data would need to reach Redshift or Snowflake. The closer your sources sit to Google, the stronger the case.

BigQuery shortens the path to machine learning on warehouse data

BigQuery ML lets analysts build and run predictive and classification models in SQL, against data already in the warehouse, without exporting it to a separate stack. Here Jack’s detail earns its place:

“Your data engineers and analysts can easily put together a machine learning model using the same data set that you have.”

For teams that want to test models on warehouse data before standing up dedicated ML infrastructure, that removes a whole staging step.

 

When BigQuery is the wrong tool for the job

BigQuery is an analytical warehouse, and several common workloads sit outside what it does well.

BigQuery handles analytics, not real-time transactions

BigQuery’s strength is querying large volumes of data for analytics and machine learning, and that includes historical transaction data. It supports ACID transactions on data in its own storage, so that is not the constraint. The constraint is real-time throughput: high-frequency row-level updates and constant small writes run slowly and cost more than they should. Because a transaction or DML statement that mutates a table blocks others on the same table from running at the same time.

BigQuery is not built for sub-second, customer-facing queries

As a batch analytical engine, BigQuery does not deliver the sub-second response that operational dashboards and customer-facing apps expect at high concurrency. For that pattern, a low-latency store like Bigtable or a purpose-built real-time engine fits better. Pointing BigQuery at an application-serving workload usually ends in slow queries and a surprising bill.

BigQuery needs active cost governance, which not every team has

On-demand pricing pushes cost control down to the level of every query and every analyst, and BigQuery does not offer a hard spending cap at the account level. The guardrails are per-query and per-project: a maximum-bytes-billed limit on a query, custom daily quotas, and billing alerts. A team that sets and enforces those controls keeps spend predictable. A team that opens self-serve access without them is the wrong fit, because nothing in the platform stops a careless query from becoming a recurring bill.

 

 

 

The cost variables that decide whether a BigQuery migration pays back

Jack’s assessment weighs the same factors against the workload a team actually runs, starting with “what are the added costs” and “who is managing infrastructure around this.” Four variables decide whether the move pays back:

 

 

  • Operational overhead you remove. The engineer-hours your current warehouse consumes in tuning, vacuuming, and capacity planning are real money that a serverless model gives back.
  • Egress to leave your current cloud. Moving data off AWS or Azure incurs outbound transfer charged by the cloud you are leaving, around $0.08 to $0.12 per GB on standard tiers, as of 2026. Negligible at a few terabytes, but material at migration volumes.
  • Migration rework. SQL dialects differ, partitioning and distribution strategies do not carry across, and undocumented jobs have to be found and rebuilt before cutover. This is the line item teams underestimate most.
  • Workload shape against the pricing model. Spiky querying suits on-demand pricing while steady, heavy processing moves to committed capacity, a shift most analyses place around 300 to 500 TiB processed per month.

How to test BigQuery on your own data before you commit

You can answer the migration question with evidence, before moving anything into production.

Step 1: Profile your current workload with the migration assessment

Run the free BigQuery migration assessment against your exported query logs and metadata. It returns a report that models your BigQuery cost against today, shows which queries translate automatically versus which need manual rework, and flags tables with no writes. That answers the rework and fit questions before you move a single table.

Step 2: Run your real queries in place with BigQuery Omni

BigQuery Omni runs the engine inside your AWS or Azure region and queries data in S3 or Blob Storage without copying it, including the EU regions in Ireland and Frankfurt. Point it at one real workload, run the queries you care about, and measure against your current setup. Know the limits: Omni needs on-demand or Enterprise edition pricing, caps cross-cloud join transfers at 60 GB, and charges egress when results cross clouds.

Step 3: Translate and validate a representative slice of queries

Use the SQL translation and data validation tools on your heaviest and most depended-on queries, rather than the whole estate. The translation report quantifies the manual effort honestly, and validation confirms the converted queries return the same results. This turns the rework cost from a guess into a number.

Step 4: Decide on the measured result

Compare the measured cost, the rework effort, and the workload fit against staying put. If the pain is transactional or real-time serving, the honest answer may be to keep the warehouse and solve the problem elsewhere. Jack’s own first step is the same conversation: a “departmental meeting, get an understanding of what’s going well and what’s going wrong.”

 

Talk to a Revolgy expert before you migrate to BigQuery

If you have read this far and still cannot tell whether your workload is one where BigQuery pays back, or whether it is the wrong tool for what you run, that uncertainty is the starting point for an Architecture Clinic with Revolgy.

Revolgy has profiled both kinds of workload: the ones where the move cut costs and removed operational toil, and the ones where the honest call was to stay put. That is the judgment worth borrowing before you spend anything on moving data.

Revolgy runs free 45-minute 1:1 working sessions with senior Google Cloud architects. Bring your current warehouse setup, your heaviest queries, and the question you cannot answer internally. Within 24 hours of the session, you receive a structured summary of the observed gaps and recommended next steps.

 

 

 

Frequently asked questions

Is it cheaper to move from Redshift or Snowflake to BigQuery?

Not automatically. At the volumes that prompt a migration, egress and the cost of rebuilding pipelines often outweigh any difference in compute rates, so the saving depends on your workload shape and how much you spend maintaining your current setup. Profiling the workload is the only way to get a real number.

Can I use BigQuery without moving off AWS or Azure?

Yes. BigQuery Omni runs BigQuery compute inside your AWS or Azure region and queries data in Amazon S3 or Azure Blob Storage without copying it to Google Cloud. It is available in a fixed set of regions, including AWS Europe in Ireland and Frankfurt.

When should you not use BigQuery?

When your workload is transactional, real-time, or customer-facing at high concurrency. BigQuery is an analytical engine and does not provide ACID transactions or sub-second serving, so high-frequency updates and application-serving belong in Cloud SQL, Spanner, or Bigtable.

How long does a BigQuery migration take?

It depends on workload complexity and hidden dependencies rather than data volume alone. The free BigQuery migration assessment and a query-in-place test with BigQuery Omni can give you a read on cost and effort in days, before any production migration starts.

Does BigQuery lock you into Google Cloud?

Yes. BigQuery runs only on Google Cloud, though BigQuery Omni lets you query data in AWS and Azure in place. If cross-cloud portability is a firm requirement, Snowflake is the cloud-agnostic alternative.