Databricks model serving pricing: $/DBU on AWS, Azure and GCP by region
Databricks Model Serving (billed as Serverless Real-Time Inference) has one of the simplest price sheets on the platform: a single rate per DBU that depends on cloud and region, not on tier. The catch is how many DBUs an endpoint burns per hour, especially on GPUs.
The short answer
- The cheapest list rate is $0.07 per DBU, and all three clouds hit it in their main US regions: AWS US East (N. Virginia), US East (Ohio) and US West (Oregon); Azure East US, East US 2, North Central US, West US 2 and West US 3; Google Cloud Iowa, Oregon, South Carolina, Virginia and Nevada.
- Outside those regions the rate climbs. Brazil is the most expensive standard commercial region on AWS ($0.112) and Google Cloud ($0.11), about 60% above the US rate. On Azure, Switzerland West is $0.12 and the two Jio India regions are $0.1702.
- Tier doesn't change the price. On AWS and Google Cloud, Premium and Enterprise list the same model serving rate in every region tracked. Azure only sells it on Premium, which Databricks says corresponds to Enterprise on AWS and GCP.
- Model serving is serverless, so there's no separate cloud VM bill to add on top, unlike classic clusters.
Model serving $/DBU by region
AWS (Enterprise and Premium are identical), 15 regions:
| Region | $/DBU |
|---|---|
| US East (N. Virginia), US East (Ohio), US West (Oregon) | $0.070 |
| AP (Mumbai) | $0.074 |
| Canada, Europe (Ireland) | $0.078 |
| Europe (London) | $0.081 |
| Europe (France), US West (N. California) | $0.082 |
| Europe (Frankfurt) | $0.084 |
| AP (Seoul) | $0.086 |
| AP (Singapore), AP (Sydney) | $0.088 |
| AP (Tokyo) | $0.090 |
| SA (Brazil) | $0.112 |
Google Cloud (Enterprise and Premium are identical), 14 regions:
| Region | $/DBU |
|---|---|
| Iowa, Oregon, South Carolina, Virginia, Nevada | $0.07 |
| Canada (Quebec), Belgium, England, Frankfurt | $0.08 |
| Singapore, Tokyo, Sydney, Dammam | $0.09 |
| SA (Brazil) | $0.11 |
Azure (Premium, pay-as-you-go retail price), 40 regions. Selected regions:
| Region | $/DBU |
|---|---|
| East US, East US 2, North Central US, West US 2, West US 3 | $0.070 |
| Central India, Sweden Central | $0.074 |
| North Europe, Canada Central | $0.078 |
| UK South | $0.081 |
| West Europe, Germany West Central | $0.084 |
| Australia East, Southeast Asia | $0.088 |
| Japan East | $0.090 |
| Brazil South | $0.112 |
| Switzerland West | $0.120 |
| Jio India West, Jio India Central | $0.1702 |
Every region and its price history is on the history pages for AWS, Azure and Google Cloud.
What a GPU endpoint costs per hour
The per-DBU rate is only half the bill. Databricks publishes how many DBUs each GPU serving size uses per hour on its model serving pricing page. Multiplied by the $0.07 US list rate:
| GPU size | DBUs/hour | $/hour at $0.07 | 730 hours (a month, always on) |
|---|---|---|---|
| T4, 1 GPU | 10.48 | $0.73 | $536 |
| A10G, 1 GPU | 20 | $1.40 | $1,022 |
| L40S, 1 GPU | 44.86 | $3.14 | $2,292 |
| H100, 1 GPU | 100 | $7.00 | $5,110 |
| A10G, 4 GPU | 112 | $7.84 | $5,723 |
| H100, 8 GPU | 800 | $56.00 | $40,880 |
Region matters at this scale. The same 1-GPU H100 endpoint at the AWS Brazil rate ($0.112) works out to $11.20 an hour, or $8,176 for an always-on month.
Special SKUs that cost more
- SAP: Model serving for SAP Databricks is billed under its own SKU at $0.125 per DBU in AWS US East (N. Virginia) and Google Cloud Iowa, about 79% above the standard rate in the same regions, and up to $0.199 in AWS Brazil.
- Government: AWS GovCloud (US Gov West 1) and Azure US Gov Arizona and US Gov Virginia list $0.088 per DBU.
- Anthropic models on Azure: Azure lists separate meters, "Anthropic Serverless Inference" at $0.105 per DBU (global) and "Model Serving for Anthropic" at $0.14 per DBU (global, effective September 1, 2026). Check which meter your usage lands on before applying the standard rate.
Has model serving gotten more expensive?
Not in the data. In the Azure regions with the oldest price records, including East US ($0.07) and West Europe ($0.084), the current retail price has carried the same effective date since April 1, 2023. DBU Watch's daily checks have detected no model serving list-price change since tracking started on October 15, 2026. Azure regions added from 2024 to 2026 each list a single price with no earlier one. See the price history page for any future change, or download the full history.
Keeping model serving costs down
- Put endpoints in a $0.07 region when latency and data residency allow; a region at $0.09 costs about 29% more for the same DBUs.
- Size GPUs to the model. Going from a 1-GPU A10G (20 DBUs/hour) to a 1-GPU H100 (100) is five times the DBUs.
- Always-on hours drive the monthly bill. An endpoint that serves traffic only part of the day costs far less than the 730-hour figures above, so check whether scale-to-zero fits the workload's latency needs.
- For batch scoring that doesn't need a live endpoint, compare against jobs compute with the serverless vs classic guide. New to DBUs? Start with Databricks pricing explained.
Prices are public list prices (AWS and Google Cloud from Databricks' pricing data; Azure from the Azure Retail Prices API) as of October 27, 2026. Discounts and committed-use contracts aren't included.
Get the monthly pricing digest
Every Databricks price change, once a month. Free.
We only use your email for the monthly digest. Unsubscribe anytime with one click. Privacy