Snowflake and Databricks Climb the AI Stack: Build vs Buy Now

Any platform lead sitting on a multi-year Snowflake or Databricks renewal in the next two quarters has a sharper question to answer than they did 12 months ago. The two vendors are no longer pitching analytic warehouses with AI features bolted on. They are pitching themselves as the operating substrate for enterprise agents, and the price tag that comes with that ambition is going to land on a CFO's desk well before the architecture is proven.

That reframing matters because it changes who you hire, what you commit to, and how exposed you are to a vendor whose roadmap is now entangled with frontier model economics. The decision window is short. Both companies are about to use their summit stages to lock in that narrative.

Key Details

As SiliconANGLE reported in a Breaking Analysis by Dave Vellante and George Gilbert, Snowflake Summit lands the week after May 30, with Databricks Data + AI Summit roughly two weeks behind it. The framing of both events is no longer about query performance or lakehouse versus warehouse. It is about who owns the layer where enterprise agents actually live.

The analysis lays out a five-piece AI software stack that builds on concepts from Geoffrey Moore: a system of engagement on the front, a system of agency on top, a system of intelligence in the middle, and data platforms plus systems of record underneath. The system of engagement is described as the new front end where humans and agents interact. The system of agency is where agents perceive, reason, decide, act and learn. The system of intelligence, the green layer in the middle, is where business knowledge, rules and context get organized so agents can operate on them.

Snowflake Intelligence and Databricks Genie are the products being positioned for that middle layer. Both vendors, the authors argue, crossed the Rubicon about a year ago, meaning they stopped behaving like pure data platforms and started moving up. Both are extending into governance through catalogs. On the consumer-facing side, the analysis points to ChatGPT moving toward Codex and Claude evolving toward Cowork as evidence that the engagement layer is also hardening.

The political dynamic inside enterprises has flipped too. The initial AI mandate came from CEOs and boards. The first phase of actual adoption, per the analysis, is increasingly bottom up, with individuals wiring tools into their own workflows. That gap between top-down mandate and bottom-up adoption is exactly where vendor lock-in tends to get cemented quietly.

Why This Matters for Data Teams

The honest read on this shift: if Snowflake and Databricks succeed in owning the system of intelligence, the analytics line item on your budget stops being an analytics line item. It becomes the substrate for every agent your business runs. That is a categorically different procurement conversation, and most data teams are not staffed for it.

Consider the team composition implications. A traditional analytics org is built around data engineers, analytics engineers using something like dbt for transformations, and a BI layer on top. The moment your warehouse vendor becomes the place where agent context, business rules and execution logic also live, you need people who can reason about ontology design, policy enforcement, and agent observability. Those are not the same hires. The market for that profile is thin, and the salary band is closer to ML platform engineer than analytics engineer.

Then there is the build-versus-buy question, which has gotten harder, not easier. A year ago you could reasonably argue that Snowflake and Databricks were interchangeable for most analytic workloads, with the choice driven by existing skills and contract use. Once Snowflake Intelligence and Genie become the host for your business semantics, switching costs compound. The semantic layer, the catalog, the agent definitions, the access policies, they all become vendor-specific assets. That is not a portable artifact the way a SQL model is.

My take is that data teams who treat this as just another feature release will wake up in 18 months locked into a stack they did not consciously choose. The teams that come out ahead will treat the semantic and ontology layer as a first-class architectural concern, owned independently of whichever vendor is hosting the compute this quarter.

Industry Impact

For iGaming, fintech and ad-tech platforms, the regulatory exposure here is the part that does not show up in a vendor pitch deck. If agents start touching production data, invoking tools and executing work, your General Counsel's risk register grows a new category overnight. Every jurisdiction with a data residency rule, every license condition that constrains how customer data is processed, every audit trail requirement, all of it now needs to extend into the agent layer. The system of intelligence is also the system of liability.

This is where Snowflake and Databricks moving into governance with catalogs becomes load-bearing. Catalogs are not just metadata anymore. They are where you prove to a regulator that an agent acting on a customer record was operating within policy. Teams running licensed products should be reading these summit announcements with their compliance leads in the room, not just their data leads.

The hiring market implication is equally concrete. The bottom-up adoption pattern means individual engineers are already building personal agents against company data, often without procurement or security in the loop. That genie does not go back in the bottle. Heads of Platform need to decide this quarter whether to formalize that activity into a governed internal program or keep treating it as shadow IT and absorb the breach risk. There is no third option that ages well.

For OLAP-heavy shops that have invested in alternatives like ClickHouse for real-time analytics, the calculus is slightly different. The analytic engine choice remains defensible on cost and latency grounds. The question becomes where the semantic and agent layer sits if your warehouse vendor is not racing to provide one.

What to Watch

The Head of Platform at any series-B or later fintech should be asking their CFO this week a very specific question: what does our committed spend with Snowflake or Databricks look like over the next 24 months, and how much of that commitment was made on the assumption that we are buying an analytic warehouse rather than an agent substrate? If the answer is "we have not modeled that distinction," the renewal terms being negotiated right now are almost certainly mispriced against the value being delivered.

Three signals are worth tracking out of the summit cycle. First, watch how aggressively each vendor prices the agent-adjacent features versus the core compute. Bundling tells you where they expect lock-in to materialize. Second, watch catalog interoperability claims. If neither vendor commits to portable semantic definitions, treat that as a deliberate choice. Third, watch how the application vendors respond. SAP, Salesforce and ServiceNow have their own claims on the system of intelligence, and the boundary war between data platforms and application platforms is where the next round of integration cost gets decided.

Teams evaluating their analytics stack right now should be asking themselves a simpler question: if our warehouse vendor becomes the host of our business logic, are we comfortable with that, and have we priced the exit?

Key Takeaways

• Snowflake Intelligence and Databricks Genie are positioning their vendors as the system of intelligence layer, not just data platforms, ahead of back-to-back summits.
• The five-layer stack borrowed from Geoffrey Moore (engagement, agency, intelligence, data platforms, systems of record) is a useful frame for procurement conversations, not just architecture diagrams.
• Bottom-up adoption of personal agents is outpacing top-down mandates, which means shadow IT risk is already in your environment whether or not you have approved it.
• Catalog and governance investments by both vendors turn semantic layers into vendor-specific assets, compounding switching costs in ways analytic SQL never did.
• Platform leaders renewing contracts in the next two quarters should renegotiate based on the agent substrate value, not the warehouse value, or accept being mispriced.