THE OPEN DATABRICKS ALTERNATIVE

Open, faster, flexible lakehouse that runs anywhere

Same Apache Spark™, same SQL, on open tables.
Faster than Photon without the Photon tax, on any Kubernetes you already run.

GET STARTED FREE

Where Databricks runs versus where Onehouse runs. Databricks: serverless preferred, hyperscaler clouds only, so AWS, Azure and Google Cloud on Databricks-managed compute, with on-prem, sovereign cloud and neocloud not available. Onehouse: bring your own cloud or any Kubernetes in your own account, on AWS, Azure and Google Cloud in your VPC, on-prem on your Kubernetes, in sovereign clouds with your region and your keys, and on neoclouds next to your GPUs. The same production ETL job runs in 1 hour 36 minutes at $275 per run on Databricks Photon and in 1 hour 17 minutes at $78.70 per run on Onehouse Quanton: 25% faster at a third of the cost.

How Onehouse stacks up

Shot for shot. Faster ingest, faster Spark, faster queries, on open tables any engine can read, in your own account with no DBUs.

Onehouse compared with Databricks, capability by capability
Capability	Databricks	Onehouse	How it is different
Streaming / ingest	Lakeflow	OneFlow	Lowest-latency CDC and streaming ingest at lakehouse scale. Writes to any catalog. No lock-in.
ETL & transformations	Photon, Lakeflow Spark Declarative Pipelines	Quanton SQL / Spark jobs	25% faster than Photon at a third of the cost. 4x faster Apache Iceberg™ merges with index-aware joins.
Storage & table management	Unity Catalog managed tables, Predictive Optimization	Universal catalog, Table Optimizer	Compaction and clustering with 2–4x less compute. Any engine, any catalog reads and writes the tables.
Analytics / low-latency querying	SQL Warehouses, Databricks SQL Serverless	Lakegres, Open Engines	Sub-second Postgres endpoint on your lakehouse tables. 6x faster than SQL Serverless.
Agents & AI access	Unity AI Gateway, on Databricks compute only	AI Gateway + Lakegres	Model routing and MCP tools in your VPC, on-prem or neocloud. Traces land in open tables.
Deployment & cost	Serverless preferred; hyperscaler clouds only	BYOC or any Kubernetes, incl. on-prem and neocloud	Compute and data stay in your account. Your cloud discounts apply in full.
Spark on your own Kubernetes	Photon runs only on Databricks-managed compute	Quanton Kubernetes Operator (BYOK)	Drop-in for the Kubeflow Spark Operator, 4–6x faster than open-source Spark. Any cluster, any cloud, on-prem.

Open. Faster. Flexible. Cheaper.

Open

Apache Iceberg™, Delta Lake and Apache Hudi™ in your buckets, registered with Glue, Unity, Snowflake and Polaris. Databricks treats Iceberg it does not govern as read-only. Onehouse writes to all of them.

Faster

25% faster than Photon on production ETL, 6x faster than open-source Spark on TPC-DS, 4x faster Iceberg merges. Measured on TPC-DS and on a customer's own pipelines, not a demo.

Flexible

Any Kubernetes: EKS, GKE, AKS, OpenShift, on-prem, air-gapped, neocloud. Same spark-submit, no code changes, and leaving is one config line.

Cheaper

Pay for data processed, not DBUs. On TPC-DS the same work costs 5.3x more on Databricks All-Purpose compute and 2.7x more on SQL Serverless.

Streaming and CDC ingestion that stays open

Lakeflow lands data in Databricks' storage, catalog and engine. OneFlow lands Iceberg or Hudi tables in your buckets with record-level indexing and merge-on-read, and registers it with Glue, Unity, Snowflake and Polaris in one pass. In production that is 350+ TB a day of CDC on a single customer, with every engine reading the same tables.

OneFlow source configuration screen in the Onehouse console

TPC-DS price/performance, Onehouse at $1.00: Databricks All-Purpose $5.27, Snowflake Gen2 $4.87, EMR Serverless $3.55, Databricks SQL Serverless $2.74

Faster than Photon, without the Photon tax

Quanton runs real Apache Spark with no per-cluster premium and no code changes. The chart shows what that buys per dollar on TPC-DS.

