What’s New in Snowflake Interactive Analytics: Spring 2026
Since the general availability (GA) of Snowflake Interactive Analytics on AWS in December 2025, and the subsequent expansion to Azure and GCP, auto-suspend (GA), auto-resume (GA), Search Optimization (GA) and storage lifecycle policies (GA), we've been learning a tremendous amount from customers deploying interactive tables and warehouses in production. Today, we're sharing four new capabilities that directly address feedback we've heard from customers operating real workloads at scale:
- A fallback path for mixed query workloads (no more five-second timeout)
- AI-assisted clustering key selection
- Cortex Code AI Skills to accelerate deployment of Interactive Analytics
- Replication (GA) and Auto-Scaling (GA)
The challenge of mixed workloads
Interactive warehouses are purpose-built for speed. They are optimized for short, highly concurrent reads, the kind that power dashboards, data APIs and AI agents querying structured data in real time.
But production workloads are rarely uniform. A retail analytics platform might serve thousands of dashboard queries per second, while occasionally an analyst fires a complex ad-hoc query. A fintech platform might serve sub-second analytical dashboards alongside periodic reconciliation queries that scan much larger ranges of data. Before today, the only option was a compromise: accept hard failures when an occasional heavy query runs long. This is not acceptable in production.
Feature 1: Fallback warehouse — no more five-second timeout
We're introducing Fallback Warehouse support for interactive warehouses. With a single configuration line, you can now designate a secondary warehouse to retry queries that exceed the timeout on your interactive warehouse:
ALTER WAREHOUSE interactive_wh SET FALLBACK_WAREHOUSE = 'fallback_wh';When a query running on interactive_wh exceeds 5 seconds, instead of returning an error to the client, Snowflake transparently retries that query on fallback_wh. The client sees a result and no disruption. No error handling. No manual retry. No application changes required.
A few important details:
- The fallback warehouse can be any warehouse type and any size
- The feature introduces zero overhead for queries that complete within the primary timeout
This pattern lets operators keep interactive warehouses tuned aggressively — high concurrency, fast SLA — without creating a cliff edge for workloads that occasionally need more compute. The discipline of the interactive warehouse is preserved; the flexibility of the broader Snowflake platform absorbs the outliers.
Why this matters: In observability, BI and operational analytics, the P99 query matters as much as the P50. Fallback Warehouse gives you a top-level mechanism to handle both, within a single unified platform, without building retry logic into your application layer.
Feature 2: AI-assisted clustering key selection
Clustering keys are fundamental to interactive table performance. They determine how data is physically organized on disk, which micropartitions are pruned away at query time, and ultimately, how consistently your workload hits sub-second latency targets.
Choosing the right clustering key requires balancing multiple factors: the filter patterns in your queries, the cardinality of candidate columns, data distribution, refresh cadence and the cost implications of clustering churn. It's one of the most impactful decisions in the interactive tables setup experience, and one of the most common sources of misconfiguration we see in production deployments.
Critically, clustering keys on interactive tables cannot be changed after creation. Unlike standard tables, where you can iterate freely, an interactive table's clustering key is fixed at creation time. Choosing wrong means recreating the table.
We have released an interactive clustering skill for Cortex Code that brings expert-level clustering guidance directly into your development workflow. You can ask natural-language questions like:
- "Help me choose a clustering key for this interactive table"
- "What columns should I cluster by given these queries?"
- "Is my current clustering key well-suited for this workload?"
Paste a DDL, share representative queries, and the skill reasons through the options and recommends a clustering configuration with explanation. Critically, the skill enforces an explicit approval gate before generating any CREATE INTERACTIVE TABLE DDL. Given the immutability of the clustering key, this guardrail helps ensure no table is created before the operator has reviewed and confirmed the recommendation. It also knows when to recommend Search Optimization Service as a complement for high-cardinality equality filters that fall outside the clustering key's scope.
