Databricks Review

Databricks closed a $5 billion round in August 2026 at a $190 billion valuation, with revenue running past $7 billion and growing more than 80% year over year. A vendor that size does not fail a board viability test.

The real question is whether it advances you or just rehouses work you already do. Unity Catalog gives you governance across clouds, and Lakebase is a new serverless Postgres tier aimed at AI applications. Snowflake is the cleaner pick if you only need a warehouse, but Databricks bundles ETL, machine learning, and BI on one platform built on Apache Spark.

The catch is cost discipline. Usage-based billing with no upfront commitment scales fast both ways, and a board will ask why the bill jumped a quarter. Pilot one data team on pay-as-you-go, watch the spend, then negotiate a Committed Use contract.

Databricks was founded in 2013 by the team behind Apache Spark, and the platform reflects engineers who think in layers. Delta Lake is the open storage format, Unity Catalog is the governance plane, and the compute sits above both. For a CTO choosing a data substrate through 2029, that separation matters — your tables stay in open Parquet, so the lock-in lives in the management layer, not the data itself.

The craft ceiling here is genuinely high. Lakebase, a serverless Postgres integrated with the lakehouse, and Genie, the natural-language query layer, show a company building toward agent-era applications rather than relabeling a warehouse. Against Snowflake, the strategic edge is owning data engineering and ML on the same governed plane instead of stitching two vendors together.

But the catch is cost governance. Per-second usage billing with no flat subscription tier scales beautifully and unpredictably, so adopting Databricks means standing up FinOps discipline on day one. Committed Use Contracts soften the rate, not the need for guardrails.

Databricks does not sell seats. It sells compute, metered in DBUs at per-second granularity. There is no flat sticker price — only published per-DBU list rates, a calculator and two paths: Pay as you go, or Committed Use Contracts.

The rate spread matters. SQL Serverless runs about $0.70/DBU; SQL Classic about $0.22. A query that finishes ten times faster on the pricier tier can still cost less. The catch is forecasting — without governance, DBU spend drifts, and the invoice you cannot predict is the real exposure, not the sticker. Compare Snowflake, which meters credits the same way. Committed Use Contracts trade flexibility for a discount, so model your floor before signing.

ROI is measurable per workload. Unity Catalog ties cost back to teams and jobs. Vendor risk is negligible: an August 2026 round at a $190B valuation, a $7B+ revenue run-rate, >80% growth.

A data engineer judges a platform by the PySpark job that failed overnight, not the keynote. Databricks handles that case. Collaborative Notebooks run Python, Scala, and SQL side by side, and Delta Lake gives the data lake real ACID transactions, so a half-finished write does not poison the next morning's ETL run.

The workflow fit is genuine. Unity Catalog keeps lineage and access control in one surface. Automated cluster management means you size a job once instead of babysitting infrastructure. The Notebooks load fast and the docs read like engineers who actually run pipelines wrote them.

The catch is the meter. Pricing is usage-based at per-second granularity across product SKUs, so a cluster left idle becomes a surprise line item, and cost forecasting takes real discipline. Lakebase, the serverless Postgres that hit GA in 2026 on acquired Neon technology, is promising but still young.

Databricks is the kind of tool you do not casually try. Pay-as-you-go billing is per-second, which sounds friendly, but the only free plan is the non-commercial Free Edition, so evaluating it for work means a 14-day trial with up to $400 in credits before pay-as-you-go billing starts. That is the trade for what you get: one workspace that runs ETL, notebooks, and SQL without bouncing between three products.

Month three is where it earns its keep. Unity Catalog handles governance across clouds so permissions do not slowly rot, and the collaborative notebooks take Python, SQL, R, and Scala in the same file. Genie lets analysts ask questions in plain language instead of waiting on the data team. Amazon SageMaker covers the ML side but does not feel this stitched together.

The catch is weight. Cluster management is automated, but the platform assumes a real data team behind it, and per-second billing with no minimums still adds up fast at scale. Founded in 2013 by the Apache Spark creators and now valued at $190 billion, this is not a tool you outgrow.

The data-platform graveyard has names on its headstones. Databricks isn't near it. Founded 2013 by the Apache Spark crew, it closed a round at a $190 billion valuation in August 2026. Survivor numbers, not pitch-deck numbers.

The lakehouse story holds up. Unity Catalog handles governance across clouds, and Delta Lake supplies the open table format underneath. Newer bets like Lakebase, the serverless Postgres layer, lean hard into AI-era framing. The catch is the marketing — "the first serverless Postgres database integrated with the lakehouse" is the kind of superlative I'd discount. Against Snowflake the differentiation is real; against Amazon SageMaker it's a broader platform.

The yellow flag is pricing. Usage-based, per-second billing, a free tier limited to non-commercial use, no committed-use numbers. Costs are knowable only after you're already running workloads.