Snowflake vs Databricks: $190B Valuation Rides on Postgres Push

The numbers behind a lopsided headline

One comparison keeps showing up in coverage of the data-platform race this autumn: Databricks' $190 billion valuation set next to Snowflake's $4.7 billion in revenue. It is a striking pair of figures, but it measures two different things. One is what private investors paid for a stake in Databricks. The other is a year of audited sales at Snowflake. Taken alone, each figure is accurate. Set side by side, they make the gap between the two companies look much wider than it is.

The Databricks figure is solid. On August 13, 2026, the company closed a $5 billion round led by Coatue at a $190 billion valuation and said its annual revenue run rate had passed $7 billion in the second quarter.24 Blackstone, MGX, T. Rowe Price and Sixth Street Growth also joined the round, and Databricks said more than 1,000 customers now spend over $1 million a year with it.24 The valuation has climbed fast: $62 billion in December 2024, $134 billion after the Series L in December 2025, and an early $188 billion figure when Coatue's first $3 billion commitment was reported in July.25

The Snowflake figure needs more care. Snowflake's fiscal 2026 revenue was $4.68 billion in total, of which $4.47 billion was product revenue, and those numbers are audited. Databricks' roughly $7 billion is an annualized run rate the company reports itself.12 One valuation tracker makes the same point: Databricks' revenue comes from company statements to the press, not from a prospectus.28 Snowflake has also grown since that fiscal year ended. In September it raised its fiscal 2027 product revenue forecast to $6.07 billion, or 36% growth, up from $5.84 billion.210

Both companies are growing faster, not just Databricks

The usual story is that Databricks is pulling away while Snowflake stalls. The 2026 numbers don't fully support that. Databricks' growth did speed up. It reported a $5.4 billion run rate growing more than 65% in February, then about $6.9 billion growing roughly 80% by June, according to CNBC figures cited in later reports.2526 Snowflake sped up too. Its second-quarter product revenue rose 37% to $1.49 billion, the third quarter in a row of faster growth, and it expects 37% to 38% growth next quarter.29 Its growth rate was 30% at the end of fiscal 2026, when early guidance pointed to only 27% for the year.5

Reports differ on valuation multiples, depending on which Snowflake market value and which revenue figure they use. One June analysis put Snowflake at about 7 times forward sales and Databricks at about 32 times, assuming a $175 billion price.23 A September analysis put Databricks at about 27 times its run rate. It put Snowflake at about 30 times, using a market value near $116 billion.28 The jump in Snowflake's stock after earnings explains much of the gap: shares rose more than 20% after hours on September 2.1 Our reading is that public investors now price Snowflake's slower but accelerating growth at roughly the same multiple private investors give Databricks. That makes the "$190B versus $4.7B" framing more misleading still.

Postgres has become the main front

The products matter more than the valuations, and here the two companies are clearly converging. In 2025 both bought their way into transactional Postgres databases. Databricks paid about $1 billion for Neon in May. Three weeks later, Snowflake bought Crunchy Data for a reported $250 million.1813 Databricks Lakebase became generally available on AWS on February 3, 2026, and Snowflake Postgres followed on February 24.13

The two products are built differently. Lakebase uses Neon's design, which separates compute from storage: the database streams its write-ahead log to a storage layer over the network, so compute can shut down when idle and branch quickly.15 Snowflake Postgres runs on dedicated instances with storage tuned for transactions. Snowflake says it is standard open-source Postgres, not a fork.35 The open-source pg_lake extension connects it to the lakehouse. Released under the Apache license in November 2025, it lets Postgres read and write Iceberg tables and copy table changes into Iceberg with about 30 seconds of delay.35

