Executive Intelligence · Company Analysis · November 14, 2026
Snowflake: Can the Data Cloud Become the Enterprise AI Data Layer? As enterprises accelerate AI adoption, trusted data is becoming essential to building intelligent business applications. Snowflake is evolving its Data Cloud into an integrated platform for enterprise data, analytics, and AI. The question is whether it can become a foundational data layer for enterprise AI.
The Data Is There. The Question Is Whether the AI Workloads Will Follow.
Snowflake reported Q2 FY2027 product revenue of $1.49 billion, up 37% year over year, its third consecutive quarter of accelerating growth, and raised its full-year product revenue guidance to approximately $6.07 billion. The company ended the quarter with 14,554 customers, 829 Forbes Global 2000 accounts, and a 126% net revenue retention rate. Its AI products — CoCo (formerly Cortex Code) and CoWork (formerly Snowflake Intelligence) — reached 9,100 and 5,800 accounts respectively, and management attributed roughly half of the company's revenue-growth acceleration to AI.
Those numbers describe a company executing well. But the central strategic question is not whether Snowflake can grow. It is whether the company can convert its position as the repository and governor of enterprise data into a position as the execution layer for enterprise AI — the place where agents retrieve context, make decisions and take action, not just the place where data sits.
The evidence points to a nuanced answer. Snowflake's governed data foundation, cross-cloud architecture and model-neutral positioning are genuine advantages for enterprises that want to run AI on their own terms. Its AI portfolio has moved quickly from feature checkboxes to an integrated stack spanning inference, development, agents, and governance. But the competitive landscape is closing fast. Databricks has staked out a lead in systems of execution with its Lakebase transactional database and LTAP architecture. And Snowflake's own operational database strategy remains "a tad incoherent," in the words of one Everest Group assessment, with the older Unistore/Hybrid Tables approach coexisting alongside the newer Snowflake Postgres direction.
This analysis examines Snowflake's evolution, platform architecture, AI portfolio, partnerships, competitive position and business model. The central finding: Snowflake has the right data foundation for enterprise AI, but it has not yet proven that it can own the execution layer where agents operate.
1. From Data Warehouse to Agentic Control Plane
Snowflake was founded in 2012 by three engineers with backgrounds at Oracle, and listed in 2020 as one of the largest software IPOs of its era. Its original proposition was straightforward: a cloud-native data warehouse that separated storage from compute, allowing customers to scale analytical workloads independently and across multiple public clouds.
That architecture remains the foundation of everything the company has built since. Storage, compute and cloud services are decoupled, and the platform operates on AWS, Microsoft Azure and Google Cloud, making cross-cloud data access a core part of its positioning. For enterprises that did not want to be locked into a single hyperscaler's data stack, Snowflake offered a neutral layer.
The expansion since then has been deliberate. The company moved from analytics into data engineering, data sharing, application development and collaboration. With the arrival of CEO Sridhar Ramaswamy in 2024 — a former Google advertising executive — the emphasis shifted decisively toward AI. The platform was rebranded as the AI Data Cloud, and a portfolio of AI products was assembled: Cortex AI for model access and inference, Snowflake Intelligence (now CoWork) for business-user agents, Cortex Code (now CoCo) for developer agents, and the Cortex AI Gateway for governance and cost control.
At Snowflake Summit in June 2026, the company introduced the concept of an "agentic control plane" — a layer that orchestrates AI agents across enterprise data, models and applications. Ramaswamy described it as "a new browser, a space where users do not stop at querying data but orchestrate work across multiple applications". That is a significant ambition. It is also a direct challenge to the hyperscalers and to Databricks, both of which are building their own versions of the same layer.
2. Financial Performance — Growth Is Accelerating, but AI Revenue Is Blended
Snowflake's Q2 FY2027 results, reported September 2, 2026, describe a business that is growing faster than it has in two years.
| Metric | Q2 FY2027 | Context |
|---|---|---|
| Product revenue | $1.49B | +37% YoY; 96% of total revenue |
| Total revenue | $1.55B | +35% YoY |
| Net revenue retention | 126% | Supported by migrations and AI use cases |
| Remaining performance obligations | $9.0B | +30% YoY; ~54% expected in next 12 months |
| Customers >$1M TTM product revenue | 828 | +27% YoY; 48 net new |
| Forbes Global 2000 customers | 829 | 14 net new in quarter |
| Non-GAAP operating margin | 15.3% | Up from 11.1% a year ago |
| FY2027 product revenue guidance | ~$6.07B | 36% growth; raised |
The AI revenue question is harder to answer than the headline numbers suggest. Management said AI products contributed roughly half of the acceleration in growth, and that customers adopting AI products also increase consumption of the core platform. But Snowflake does not break out AI revenue as a separate line. The company reports account counts for CoCo and CoWork, not paid usage. Those are adoption metrics, not monetisation metrics. The distinction matters: an account can be active without generating material incremental consumption, and a customer can experiment with AI features without shifting production workloads onto them.
