Databricks $190B Valuation vs Snowflake $4.7B Revenue: Data War 2026

The Valuation Gap That Defines Enterprise Data in 2026

The most consequential number in enterprise data software this year is not a revenue figure but a valuation. On August 13, 2026, Databricks closed a $5 billion strategic round — led by Coatue with participation from Blackstone, T. Rowe Price and Point72 — that priced the company at $190 billion, a 42% jump from the $134 billion mark it set in December 20251725. Its most direct competitor, Snowflake, trades publicly on the NYSE at a market capitalization of roughly $114 billion to $116 billion as of mid-August1725.

The headline framing of "$190B vs $4.7B" is really a comparison across different unit types: private valuation against audited public revenue. Snowflake's fiscal 2026, which closed January 31, 2026, delivered $4.68 billion in total revenue, of which $4.47 billion was product revenue, growing about 30% year over year1918. Databricks, still private with no S-1 filed, disclosed an annualized revenue run rate above $7 billion at the time of its August raise, growing more than 80% year over year1925.

That number deserves scrutiny, because the reporting diverges meaningfully across sources. In February 2026, Databricks itself announced crossing $5.4 billion in ARR at 65% growth2120. By August, the figure had climbed past $7 billion and the growth rate had accelerated past 80%2517. Some outlets treat the February and August figures as a coherent acceleration story; others continue citing the $5.4 billion number months later. The defensible read: Databricks' self-reported run rate roughly doubled in a year, while Snowflake's audited growth rate sits in the high twenties to low thirties depending on the quarter2123. The revenue race has effectively flipped — Databricks, long the smaller challenger, now claims the larger and faster-growing base.

Why Private Markets Pay a Premium for Growth

The multiples are surprisingly close. Databricks' $190 billion on roughly $7 billion of ARR works out to about 27x revenue; Snowflake's $114 billion on $4.47 billion of product revenue is around 25x17. The premium, then, is not about multiple expansion — investors are pricing Databricks' 80% growth roughly 2.5 times above Snowflake's 30%17. Snowflake's net revenue retention of 125% to 126% remains excellent by public-market standards, and its remaining performance obligations of $9.2 billion to $9.8 billion are growing 38% to 42%2118. Databricks reportedly sustains 140%+ net retention and claims positive free cash flow despite estimated operating losses2122.

One analyst framing is worth keeping in mind: a $190 billion private valuation is not a market price. It is the price one set of crossover investors agreed to pay for a slice of a company whose terms Databricks controlled, with no daily price discovery forcing honesty17. Snowflake's figures, by contrast, are SEC-audited. The comparison is real but asymmetric — a point that cuts in both directions.

Database Releases: A Blistering Ship Cadence on Both Sides

The financial race is downstream of a product race, and 2026 has been an unusually heavy release year for both platforms.

On Snowflake's side, the release notes tell a story of AI features maturing into governance surfaces. Cortex Search — the company's hybrid vector-plus-keyword retrieval engine — hit general availability with multi-index support and custom vector embeddings in March 2026, letting customers bring their own pre-computed embeddings from third-party models into Snowflake-managed indexes12. Cortex Agents went generally available on September 1619. Cortex Code, an agentic assistant for SQL and Python development inside Snowsight, moved from preview on February 2 to general availability on March 957. The company also shipped data lineage for Cortex Agents on September 2, letting administrators trace which tables feed which agents — a governance feature that would have been unthinkable as a priority three years ago6. On the open-format side, Snowflake added Apache Iceberg version 3 support in preview in March and write support via external query engines4.

Databricks matched the cadence and arguably raised it. The June 2026 Data + AI Summit was the company's most ambitious product showing in years: Lakehouse//RT, a new serverless SQL warehouse type delivering sub-second queries — as low as 10 milliseconds on small datasets with 12,000 queries per second sustained — powered by a new engine called Reyden, which co-founder Reynold Xin called "probably the largest single innovation we have done since our introduction of lakehouse"32. The same event brought managed Iceberg v3 tables to general availability, unifying Delta and Iceberg data files at the storage layer, plus Lakebase updates including sub-second Postgres branching and cross-cloud disaster recovery32.

Vector Databases: The Category That Stopped Being a Category

The most interesting structural shift of 2026 is what happened to the standalone vector database. Neither platform bought one. Databricks agreed to acquire serverless Postgres company Neon in May 2025 for roughly $1 billion; three weeks later Snowflake acquired Postgres specialist Crunchy Data for a reported $250 million9. Neither buyer went after Pinecone, Weaviate, Qdrant or LanceDB. Both chose Postgres infrastructure9.

The reason is now visible in the release notes. Databricks' Lakebase Search, announced in beta at the 2026 summit, delivers hybrid vector and full-text retrieval natively inside Postgres with 32x compression and support for over a billion vector indexes32. Snowflake's Cortex Search handles hybrid retrieval over Snowflake-managed indexes, which for most warehouse-grounded RAG use cases removes the need for a separately hosted vector database entirely3. One widely cited analysis puts it bluntly: for most teams, the question in 2026 is no longer "which vector database," but "what requirement forces this data out of the primary data platform at all"9.

Meanwhile, upstream Postgres keeps improving as the substrate. pgvector 0.8 shipped iterative index scans that fix the long-standing overfiltering problem with filtered vector queries, parallel HNSW builds that cut reindexing time 30-50%, and halfvec quantization that roughly halves storage10. A TigerData/Timescale benchmark reported Postgres with pgvectorscale achieving 471 queries per second at 99% recall on 50 million vectors versus Qdrant's 41.47 — a widely repeated 11.4x figure that, notably, traces to a single benchmark rather than independent reproductions139. Security patching continues apace, with pgvector 0.8.2 (February) fixing a parallel-HNSW buffer overflow and 0.8.7 (October) fixing an IVFFlat build flaw that could enable arbitrary code execution1214. Neon, now Databricks-owned, made Postgres 18 the default for new projects in June and cut storage pricing from $1.75 to $0.35 per GB-month post-acquisition11.

Streaming and the Lakehouse Endgame

Streaming is where the architectural philosophies diverge most sharply. Databricks spent 2026 closing its real-time gap. AUTO CDC flows for streaming tables went generally available in June, handling out-of-order change data capture records as SCD type 1 or type 2 without MERGE logic28. ZeroBus, Databricks' answer to Apache Kafka "when the lakehouse is your sole sink," reached GA alongside Spark Declarative Pipelines' Real-Time Mode, which folds millisecond-level streaming into the same programming model as batch — eliminating, in Databricks' telling, the need for a separate engine like Flink3228. Lakehouse//RT then extends that to the serving layer28. Snowflake's Snowpipe Streaming delivers seconds-level latency, with Snowflake countering on simplicity: serverless ingestion and the same governance model, just without the millisecond ambitions18.

The Reading That Matters

Committing to a verdict: the "Snowflake vs Databricks" framing of 2023 has quietly become a category-versus-category question. Both companies converged on open formats, both absorbed Postgres, both shipped agents, and both now compete for the same data-layer budget that AI applications demand2624. Databricks' $190 billion is a bet that the data platform that trains models, runs pipelines, serves analytics and handles transactional Postgres workloads in one system wins the era2319. Snowflake's $114 billion is the bet that governed simplicity compounds. The multiples are nearly identical — what public markets are really adjudicating, whenever Databricks finally files, is whether 80% growth at $7 billion scale was worth paying twice the price for.