How Snowflake Makes Money: Consumption Pricing, NRR, and the Path to Profitability

Executive Summary

Snowflake is one of the most closely watched companies in enterprise software — both because its consumption-based pricing model is genuinely differentiated and because the path to GAAP profitability has been slower and more complex than the bull case assumed at the time of its record-breaking 2020 IPO. In fiscal year 2026 (ending January 2026), Snowflake is expected to generate approximately $3.5B in product revenue, growing at roughly 25-28% year-over-year, with a net revenue retention rate that has declined from its peak of 178% to the 127-130% range. Understanding the revenue model — especially the consumption mechanic — is essential to forming a view on whether this business is a durable compounder or a growth story in secular deceleration.

The Business Model in Plain English

Snowflake sells cloud data warehousing, data lake, and broader data platform capabilities as a service. Unlike traditional SaaS, which charges a fixed monthly or annual subscription fee regardless of usage, Snowflake charges by the compute credit consumed when customers query or process data. Storage is charged separately at near-commodity rates.

The implication is significant: Snowflake's revenue goes up when customers run more queries and processes on its platform, and down when they optimize or reduce workloads. This creates a fundamentally different relationship between customer value delivery and revenue recognition than subscription SaaS.

Customers are typically sold on-demand credits or, more commonly for enterprise accounts, pre-purchased capacity contracts — commitments to spend a minimum amount over a defined period (typically 1-3 years). The pre-purchased capacity model provides revenue visibility but also creates complex dynamics around consumption ramp, overage, and renewal negotiation.

Revenue Streams Breakdown

Within product revenue, the split is approximately:

• Compute (virtual warehouse credits): ~75-80% of product revenue
• Storage: ~10-12% (charged at ~$23/TB/month compressed, declining as cloud providers cut storage costs)
• Data transfer / Marketplace / other: ~8-12%

Snowflake Marketplace — where data providers sell curated datasets directly to Snowflake customers — is a growing but still small segment. It represents significant long-term optionality: Snowflake earns a revenue share (approximately 25%) from marketplace transactions without bearing data collection costs.

Snowflake's Iceberg Tables initiative, which allows customers to use open-format storage (Apache Iceberg) that they own rather than Snowflake's proprietary format, is a strategic concession to customer data portability demands. In the near term this is neutral-to-positive for compute revenue; in the long term it modestly reduces switching costs.

Unit Economics

Snowflake's unit economics are strong at the gross margin level but the profitability picture requires careful interpretation:

The NRR story is the central debate in Snowflake analysis. At 178% NRR in FY2022, the company was essentially printing money from its existing customer base — each cohort was expanding so rapidly that Snowflake barely needed new logo acquisition to hit 80%+ growth rates. NRR has since compressed as:

1. Large enterprise customers optimized their Snowflake spend (data teams got better at query optimization)
2. New customer cohorts are smaller companies with lower organic expansion trajectories
3. Economic pressure in 2023-2024 caused many customers to scrutinize cloud spend

Even at 127-130%, Snowflake's NRR is elite — it compares favorably to most SaaS peers. But the direction of travel matters as much as the absolute level.

Customer concentration is modest — the top 10 customers represent less than 10% of revenue. The $1M+ ARR customer count has grown to approximately 570 customers as of recent quarters, representing roughly 65-70% of total product revenue. This is a healthy enterprise customer distribution.

Why the Model Is Durable

1. Data gravity. Once a customer has migrated petabytes of data to Snowflake and built hundreds of data pipelines on it, the switching cost is enormous. Data gravity is not just about moving bytes — it is about re-engineering the entire analytics stack, retraining teams, and rebuilding trust with downstream users of that data. Snowflake's churn rate is approximately 2-3% annually, which is exceptional for any enterprise software product.

2. Cross-cloud and multi-cloud positioning. Snowflake runs on AWS, Azure, and GCP and allows customers to share data across clouds and across organizations. This cross-cloud utility is genuinely difficult for hyperscaler-native alternatives (Redshift, BigQuery, Synapse) to replicate — Amazon is not inclined to help you share data easily with Azure.