The margin story is also more complicated than it appears. Non-GAAP operating margin expanded, but gross margin contracted 120 basis points to 71.8%. CFO Brian Robins attributed the compression to the higher mix of AI workloads, which "currently carry lower contribution margins". That is a structural issue, not a temporary one: if AI becomes a larger share of Snowflake's business, the company will have to find ways to improve AI workload economics or accept a lower-margin mix.
3. The AI Portfolio — A Stack, Not a Feature
Snowflake's AI portfolio has matured from a set of loosely connected features into a coherent stack. The components serve different users and different stages of the AI workflow, but they share a common foundation: governed enterprise data.
Cortex AI is the managed inference layer. It provides access to models from multiple providers — Anthropic, OpenAI, Google and Snowflake's own Arctic family — through SQL functions and APIs. The strategic choice here is model neutrality: Snowflake does not require customers to standardise on a single foundation model, and the Cortex AI Gateway can route tasks to different models based on customer-defined policies and performance data. That is attractive for enterprises that want to avoid lock-in and optimise costs.
CoWork (formerly Snowflake Intelligence) is the business-user agent. It allows non-technical users to ask questions of governed enterprise data in natural language and receive insights without writing SQL. At Summit, Snowflake added features including User Skills for personalisation, Deep Research for multi-step analysis across structured and unstructured data, and Cortex Sense for joining data with business definitions. CoWork reached 5,800 accounts in Q2, up nearly 11% sequentially.
CoCo (formerly Cortex Code) is the developer agent. It provides AI-assisted development for building and managing data and AI workflows. New capabilities include Datastream for bringing real-time Kafka data into AI applications, desktop and mobile versions, Automations for recurring workflows, and prebuilt Skills. CoCo surpassed 9,100 accounts, adding more than 2,000 net new accounts in the quarter.
Snowflake Postgres is the newest strategic expansion. It moves Snowflake beyond analytical workloads into transactional application data — the operational database layer where agent state, application state and low-latency workloads live. That is a direct challenge to Databricks' Lakebase and to the operational databases that enterprises have historically kept separate from their analytical platforms. Everest Group's assessment is pointed: the coexistence of the older Unistore/Hybrid Tables approach with the newer Snowflake Postgres direction "creates two potential pathways to the AI database," making the destination "less immediately legible".
The Cortex AI Gateway is the most strategically important product in the portfolio, and the least discussed. It is a centralised layer for governing agent activity, model access and AI consumption costs. In an enterprise AI deployment, the gateway is where policy is enforced, where costs are attributed, and where the boundary between what an agent can and cannot do is defined. If Snowflake can make the gateway the control point for enterprise agents — including agents that interact with systems outside Snowflake — it becomes the governance layer for the agentic enterprise. That is a much larger opportunity than selling inference. But the gateway currently governs agents and connections routed through it; it does not yet provide estate-wide discovery and management of unmanaged agents.
4. Model Partnerships and the AWS Relationship
Snowflake's model strategy is deliberately plural. The company has partnerships with Anthropic, OpenAI, Google Cloud and NVIDIA, giving customers access to Claude, GPT and Gemini models alongside Snowflake's own Arctic models. That neutrality is a differentiator against hyperscalers that prefer their own models and against Databricks, which has its own model portfolio through MosaicML.
The AWS relationship is the most consequential and the most revealing. In May 2026, Snowflake committed $6 billion in AWS Graviton compute and AI spending over five years, alongside joint investments in customer success, migrations and go-to-market. The deal makes Snowflake one of AWS's largest CPU-based computing customers. It also illustrates the structural tension in Snowflake's business: the company can operate across clouds, and it can offer model neutrality, but it runs on infrastructure it does not own and must manage substantial underlying computing costs. The $6 billion commitment is a bet that AWS's Graviton chips provide the best price-performance for Snowflake's workloads — but it also deepens the dependency.
