When the Warehouse Starts Thinking for Itself
Snowflake’s Cortex AI suite has been expanding quietly but steadily, and the direction it’s moving points directly at one of the most embedded parts of the modern data stack: the transformation layer that dbt Labs has spent years building its business around. Cortex now offers in-warehouse machine learning, natural language querying, document intelligence, and LLM-powered functions that run natively inside Snowflake’s compute environment. The more capable Cortex becomes, the harder it is to justify the extra step of routing logic through dbt Cloud.
dbt Labs built its dominance by solving a real problem – analysts needed a clean, version-controlled way to write SQL transformations without touching infrastructure. It worked, and a generation of analytics engineers made dbt the default. But Snowflake has spent the past eighteen months building a layer that doesn’t just sit beneath those transformations – it increasingly absorbs them.
What Cortex Actually Does Now
Cortex is not a single product. It’s a collection of AI capabilities embedded directly into Snowflake’s platform – Cortex Analyst for natural language to SQL, Cortex Search for semantic search over unstructured data, Cortex Guard for AI safety filters, and a set of LLM functions that let users call models like Mistral and Llama directly inside SQL queries. This isn’t a chatbot bolted onto a dashboard. It’s AI running at the compute layer, where the data already lives.
The strategic weight of that positioning matters. When AI inference happens inside the data warehouse itself, the workflow for an analytics team changes. Instead of extracting data, transforming it with dbt, loading it into a feature store, and then calling an external model, the whole chain can collapse into a single SQL statement. Fewer hand-offs mean fewer points where a third-party tool earns its seat at the table. dbt’s seat is exactly the kind that Cortex is quietly pulling away.
dbt’s Moat and Where It’s Thinning
dbt Labs doesn’t compete on compute – it never tried to. Its moat is the developer experience: modular SQL, lineage graphs, testing frameworks, documentation that auto-generates from model metadata, and a community of analytics engineers who think in dbt concepts. That ecosystem is real and deep, and it won’t dissolve because Snowflake added some AI functions. But the moat gets thinner the further up the stack Snowflake climbs.
The pressure shows up most clearly in the transformation-to-insight journey. Historically, that journey had natural gaps where tools like dbt, Looker, or Sigma inserted themselves. Cortex Analyst is designed to close the gap between raw data and a business answer by letting a user ask a question in plain English and get a SQL query generated, executed, and returned – all within Snowflake. That’s not a perfect replacement for a dbt project, but it does make some of the lighter dbt use cases feel like overhead.
dbt Labs has responded by pushing harder into its semantic layer – dbt Semantic Layer and MetricFlow are meant to make dbt the canonical place where business metrics get defined, regardless of what BI tool or AI interface sits on top. The logic is sound: if every tool agrees that “revenue” means the same thing because dbt says so, then dbt stays central even when Snowflake handles the AI. The question is whether Snowflake’s customers actually adopt that standard or just define their metrics inside Cortex Analyst and move on.
There’s also the cost conversation that Snowflake’s sales teams are well-positioned to have. dbt Cloud charges by seat and by usage, sitting on top of infrastructure costs that customers are already paying to Snowflake. When Snowflake’s account executives can point to native AI capabilities that are already included in compute spend, the CFO’s instinct is predictable. Consolidation narratives are winning in enterprise software right now, and Snowflake has a consolidation story to tell.
The Broader Data Stack Squeeze
dbt Labs isn’t the only company feeling this kind of pressure from cloud platforms expanding their native capabilities. The entire category of best-of-breed data tools – orchestrators, transformation layers, observability platforms – is navigating a moment where the hyperscalers and major cloud warehouses are building aggressively into the middle of the stack. Snowflake’s platform ambitions mirror what Databricks is doing with Unity Catalog and its own AI functions, which means the squeeze on independent tooling is coming from multiple directions.
For dbt Labs, the timing is complicated by the fact that the company raised at a high valuation during the peak of data tooling enthusiasm and now has to justify that number in a market that’s more skeptical of point solutions. A pivot toward being an AI-native semantic layer is the right strategic instinct – it’s a higher-value position than transformation alone – but executing it while Snowflake is actively filling the same space from below requires precise positioning and fast product movement.
What Comes Next for Analytics Engineers
Analytics engineers themselves are watching this carefully. Many have careers built on dbt fluency, and there’s a real professional stake in whether the tool remains central to enterprise data work. The honest read is that dbt’s workflow discipline – version control, testing, documentation – still offers things that Cortex doesn’t replicate, especially for teams managing complex pipelines with hundreds of models and strict governance requirements. Cortex Analyst is fast and accessible; a well-maintained dbt project is auditable and reliable in a different way.
But “auditable and reliable” is a maintenance pitch, and AI is selling possibility. Snowflake’s messaging around Cortex leans into the promise of democratizing data access – giving business users direct answers without waiting for an analytics engineer to build a model. That framing, whether or not it fully delivers, shifts the perceived value of the analytics layer in ways that favor the platform offering the AI and complicate the position of the tool that sits between the platform and the question.
The coming eighteen months will likely force dbt Labs into a harder choice about whether to remain a transformation-and-semantics specialist for power users or to build outward into enough of the data workflow that it can make its own consolidation argument to enterprise buyers. Half-measures in that direction could leave it too complex for the Cortex-curious and too limited for the teams that need full pipeline orchestration. Snowflake, for its part, doesn’t need Cortex to fully replace dbt – it just needs enough customers to ask whether they still need both.