3. Data Sharing and Snowpark ecosystem. Snowflake Data Sharing allows organizations to share live data with partners, customers, or regulators without copying data. Snowpark allows customers to run Python, Java, and Scala workloads natively. Both features increase the surface area of use cases on the platform, which drives incremental credit consumption.

4. AI and ML workloads. Snowflake Cortex AI, announced in 2024 and expanding through 2025-2026, enables LLM inference, embedding generation, and ML training on Snowflake-managed data. AI workloads are compute-intensive — a single LLM inference run consumes orders of magnitude more credits than a traditional SQL query. If enterprises run AI workloads on their Snowflake data (which many will), this could meaningfully re-accelerate consumption growth.

Comparison to Closest Competitors

Databricks is the most threatening competitor. It has grown faster than Snowflake over the past 18 months, driven by the lakehouse architecture (combining data lake flexibility with warehouse performance) and its leading position in ML/AI workloads (MLflow, Delta Lake). Databricks's pending IPO will sharpen the competitive narrative significantly.

Hyperscaler alternatives (BigQuery, Redshift) are deeply discounted or bundled for customers committed to a single cloud, which limits Snowflake's pricing power in single-cloud enterprise deals.

What the Model Looks Like at Scale

At $10B+ revenue (a realistic 5-year target if growth continues in the 20-25% range), Snowflake's economics should look like:

• Product Gross Margin: 80-82% (further cloud cost optimization)
• FCF Margin: 35-40% (operating leverage on G&A, R&D, and S&M)
• Non-GAAP Operating Margin: 25-30%

The critical variable is compute cost deflation vs. AI workload growth. If enterprises shift significant AI inference workloads to Snowflake Cortex, revenue per credit effectively rises (higher-value workloads). If customers continue optimizing SQL queries, the organic expansion rate per customer remains muted.

Snowflake's sales motion is also shifting. Historically, land-and-expand was almost automatic — customers expanded organically. Now the company must invest more in structured expansion selling, which is showing up in S&M expenses growing faster than revenue in recent periods.

Red Flags and Risk Factors

1. NRR compression is not over. The NRR peaked at an unsustainable level. Current guidance implies it stabilizes in the 127-130% range, but if economic conditions tighten or if Databricks captures incremental workloads in existing Snowflake accounts, further compression to 120% or below is possible.

2. GAAP profitability remains elusive. Snowflake's stock-based compensation is exceptional in scale — SBC has run at 30-35% of revenue. While FCF is positive, GAAP losses create an overhang and make traditional earnings-based valuation impossible.

3. Competitive intensity from Databricks. The lakehouse model is winning new workload categories. If Snowflake is losing the incremental data + AI workload to Databricks, the long-term ceiling on NRR is lower than the bull case implies.

4. CEO transition risk. Frank Slootman's departure in February 2024 removed one of the most respected enterprise software operators from the helm. New CEO Sridhar Ramaswamy (former Google Ads SVP) brings AI expertise but less enterprise infrastructure operating experience. Execution risk is elevated during leadership transitions.

5. Valuation. At 15-18x forward revenue (typical 2025-2026 trading range), Snowflake is priced for continued strong execution. Any quarter where product revenue growth decelerates meaningfully below guidance produces severe multiple compression.

Takeaways for Investors

1. The consumption model is a double-edged sword — extraordinary when workloads grow, painful when customers optimize. The AI workload thesis is the bull case re-accelerant.
2. NRR is the metric that matters most — monitor it closely for signs of stabilization vs. continued compression.
3. Databricks is a genuine threat, not FUD. The competitive landscape in data platforms is the most contested in enterprise software in 2026.
4. FCF is the right profitability lens — GAAP losses overstate the cash burden because SBC is a real but non-cash cost. FCF margins approaching 30% are strong.
5. The Cortex AI bet is the highest-conviction long-term catalyst — if Snowflake successfully converts its installed base to AI workload execution on the platform, the consumption re-acceleration could be substantial.
6. Position sizing matters — Snowflake is a high-conviction, high-volatility name. Execution risk is elevated post-CEO change. Appropriate for growth portfolios; size accordingly.