Snowflake and Databricks Are Converging on the Same AI Data Platform
Snowflake’s Q1 FY2027 product revenue reached $996 million — its first brush with the $1 billion quarterly mark — growing 26 percent year-over-year under CEO Sridhar Ramaswamy, who has spent the 18 months since taking over repositioning the company’s roadmap around AI workloads. In the same period, Databricks crossed $3.5 billion in annualised revenue, growing above 50 percent year-over-year, and is preparing for an IPO that analysts are pricing in the $80-100 billion range. Snowflake’s Q1 FY2027 investor materials and Databricks’ most recent funding disclosures confirm that the two largest independent data platforms are now competing for the same enterprise budget category — AI-ready data infrastructure — despite having been built for different purposes from opposite ends of the same problem. The convergence is creating a consolidation moment in enterprise data architecture that CIOs and chief data officers are being forced to navigate without a clear answer about which platform wins.
Snowflake was built as a cloud-native SQL data warehouse: governed, performant, accessible to business analysts through standard SQL interfaces, priced on compute and storage consumption. Its architecture separated storage from compute in a way that made it radically more flexible than on-premise data warehouses and drove one of the most successful enterprise software IPOs in history in 2020. Databricks was built as a unified analytics platform on top of Apache Spark, designed by the academic team that created Spark and optimised for data engineering and machine learning workloads that required Python notebooks, distributed compute, and direct access to raw data lakes rather than warehouses. The two platforms attracted different buyers — Snowflake’s SQL-first approach won with analytics and BI teams; Databricks’ code-first approach won with data science and ML engineering teams — and for several years co-existed without direct competition in most enterprise accounts.
Snowflake and Databricks Started at Opposite Ends of the Same Stack
The architectural gap that kept the two platforms non-competing has closed from both sides. Snowflake added Snowpark — a framework enabling Python, Java, and Scala workloads to run directly in Snowflake — and acquired Neeva (an AI search company) to accelerate its AI feature roadmap. Snowflake Cortex, the company’s AI/ML layer, provides large language model inference, text-to-SQL capabilities, and document AI directly within the Snowflake environment, enabling analysts who have never written Python to run LLM-powered queries against their governed data. Databricks, moving in the opposite direction, added Databricks SQL — a high-performance SQL warehouse that competes directly with Snowflake’s core competency — and has aggressively marketed the Lakehouse architecture as a unified replacement for the Snowflake-plus-Databricks two-platform approach that many enterprises currently operate.
The strategic logic of both movements is the same: the AI era has elevated the importance of the data layer, and the platform that wins the AI data layer wins a multi-decade renewal of enterprise software spend. Cloud infrastructure capex is growing at rates that reflect AI workload growth, and both Snowflake and Databricks are positioned to capture the application layer above that infrastructure if they can deliver the governed, accessible, AI-augmented data platform that enterprises actually need. The problem for enterprise buyers is that both platforms are credibly claiming to be that platform, and neither has yet demonstrated that its historically weaker side — Snowflake’s ML credentials, Databricks’ governance and BI credentials — has fully caught up to the other’s core strength.
What Cortex and Mosaic AI Actually Deliver
Snowflake Cortex and Databricks Mosaic AI are the respective AI product layers that each company is betting on to differentiate in the AI era. Cortex provides LLM functions accessible via SQL: COMPLETE (text generation), EXTRACT_ANSWER (question answering over documents), SENTIMENT, SUMMARIZE, and TRANSLATE. These are high-level, low-friction functions that allow a data analyst to run AI against their Snowflake data without writing Python or managing model infrastructure. The value proposition is accessibility — the analyst who has been using SQL for a decade can now apply AI to their data without crossing a technical threshold they have not previously had to cross.
Mosaic AI on Databricks targets a different user: the ML engineer or data scientist who wants to fine-tune foundation models on proprietary data, run large-scale model training on distributed Databricks clusters, and deploy models into production with MLflow tracking. The Databricks approach assumes a higher technical floor and delivers deeper capability at that floor — model customisation, vector search, AI agent tooling, and the Unity Catalog governance layer that bridges ML model management with data governance. The practical division is that Cortex is winning with centralised analytics teams who need AI features without ML expertise, while Mosaic AI is winning with data science organisations that are building bespoke AI products. Enterprise AI cost management is a concern on both platforms: Cortex’s per-call LLM pricing and Mosaic AI’s GPU compute charges add cost layers that data platform budgets did not previously carry.