Databricks has put more of its stage time behind Postgres. At its June summit it said Lakebase handles 12 million database launches a day and announced disaster recovery across clouds and regions.41 It also previewed LTAP, a design that converts Postgres rows into Delta and Iceberg column formats as data is written, so analytics engines can query it without a separate pipeline.41 LTAP was announced as "coming soon" and is not yet shipping.50 Even so, Lakebase is already producing revenue: Reuters, as cited elsewhere, said it passed a $100 million run rate by August.14 Snowflake's answer at its own summit was mostly about hardening for large customers: Private Link, customer-managed keys and secrets. It also added a public preview of Postgres Data Mirroring, which copies tables into Snowflake without a separate change-data-capture setup.32

One independent Postgres commentator gives blunt advice: pick the managed Postgres that sits next to the data platform you already use, because that is effectively the only real choice.11 We agree. For most buyers this race strengthens lock-in rather than opening up options. In return, Postgres keeps gaining ground across the industry. It posted the largest DB-Engines popularity gain of any database in the first half of 2026.20 Neon, now owned by Databricks, cut storage prices from $1.75 to $0.35 per GB-month.20

Vector search moves inside Postgres

The clearest change for vector databases is that neither company bought one. Both are building vector search into what they already sell. At its summit, Databricks put Lakebase Search into beta. It combines vector and full-text search inside Postgres and uses 32x compression to support more than a billion vector indexes.4145 Coverage of the launch said it plainly: teams no longer need a separate vector database such as Pinecone or Weaviate for AI retrieval.48 Snowflake's summit coverage focused more on governance and meaning than on its own vector engine. Its headline items were Cortex Sense, a system that interprets business data for AI agents, and Horizon Context, which collects business context from systems including PostgreSQL.3134

The implication is our own analysis. If billion-scale vector search comes free inside a managed Postgres that sits next to a company's governed data, standalone vector stores will need a very strong performance case to win spending. Databricks currently has the more concrete product here, though it is still in beta.

Streaming: both target Kafka

Both companies also went after streaming infrastructure this summer with similar products. Snowflake announced Datastream, a fully managed streaming service that is compatible with Apache Kafka. Data sent through it picks up Snowflake's governance and lineage tracking automatically.3138 Snowflake estimates the real-time data market at $128 billion.39 Reports disagree on how ready Datastream is. Some list it as private preview.32 At least one tracker describes it as still in development and only heading toward private preview.40

Databricks' Zerobus is further along. It is generally available as a push-based service for loading event data, handling more than 10 GB per second per table, with Kafka-compatible interfaces in beta.41 Databricks says Zerobus does not replace Kafka for spreading events across many microservices. It replaces Kafka only where the lakehouse is the sole destination.46 Snowflake also keeps Snowpipe Streaming as its direct row-by-row loading interface.35 On complex streaming transformations, comparisons still favor Databricks' Spark-based tools.13

Streaming companies should take note. The two biggest independent data platforms now both offer a built-in alternative to running Kafka just to move data into the warehouse. That is the simplest use of Kafka, and also the most common.

Iceberg narrows the lock-in argument

At the storage layer, Snowflake made Apache Iceberg v3 generally available. Through Horizon Catalog, which is built on Apache Polaris, outside engines can now read and write Snowflake-managed tables.3134 Databricks made managed Iceberg v3 tables generally available at its own summit, alongside Delta.41 Reports again differ on details. One said outside-engine writes were still in public preview during Snowflake's summit, while others describe the feature as generally available.3936 Either way, Databricks' old advantage as the open-format option is smaller than it was.12

Our reading

The valuation gap mostly reflects how fast Databricks is growing and how much investors expect from AI. Databricks' AI products alone had a $1.4 billion run rate by February.22 On products, the two companies are now close. Both sell managed Postgres, Kafka-style streaming, Iceberg v3 and agent platforms. Databricks has shipped more on Postgres and vector search. Snowflake has more audited revenue behind it and has now accelerated for three quarters straight. The deciding question for 2027 is which side can turn its operational database into everyday infrastructure for AI agents first. On current evidence, Databricks is slightly ahead, and Snowflake's growth suggests it is not falling behind.