Feature 3: The Snowflake Interactive Skill — an AI copilot for the full experience
The clustering skill above addresses one high-stakes moment in the interactive tables journey. But the reality is that getting an interactive workload production-ready involves many more decisions: how to size and configure a warehouse, how to handle data freshness, how to structure queries for the 5-second timeout constraint, how to implement data mutation patterns on tables that don't support UPDATE or DELETE natively, and how to diagnose issues when something goes wrong.
Historically, navigating all of this required either deep Snowflake expertise or significant trial and error. We are changing that. We have released the Snowflake Interactive Skill in Cortex Code, a comprehensive, intent-aware AI assistant that guides developers through the complete interactive tables and warehouses lifecycle. Rather than a single-purpose helper, this is a full workflow skill that routes automatically to the relevant guidance based on what you're trying to do.
How it works: The skill listens to your intent and routes to the appropriate workflow. You don't need to know which sub-skill to invoke. Just describe what you need:
| What you say | What happens |
|---|---|
| "I'm new to interactive tables, help me get started" | Walks you through sizing, table selection and first-time setup |
| "Create an interactive table from my orders table" | Guides you through static vs. dynamic choice, clustering selection, warehouse sizing |
| "Add tables to my interactive warehouse" | Handles the ALTER WAREHOUSE ADD TABLES workflow end-to-end |
| "My queries are timing out after 5 seconds" | Diagnoses the cause and recommends remediation |
| "I need to update data in an interactive table" | Explains the standard + dynamic pattern and generates the DDL |
| "Help me benchmark interactive vs. standard table performance" | Sets up a proper comparison query and measures results |
| "What should I cluster by?" | Loads the clustering sub-skill and applies workload-driven heuristics |
Guardrails baked in: The skill enforces stopping points before any mutation. Read-only diagnostics and query generation run freely, but no CREATE, ALTER or ADD TABLES statement is executed without explicit user approval. This is especially important in a context where:
- Clustering keys are immutable after creation
- Adding a table to a warehouse triggers a cache warming operation
- Warehouse sizing directly affects both latency and cost
What this represents: The interactive skill is part of a broader architectural shift in how Snowflake delivers platform expertise. The goal is not to document what's possible. It is to make the right path the default path. When you open Cortex Code and describe an interactive analytics problem, you should receive production-quality guidance, not a pointer to documentation. For teams adopting interactive tables for the first time, this skill reduces the time from "I want sub-second dashboards" to "my dashboard is live" from days to hours. For experienced teams, it surfaces best practices and catches common configuration errors before they become production incidents.
Feature 4: Enterprise readiness — replication and auto-scaling now generally available
Performance at sub-second latency is table stakes. But for organizations running Interactive Analytics as a core part of their business, powering customer-facing products, operational dashboards, or SLA-bound data APIs, the question isn't just how fast, it's how reliable. Two capabilities that move Interactive Analytics into the enterprise readiness tier are now generally available.
Replication and Failover (generally available): For mission-critical applications, downtime is not an option. Interactive tables can now be included in your primary replication and failover groups, just like standard tables. This means you can replicate your interactive data assets across regions or cloud providers and fail over to a secondary environment with the same guarantees you rely on for the rest of your Snowflake data. This is particularly significant for regulated industries and customer-facing applications where RTO and RPO requirements apply to every data tier, not just transactional systems. Interactive tables are no longer a special case in your business continuity architecture; they're a first-class participant.
Auto-Scaling for Interactive Warehouses (generally available): Interactive warehouses can now scale automatically in response to concurrency demand, removing the need to manually adjust cluster counts during peak load. This brings interactive warehouses to parity with the workload management capabilities that Snowflake customers have grown to love.
Conclusion
As always, if you've already used Snowflake, the mental model is familiar. What's new is how much further the platform now carries you — from the first table creation decision to a fully replicated, auto-scaling, production-hardened deployment.
Our commitment is to continuously improve and listen to feedback our customers give us to build a great product. Since GA, we've rolled out many new features and we aren't stopping. Stay tuned, there's more to come.
Authors
