Google Cloud 1 renamed Dataproc 1 to Managed Service for Apache Spark 129 (Modern Data Stack Layers); its pricing page now reads "Managed Service for Apache Spark (formerly Dataproc)". It keeps the two deployment models. Clusters run on Compute Engine VMs that you pay for, plus a management fee of $0.010 per vCPU-hour across all nodes, billed per second with a one-minute minimum. The serverless model (formerly Dataproc Serverless) bills per second in Data Compute Units: $0.06 per DCU-hour on the standard tier and $0.089 on the premium tier in us-central1, plus shuffle storage from $0.000054795 per GiB-hour. Google's own example prices a batch job with 12 cores (a 4-core driver and two 4-core executors) as 12 DCUs. Both models offer the Lightning Engine, Google's vectorized execution engine, which Google claims runs up to 4.9 times faster than open-source Spark; on clusters it costs an extra $0.0025 per vCPU-hour from 1 June 2026.
A serverless batch is one gcloud command; the same orders_daily.py reads and writes Cloud Storage (adapted from Google's quickstart; not run here):
gcloud dataproc batches submit pyspark gs://booknest-jobs/orders_daily.py \
--region=us-central1 \
-- gs://booknest-lake/marts/orders_dailyThe command group is still called dataproc after the rename. Google sizes and scales the workers, bills the DCUs used per second, and keeps logs and the Spark UI in the console.