4xfaster Iceberg merges, with index-aware joins and low-shuffle MERGE INTO

25%faster than Photon on a public fintech's production ETL, as well as TPC-DS 10 TB

2.9xthe DBUs Photon emits over standard Spark on the same instance. Quanton has no special tax.

Table management that does not require Unity Catalog

Databricks' own documentation makes Unity Catalog mandatory for Iceberg: any table governed by Glue, Polaris or Nessie is a "foreign Iceberg table", read-only, with no partition evolution and no branching. Table Optimizer runs incremental clustering, intelligent compaction and columnar compaction asynchronously with 2–4x less compute, on tables every engine and catalog can read and write. At a $700B+ pension fund it sped up Athena 1.7x, Spark 9x and Trino 12x on the same tables, without changing a query.

Onehouse console, clustering settings for a table: 3.6 TB optimized, 1 minute 34 seconds average clustering duration, 15 minutes between runs, with keys, layout strategy and frequency configurable

Onehouse console, compaction settings for a table: 15 minutes to compaction, 56 seconds average compaction duration, 68 file groups compacted, with frequency and bytes per compaction configurable

Onehouse console, cleaning settings for a table: 357 GB of space reclaimed, 18 seconds average cleaning duration, 15 minutes between runs, with frequency and time-travel retention configurable

Query time in seconds, Databricks SQL Serverless versus Onehouse Lakegres: point lookups on fact tables 21.1 versus 3.5; joins against a large dimension table 381 versus 63; joins against a small dimension table 288 versus 47

Database-speed queries directly on open Iceberg tables

Lakegres serves your Iceberg and Hudi tables through a Postgres wire-compatible endpoint, with record-level indexes and columnar caching. Against Databricks SQL Serverless: point lookups 3.5s versus 21.1s, joins 63s versus 381s. Same tables, no copy into a warehouse.

The only lakehouse that will keep up with your AI agents

Agents think by querying: bursts of point lookups and scans at hundreds of QPS. Lakegres is built for that traffic, with sub-second SQL and vector search on your open tables. Onehouse AI Gateway routes any model from any provider with failover and cost controls, and its built-in MCP server optimizes context queries, so answers come back more accurate, faster and cheaper. Traces land as open tables. Unity AI Gateway runs only on Databricks compute and charges per GB to log them.

Stacked bar chart of the cost of the same TPC-DS work, normalized: Databricks SQL Serverless $2.74 in platform fees; Onehouse BYOC $1.00 at list price, $0.30 Onehouse fees plus $0.70 your infrastructure; $0.72 with a 40% reserved-instance discount, 3.8x cheaper than Serverless; $0.51 with 70% of compute on spot instances, 5.4x cheaper.

Same TPC-DS work, cost normalized to Onehouse at list price. Reserved and spot pricing cut the infrastructure share by 40 to 70%.

Your cloud, your discounts, or no cloud at all

Databricks runs on its own compute in the hyperscaler clouds, with serverless as the preferred path. Onehouse runs on a Kubernetes compute runtime in your account: virtual clusters with queues and resource isolation, your negotiated rates. Your reserved-instance and spot discounts apply to the infrastructure share, 70% of the bill at list price, so the same work lands 3.8 to 5.4x under Databricks SQL Serverless. The same runtime deploys on-prem, air-gapped, or on a neocloud next to your GPUs.

Keep your Kubernetes. Swap the operator.

Photon runs only on Databricks compute. The Quanton Kubernetes Operator matches the Kubeflow Spark Operator spec and deploys driver and executor pods with Quanton instead of open-source Spark: 4x faster merges, 6x faster on TPC-DS, same spark-submit, your existing EKS, GKE, AKS or on-prem cluster. Reverting is one config line.

Your existing Kubernetes cluster on EKS, GKE, AKS, OpenShift or on-prem: the same spark-submit and SparkApplication spec go to the Quanton Operator, a drop-in for the Kubeflow Spark Operator, which launches Spark driver and executor pods running Quanton against your object storage on S3, GCS, ADLS or on-prem. 4 to 6x faster than open-source Spark, revert in one config line.

Go deeper on the open alternative

Blog

Databricks Iceberg support has a catch. It's called Unity Catalog.

Read article

Blog

Introducing Onehouse AI Gateway: routing without chains and tolls

Read article

Blog

Introducing the Quanton Kubernetes Operator for Apache Spark™

Read article

Keep your data. Lose the Photon tax.

get started today