Methodology & sources
Every current price on this site comes from a public price source that we pull automatically. We never type current prices in by hand. The one exception is the historical backfill (see Price history), where each value is linked to the archived page it was read from.
Sources
- Azure: the official Azure Retail Prices API (
serviceName eq 'Azure Databricks'), all regions, all meters. - AWS & Google Cloud: the pricing data that powers the instance-type calculator on databricks.com/product/pricing.
- Azure VM prices & DBU ratings (calculators only): Azure's public pricing-calculator feed.
- AWS EC2 prices (calculators only): on-demand Linux, shared-tenancy list prices from the public JSON behind AWS's EC2 On-Demand pricing page, for the 569 instance types Databricks supports in 17 commercial regions. Spot-checked against the official AWS Price List bulk API. GovCloud and a few retired types (i2, r3) aren't priced.
- Google Compute Engine prices (calculators only): on-demand list prices read from Google's official VM pricing pages (general-purpose, compute-optimized and memory-optimized), for 149 of the machine types Databricks supports in 15 regions. Google's Cloud Billing Catalog API needs an API key, so we use the public pages. GPU machine types aren't priced yet, and "-lssd" variants are shown at the standard machine price (Local SSD extra).
Price history
The price history combines three kinds of observations, each tagged with its source in the history CSV:
- DBU Watch daily captures (since October 15, 2026): every price we pull, stored as dated snapshots in the public repo. Because the job runs daily and records any change, a change between two captures is dated to the day it was detected.
- Azure effective dates: each Azure Retail Prices API record carries an
effectiveStartDate, Microsoft's own date for when the current price took effect. We plot the current price from that date. The API doesn't publish the price before it, so earlier Azure prices only appear where an archived page shows them. - Internet Archive captures (2019 to 2025): public Wayback Machine copies of databricks.com's AWS and Google Cloud pricing pages, an archived copy of the databricks.com pricing feed, and Azure's pricing page. Values were read from each capture and are linked to it; the capture date is when the price was listed, not when it changed. When a page called a product by an older name (e.g. "Data Engineering" in 2019, "SQL Compute (Preview)"), the CSV notes the original label.
Caveats: archived pages exist only for some dates, list only headline rates (mostly US/"all regions" classic compute), and from 2023 databricks.com renders prices with JavaScript, so captures between early 2023 and the feed capture of September 2025 carry no readable prices. Between two observations months apart we don't know when a price changed, and the charts draw those stretches dashed. We never estimate or interpolate a price.
Process
Once a day a script downloads each source, normalizes it to cloud, product, tier, region, unit and price, and compares it with the previous snapshot. Price changes, new SKUs and removed SKUs go to the latest changes log on the price history page, and a dated snapshot is saved to the history archive. If a source fails, we keep its previous prices instead of recording false removals.
Independence
DBU Watch is independent. It is not affiliated with or endorsed by Databricks, Inc., Microsoft, Amazon or Google. "Databricks" is a trademark of Databricks, Inc.; other names are trademarks of their owners. Prices are public list prices in USD, for information only; your contract and the official price pages always win.
Data license & downloads
The DBU Watch datasets (latest prices as JSON, full price history as CSV) are free to reuse under the Creative Commons Attribution 4.0 (CC BY 4.0) license, with attribution to "DBU Watch" and a link to dbuwatch.com. Downloads require a free email signup. Prices themselves are facts published by Databricks and the cloud providers.
Download data (CSV) Latest prices (JSON)
Caveats
- These are public list prices in USD. Negotiated discounts, committed-use contracts and DBCU pre-purchase aren't reflected.
- Classic compute is billed twice: Databricks bills DBUs and your cloud provider bills the VMs. The price table shows the DBU part. The calculators add the VMs at on-demand list prices. They leave out storage and disks, networking and egress, Savings Plans, reserved instances, committed-use discounts, spot VMs and negotiated discounts.
- VM prices refresh daily with the rest of the data (last VM refresh: AWS 2026-10-17 12:57 UTC, GCP 2026-10-17 12:57 UTC). If a VM price source fails, the calculators keep the last good prices.
- For AWS and GCP, classic compute has one list price for every region ("All regions"). Serverless and SQL prices vary by region.
- Where a source lists two values for the same price (for example 1.24 and 1.243), we keep the more precise one and flag real conflicts in the raw data.
Last refresh: Oct 17, 2026 12:57 UTC. Source errors on last run: none.