Microsoft Fabric as the Third Competitor
The Snowflake-Databricks duopoly framing obscures a significant third force: Microsoft Fabric, announced in 2023 and generally available since late 2023, which attempts to unify the data engineering, analytics, and AI layers within Microsoft’s existing enterprise ecosystem. Fabric integrates OneLake storage, Synapse Analytics, Power BI, Azure ML, and Real-Time Intelligence into a single governance and management surface. For enterprises already paying for Microsoft Azure and Microsoft 365, Fabric’s pricing is bundled in ways that make standalone Snowflake or Databricks economics harder to justify to a CFO — not because Fabric has matched either platform’s depth, but because the incremental cost of Fabric for an existing Microsoft customer is often near zero relative to the existing enterprise agreement. Microsoft’s positioning is laid out on the Fabric product page.
Snowflake and Databricks are both aware of the Microsoft bundling risk and have positioned their independence — and their multi-cloud neutrality, running natively on AWS, Azure, and Google Cloud — as the differentiator that Fabric cannot replicate. A company standardised on Fabric is a company standardised on Azure; a company on Snowflake or Databricks can shift cloud providers without losing their data platform investment. The precedent from enterprise workflow platforms is instructive: platform independence has consistently commanded a premium when the alternative is lock-in to a single hyperscaler’s ecosystem, and the enterprise data category — where data gravity is even higher than workflow gravity — may prove more resistant to hyperscaler consolidation than adjacent categories. Whether that premium is sufficient to sustain two independent unicorns plus an IPO candidate in a category that Microsoft is bundling aggressively is the question that will resolve in the next platform purchasing cycle.
Why Enterprises Are Running Both and Whether That Can Last
The most common enterprise data architecture in 2026 is a combination of Snowflake for governed SQL analytics and Databricks for ML and data engineering, with data shared between them via open formats (Delta Lake, Iceberg, Parquet) that both platforms support. This two-platform approach is expensive — licencing both platforms for a large enterprise adds several million dollars annually to data infrastructure costs — and creates operational complexity around data synchronisation, access governance, and skills development. Databricks’ messaging has explicitly targeted this two-platform reality as the argument for consolidating onto a single Lakehouse; Snowflake’s messaging has equally explicitly targeted it as the argument for staying with Snowflake and using Cortex rather than maintaining a separate ML platform. This is exactly the kind of platform-monetisation overlap hedge funds were exiting tech to avoid.
The two-platform situation will not last indefinitely: at some point in the next two to three years, the enterprise organisations that currently run both will face a renewal cycle in which one platform’s AI capabilities have become strong enough to justify consolidation, and the switching-cost analysis will tip toward whichever platform has closed the capability gap more convincingly. Which direction that consolidation goes — Lakehouse unifying data engineering and analytics, or cloud data warehouse expanding into ML — will determine which of the two companies captures the majority of the enterprise AI data infrastructure category that both are competing to own.
When Two Competitors Converge on the Same Architecture, the Category Wins
Shane Parrish at Farnam Street builds on Charlie Munger’s observation that the best mental models force you to look at a situation from a different level of abstraction than the one that feels most natural. The natural way to read the Snowflake-Databricks convergence story is as a competitive battle — two well-funded companies fighting for the same enterprise data contracts. The second-order read is more useful: when two strong competitors converge on an identical architecture, the category they are both converging toward tends to beat the alternatives they are both abandoning.
Through 2022, the enterprise data infrastructure market had two genuine camps: cloud data warehouses optimised for structured SQL analytics, of which Snowflake was the commercial leader, and data lakehouse platforms optimised for ML pipelines and unstructured data engineering, of which Databricks was the commercial leader. Enterprises had a real architectural choice. The SQL shop and the Python shop pointed at different platforms and the platforms were genuinely different.
What the 2026 convergence eliminates is that real choice. Databricks’ SQL Analytics has closed the performance gap with Snowflake’s data warehouse sufficiently that a new enterprise buyer evaluating both platforms faces two products that can do most of what the other one does. Snowflake’s ML and Spark integration has similarly closed the gap with Databricks’ native data engineering environment. The buyer now chooses based on pricing, existing contracts, support relationships, and which sales team showed up better — not based on fundamental architectural fit.
The mental model that applies here is what Parrish calls “avoiding the obvious wrong choice” — the observation that eliminating clearly bad options is more valuable than optimising among equivalent good ones. For enterprise AI data infrastructure buyers in 2026, the obvious wrong choices (proprietary on-premise databases, first-generation Hadoop stacks, single-workload solutions) have been eliminated by the convergence. Snowflake and Databricks have each become defensible enough that either is a reasonable choice. The category — unified cloud AI data platform — has already won. Which company captures the larger share of that category over the next three years is a second-order question, and it will be decided by sales execution and switching costs rather than architectural differentiation.