Snowflake shares jumped about 30% in after-hours trading when the deal was announced. The market read it as validation of Snowflake's AI positioning and its relationship with the largest cloud provider. The longer-term question is whether the economics of the deal favour Snowflake or AWS. A five-year, $6 billion commitment gives AWS significant leverage, and it locks Snowflake into a specific hardware path at a moment when the AI infrastructure market is still evolving.
5. Competitive Landscape — Databricks Is the Benchmark
The enterprise data platform market has consolidated around two leaders: Snowflake and Databricks. Both have expanded from their origins — Snowflake from data warehousing, Databricks from data engineering and machine learning — into overlapping platforms that cover analytics, AI development, governance and operational data.
| Competitor | Strategic challenge to Snowflake |
|---|---|
| Databricks | Unified data engineering, analytics and AI development; leads in systems of execution with Lakebase and LTAP |
| Microsoft | Azure data services, Fabric and enterprise distribution; embedded in productivity workflows |
| Google Cloud | BigQuery, Vertex AI and integrated cloud AI |
| AWS | Redshift, Bedrock and broad cloud ecosystem |
| Oracle | Database relationships and enterprise AI infrastructure |
| MongoDB | Operational application data and vector search |
The Databricks comparison is the one that matters most. Databricks reported annualised revenue of $7 billion in June 2026, growing more than 80% year over year, and raised at a valuation of approximately $190 billion. That is faster growth than Snowflake's 37% product revenue growth, and it reflects Databricks' stronger position in AI development and data engineering workloads.
Everest Group's assessment is that Databricks has established "a bold architectural lead" in systems of execution. Lakebase provides a managed PostgreSQL foundation for transactional applications and agent state; Lakehouse RT targets low-latency analytical workloads; and the proposed LTAP architecture seeks to bring transactional and analytical processing together on a common lakehouse foundation. Snowflake's operational database strategy, by contrast, is described as less coherent: Unistore and Hybrid Tables coexist with Snowflake Postgres, creating uncertainty about the destination.
Against the hyperscalers, Snowflake's argument is neutrality and portability. A customer can run Snowflake on AWS, Azure and Google Cloud, and can access models from multiple providers. That is valuable for enterprises that do not want to be locked into a single cloud's data and AI stack. But the hyperscalers are improving their native offerings, and the gap between a cross-cloud platform and a well-integrated native stack is narrowing. The question is whether neutrality is worth a premium when the native option is good enough.
6. The Enterprise AI Data Layer Thesis — Foundation Without Execution
The case for Snowflake as the enterprise AI data layer rests on a simple observation: AI models are only as useful as the data they can access, and enterprises trust Snowflake with their data. The company already holds the governed, permissioned, business-context-rich information that agents need to make decisions. It has the access controls, the metadata and the audit trails that regulated enterprises require. And it has the cross-cloud architecture that lets customers run AI wherever their data lives.
That is a genuine foundation. But being a repository for enterprise data does not automatically make Snowflake the execution layer for enterprise AI. The execution layer is where agents run, where actions are taken, where state is maintained and where outcomes are measured. It is the layer that Databricks is building with Lakebase and LTAP. It is the layer that hyperscalers control through their cloud platforms. And it is the layer that Snowflake is trying to reach with Snowflake Postgres, CoCo and the Cortex AI Gateway — but has not yet proven it can own.
The evidence of progress is real but incomplete. CoCo and CoWork adoption is growing. The Cortex AI Gateway is a credible governance layer. The AWS commitment gives Snowflake the infrastructure base to scale. But the company has not disclosed AI-specific revenue, and its operational database strategy is still being assembled. The risk is that Snowflake becomes the data layer for AI that runs somewhere else — the place where context is retrieved but not the place where work is executed. That would still be a valuable business, but it would not be the control plane Ramaswamy describes.
7. Business Model — Consumption Economics Under Pressure
Snowflake's revenue model is consumption-based: customers buy credits and consume them through compute, storage and AI workloads. That model has advantages — it aligns revenue with usage and allows customers to scale — but it also creates volatility. Customers optimise their consumption, and revenue can fluctuate with workload efficiency.
The AI question is whether AI creates new consumption or simply shifts existing workloads. Management says AI adopters consume more of the core platform, and that AI products contributed roughly half of the growth acceleration. But the gross margin compression tells a more complicated story: AI workloads carry lower contribution margins, and if they become a larger share of the mix, Snowflake's profitability will be affected. The company will need to improve AI workload economics — through better model routing, caching, or more efficient inference — to maintain its margin profile.
