Snowflake vs. Databricks Reassessed: 12 Months of Pricing Drift

Twelve months ago we reviewed Snowflake and Databricks for the mid-market data team and concluded that the workload-shape — analytics-driven vs. ML-driven — was the most reliable basis for choosing between them. The conclusion still holds. What has changed in the last 12 months is the pricing leverage, the operational maturity of the multi-engine pattern, and the AI-workload pricing transparency at both vendors.

We re-tested both at four mid-market data teams during Q4 2025: two of the same organizations from last year's review (one Snowflake, one Databricks), one organization that has migrated its primary platform during the year, and one that runs both in production.

What's changed at Snowflake

Pricing leverage. The vendor's renewal posture has softened modestly over the last 6 months. Customers who push back at renewal are now seeing larger concessions than were available 12 months ago — typically 10–18% rather than the 0–5% that was common in 2024. The shift is partly competitive pressure from Databricks and partly a response to customer-side procurement maturity.

The Iceberg story has matured into operational reality. We covered this in our November piece on Snowflake's Iceberg pivot; the developments have continued. The multi-engine architecture pattern is now feasible enough that mid-market organizations are running it in production rather than in pilot.

The AI-workload pricing remains the largest single source of customer surprise. Snowflake's Cortex pricing is opaque enough that finance teams continue to flag unexpected charges at quarter-close. The vendor has improved disclosure but has not solved the problem.

What's changed at Databricks

The UX has improved meaningfully. The notebook experience, the Unity Catalog interface, and the new "ask Databricks about your data" features have produced a meaningfully more accessible product. The gap to Snowflake on operational accessibility is narrower than it was.

The pricing has stabilized at the mid-market tier after several years of compounding escalation. The Standard tier has not seen a meaningful price increase in 18 months. The Premium and Enterprise tiers have escalated modestly but well within the band of normal SaaS escalation. This is the most customer-friendly Databricks pricing posture we have tracked.

The pricing leverage has moved toward customers, slightly. Buyers should renegotiate.

The AI workload pricing remains the cleaner of the two vendors, with more predictable per-job costs and clearer disclosure of pricing for the model-serving features. Customers running production AI workloads on Databricks report fewer pricing surprises than equivalent Snowflake deployments.

On the multi-engine pattern

The pattern we identified last year — Snowflake for analytics, Databricks for ML, with Iceberg-managed shared storage — is now common enough to call mainstream at growth-stage data organizations. The operational complexity is real and the cost is non-trivial, but the architectural benefits compound for organizations whose work product spans both analytics and ML.

We have watched two test sites move from single-vendor to multi-engine architectures over the year. Both report meaningful operational benefit. Neither describes the migration as easy.

What to do at renewal

Both vendors are now genuinely negotiable. The procurement playbook for 2026 renewals: start the conversation 90 days early, model your usage profile against both vendors' current SKU structures, present a credible willingness to migrate, and push back on automatic-escalation provisions. Customers running both can use one vendor's pricing as a credible threat against the other.

The combined effect, in our reading, will be modest single-digit percentage savings against rack-rate pricing for most customers and double-digit savings for customers willing to genuinely consider migration.

The verdict

Snowflake for analytics-shaped workloads. Databricks for ML-shaped workloads. The "both" pattern for organizations that span the two. The pricing has moved enough that buyers should re-model and renegotiate. The procurement leverage has shifted slightly toward customers; this is the first year in five we can say that about the data-platform category.