The marketplace and data-sharing economics are an underappreciated part of the model. Snowflake's data marketplace reached 4,105 listings, up 21% year over year. Data sharing is a source of stickiness: once an enterprise is sharing data with partners and customers through Snowflake, the cost of moving to another platform increases. That is a moat that is not captured in the AI narrative but matters for retention.
Snowflake is executing well. The question is whether execution in the data layer is enough to win the AI layer.
The company's Q2 results are genuinely strong. Product revenue growth accelerated to 37%, the third consecutive quarter of acceleration. Net revenue retention held at 126%. Operating margin expanded. The customer base grew. These are the metrics of a business that is working.
But the strategic question is not about growth. It is about positioning. Snowflake sits on a mountain of governed enterprise data, and that data is the raw material for enterprise AI. The company has built a credible AI portfolio on top of it. What it has not yet demonstrated is that customers will run their most important AI workloads — the agents that take actions, make decisions and execute business processes — on Snowflake rather than retrieving context from Snowflake and executing elsewhere.
Databricks is the benchmark, and it has a head start in the execution layer. Lakebase and LTAP give Databricks a coherent architectural story for agents that need to write as well as read. Snowflake's answer — Snowflake Postgres, CoCo, the Cortex AI Gateway — is credible but less legible. The coexistence of Unistore and Snowflake Postgres suggests a company that has not yet decided how to approach the operational database market.
The AWS deal is a double-edged sword. It gives Snowflake access to cost-efficient Graviton compute at scale and deepens its relationship with the largest cloud provider. It also commits the company to $6 billion in spending over five years and reinforces its dependence on infrastructure it does not own. That is not a weakness unique to Snowflake — every SaaS company runs on someone else's cloud — but it is a constraint on the economics of the AI business.
The most important product in Snowflake's portfolio is the one that gets the least attention: the Cortex AI Gateway. If the gateway becomes the control point for enterprise agents — the place where permissions, costs and actions are governed — Snowflake becomes essential to the agentic enterprise regardless of where the agents run. That is a more defensible position than trying to own the execution layer directly. It is also a harder product to sell, because governance is invisible until something goes wrong.
The caveat worth stating plainly: this analysis rests on company filings, earnings calls and independent research. Snowflake does not disclose AI-specific revenue, and the AI adoption figures cited are account counts, not paid usage. The competitive comparisons are based on publicly available information and analyst assessments. The strategic read is defensible; the outcomes are not yet established. This article does not provide investment advice.
The CODEW Stat
Half. The share of Snowflake's revenue-growth acceleration that management attributes to AI products. That is a significant number, and it is the strongest evidence that the AI portfolio is translating into business results. But it is also a measure of dependence: if half of the acceleration comes from AI, and AI workloads carry lower margins, then the company's growth and its profitability are pulling in opposite directions. Snowflake's next challenge is not proving that AI can drive growth. It is proving that AI can drive growth profitably.
Sources & References
- Snowflake — Q2 FY2027 earnings release and earnings call (September 2, 2026) — investors.snowflake.com
- Nasdaq — "Snowflake Q2 Earnings Call Highlights" (September 1, 2026)
- Nasdaq — "Snowflake Q2 Earnings Beat on Product Revenues and AI Strength" (September 2, 2026)
- Everest Group — "Picking an enterprise data platform in a market with no clear winner" (September 7, 2026)
- Everest Group — "From Data Foundations to Agentic Execution: Key takeaways from Snowflake Expedition 2026" (November 10, 2026)
- DigitalToday — "Snowflake agentic OS-style strategy gains traction, AI brand overhauled" (June 1, 2026)
- DigitalToday — "AWS adds big Graviton CPU customer as Snowflake signs $6 billion five-year deal" (May 27, 2026)
- TechTarget — "Snowflake barrage adds more AI development, analysis tools" (June 1, 2026)
- Snowflake Investor Relations — "Snowflake Delivers the Foundation for Autonomous Data Engineering" (November 2, 2026)
- GuruFocus — "Decoding Snowflake Inc (SNOW): A Strategic SWOT Insight" (March 20, 2026)
Financial figures are company-reported and unaudited unless otherwise stated. AI account adoption figures are company-reported account metrics, not paid usage or production deployment counts. Snowflake does not disclose AI-specific revenue. Competitive assessments are drawn from independent research and analyst commentary and are attributed accordingly. This article does not provide investment advice.
Related Coverage
Reviewed by Erwin Castro
on
Saturday, November 14, 2026
Rating:
No comments: