Company Deep Dive #6: SNOWFLAKE
Data cloud to power analytics and AI
We’ll be diving into Snowflake today.
1. What is Snowflake and why does it matter?
Snowflake is a cloud-native data platform used by more than 11,000 organizations to store, combine, and analyze data. Think of it as a managed backbone where companies dump everything from sales transactions to product logs, then run analytics and AI on top without buying hardware or stitching together a dozen tools. It runs on AWS, Azure, and Google Cloud. And it separates storage from compute so customers pay only for what they use when they query or process data.
Why does this matter for people who build or invest in infra startups? Snowflake has become a default “home base” for analytics and AI data pipelines. When Snowflake grows, adjacent categories see more demand e.g. ETL/ELT, orchestration, observability, governance, semantic layers, BI, ML platforms, vector search, and data marketplaces. When Snowflake slows, enterprise budgets tighten around the whole modern data stack. In short, Snowflake is a bellwether for data-first infrastructure.
What the recent numbers tell you at a glance: product revenue is just under $1 billion a quarter and growing in the mid-20% range. Net dollar retention sits in the mid-120s, which is a sign customers expand once they’re in. Gross margins are mid-70s on a non-GAAP basis. Free cash flow is healthy for a high-growth platform. The headline deceleration from hypergrowth down to “merely” strong growth is real. But the customer base (Global 2000 penetration and >$1M-spend accounts) keeps expanding. That mix points to durability even as growth normalizes.
2. What Snowflake actually does (in user terms)
Snowflake hosts your core analytical data in one place. Data engineers pipe raw data from SaaS apps, databases, and event streams. Snowflake stores it. Analysts query it with SQL. Data scientists train or score models against it. Business teams consume dashboards on top. The platform handles security, scaling, and performance so teams don’t have to babysit clusters. Two design choices matter most:
Separation of storage and compute. You can park a lot of data cheaply and spin up compute for specific jobs only when needed. That’s why finance teams like Snowflake: the bill tracks usage.
Multi-cloud neutrality. Snowflake runs on the big clouds. If you’re an AWS shop (and most Snowflake workloads are), you can still avoid lock-in to Redshift. If you prefer Azure or GCP, you get the same Snowflake experience there.
Over the last 2 years, Snowflake has broadened beyond “warehouse” into an “AI data cloud”. This includes built-in features for data sharing, governance, Python, ML, and a marketplace where vendors offer data products and apps. It has also leaned into builders: programs like “Powered by Snowflake” for ISVs, startup credits and an AI accelerator, plus investments via Snowflake Ventures. That ecosystem pull is why a lot of infra startups list a Snowflake integration on day one.
3. Market tailwinds and the size of the prize
There are two forces you can rely on:
Data keeps compounding. Every product emits logs and events. Every business team wants metrics and insights. Compliance needs lineage and audit trails. Even companies that don’t consider themselves “data companies” now ingest terabytes monthly.
AI turns “nice-to-have analytics” into “must-have infrastructure”. If you want AI in production, you need curated, governed, queryable data at scale. That’s not optional.
Depending on definitions, the “data warehouse / data platform” market sits in the tens of billions and is growing quickly. What matters more than any top-down TAM slide is penetration and runway. Snowflake already serves hundreds of Global 2000 companies, yet most large enterprises still have pockets of legacy on-prem warehouses, departmental silos, and bespoke pipelines that haven’t been modernized. Mid-market penetration is even earlier. If you’re underwriting the space, the shape looks like a long adoption curve with multiple expansion vectors — more data sources, more use cases (real-time, AI, governance), and more regions and business units.
For startups, those tailwinds translate into predictable demand patterns. Once a customer standardizes on Snowflake, they go looking for better ingestion, faster transformations, cheaper storage tiers, smarter cost controls, higher-quality governance, and applications that sit directly on the warehouse. That creates room for specialized infra products so long as they reduce time-to-value or total cost of ownership versus do-it-yourself.
4. How Snowflake makes money and what its unit signals mean
Snowflake’s revenue model is consumption-based. Customers buy credits and these credits are spent when they use compute (running queries, loading data, ML jobs) and on storage. The practical result: land with a small workload, then expand as usage grows. That’s why net dollar retention sits around the mid-120s. Once a team loads critical data and connects dashboards, the organization tends to add more data and more users over time.
The company’s non-GAAP gross margins are in the mid-70s, reflecting that it resells cloud compute/storage with value-add software on top. Operating margins are positive on a non-GAAP basis but still modest. GAAP profitability remains weighed down by stock-based comp, typical for growth platforms. Free cash flow is strong, aided by upfront deals and efficient support relative to the size of the dataset under management.
The two things founders should read from these signals:
Expansion is the core flywheel. If you sell into Snowflake customers, design your product and pricing to expand in step with data and seat growth i.e. usage-based SKUs, straightforward attach, and visible ROI in the admin’s cost dashboard.
Cost visibility matters. Snowflake’s consumption can surprise finance teams if workloads aren’t governed. Tools that give spend predictability, workload scheduling, and performance tuning have tangible value. If you provide them, quantify savings in hours and dollars.
5. The ecosystem: partners, marketplace, and why co-selling works
The center of gravity is AWS. A large majority of Snowflake’s workloads run there. That deep alignment (plus similar tie-ups with Azure and GCP) leads to co-sell motion where Snowflake and the cloud provider bring each other into enterprise deals. For startups, this means two things:
If you’re building a product that complements Snowflake, become a first-class integration and explore the partner programs. “Powered by Snowflake”, the Marketplace, and Snowflake’s startup accelerator can shorten your path to customers. AWS credits targeted at Snowflake-building startups sweeten early runway.
If you’re building an alternative data platform, you’ll be competing not only with Snowflake’s direct sales team but also with its partner field resources and an installed base invested in Snowflake skills, tooling, and data models. Your wedge needs to be sharp: a 10x on a real pain point, not a 20% improvement.
The Marketplace angle deserves emphasis. It allows data vendors and app builders to sell to Snowflake customers without complex deployment. That lowers distribution friction for startups, especially in vertical data products (healthcare, fintech, climate), synthetic data, enrichment feeds, and analytics apps. For founder planning pipeline, that channel can be the difference between a long enterprise slog and a repeatable sales motion.
6. Competition and pricing dynamics
The big three cloud providers all offer their own analytic databases: Redshift (AWS), BigQuery (Google), and Synapse/Fabric (Microsoft). Databricks is the cross-cloud heavyweight with a focus on the “lakehouse” and AI tooling. Each has rational appeal:
Cloud-native warehouses integrate tightly with their own clouds, sometimes with bundle economics that are hard to ignore if a customer is “all-in” on a given vendor.
Databricks resonates where data science and ML are the main events, and where notebook-centric developer workflows are the norm.
Snowflake wins on simplicity, predictable performance, governance, and true multi-cloud neutrality, especially in organizations that need to share data across partners or acquisitions.
Pricing is rarely apples-to-apples and often comes down to total cost of ownership, not list price. A “cheaper” engine that demands more ops or yields longer runtimes can cost more overall. Where Snowflake loses, two reasons show up often: (1) a customer standardizes on the native cloud stack for consolidation simplicity (2) heavy ML shops prefer Databricks’ developer ergonomics. Where Snowflake wins, customers cite low-friction scaling, governance, data sharing, and fewer knobs to tune.
For startups, the takeaway is to assume a multi-platform world and meet customers where they are. If your product only works with one warehouse, you’ve narrowed your market more than you think. If it works better with one (e.g. deep Snowflake integration) but still functions elsewhere, you keep optionality while riding Snowflake’s distribution.
7. Financial health as a signal for investors and your sales pipeline
Snowflake’s latest prints show mid-20s growth off a multi-billion revenue base, high net retention, rising large-customer counts, mid-70s non-GAAP gross margins, and strong free cash flow. At the same time, the growth curve has flattened from earlier years. And GAAP profitability is still out of reach. Here’s how to interpret that for your own planning:
Budgets are expanding, but with scrutiny. Data teams still spend, but CFOs want predictability. Products that help customers hit value targets or control spend continue to land. The “nice-to-have” tools struggle.
Pipeline quality > pipeline size. With Snowflake growing steadily but not explosively, downstream categories don’t get automatic lift. Founders should prioritize ICP discipline and ROI-first messaging over broad, top-of-funnel experiments.
Stock and cash position fuel the ecosystem. As long as Snowflake’s free cash flow is solid and the market rewards durable growth, programs like credits, marketplace incentives, and startup acceleration remain funded. If that changes, discount rates across data infra tighten quickly.
A practical heuristic: if Snowflake’s net retention ticks down meaningfully or RPO growth slows, expect longer sales cycles across the stack two to three quarters later. Conversely, when Snowflake’s large-customer growth re-accelerates, expect higher attach rates for complementary tooling shortly after.
8. How many infra startups are actually exposed
Not every infra startup lives and dies with Snowflake, but many are intertwined. Here’s a directional breakdown based on how customers actually build:
Directly dependent (high exposure): ETL/ELT into warehouses. Reverse ETL. Data quality/observability for SQL/warehouse pipelines. Governance and cataloging tuned to warehouse metadata. Semantic layers and BI that query Snowflake. Usage optimization/cost control for Snowflake. If your product inherently assumes a warehouse and Snowflake is the most common one, you’re in this bucket. Roughly 40-45% of venture-backed data-infra startups in the market fall here by function.
Indirectly dependent (moderate exposure): ML platforms that read/write from the warehouse. Vector/search layers that sync embeddings from warehouse tables. Privacy/security overlays. Data applications “powered by” the warehouse. Vertical analytics where Snowflake is the default store. A large slice of the remainder sits here, especially in AI application infrastructure.
Loosely coupled (low exposure): Core developer tools, CI/CD, general-purpose observability for app services, container/Kubernetes tooling, and non-warehouse databases (OLTP, time-series, graph) that often live alongside but not inside the warehouse footprint. These still feel macro data headwinds/tailwinds but don’t map 1:1 to Snowflake cycles.
If you’re fundraising, assume your investor will ask: “What % of your current ARR is in Snowflake-centric accounts? And what % of your pipeline references Snowflake explicitly?”. You should know both numbers. If >60% of your ARR depends on Snowflake usage expanding, your growth will correlate to their cycle. That’s not inherently bad, but you should explain your diversification plan by warehouse, by cloud, and by use case.
Two more practical notes:
ISVs building on Snowflake’s platform (e.g. Marketplace apps) can scale faster with lower deployment friction. The trade-off is platform dependency. Price that risk into your roadmap and contracts (e.g. plan for a second platform by a certain ARR milestone).
Data vendors selling feeds directly in the Marketplace can win on distribution. The edge comes from curation and refresh SLAs. If your data is “commodity”, the Marketplace’s transparency can compress margins unless you add proprietary enrichment or quality guarantees.
9. Risks, correlations, dependencies: How to monitor them without handwaving
Macro and budget risk. When CFOs tighten spend, consumption platforms are the first to feel it. Usage throttles, postponed migrations, and smaller commitments ripple through to everyone who sells into the same teams. Watch for early signs in smaller cohorts (SMB/mid-market churn rises first) and in Snowflake’s remaining performance obligations growth. If bookings slow, downstream products will feel it.
Competitive pressure. If Microsoft’s Fabric bundle or Google’s BigQuery pricing makes “all-in” economics too attractive, some customers will consolidate. Likewise, Databricks can pull workloads where ML is central. Founders should take this as a design constraint: make your product work well across at least two data backbones and keep your total value proposition independent of any single vendor’s roadmap.
Cloud dependence. Snowflake rides on the hyperscalers’ rails. If underlying cloud costs change or relationships shift, Snowflake’s own unit economics and pricing could move. In practice, that would show up as either pass-through price changes or new SKUs that nudge customers to particular usage patterns. Products that help customers optimize spend become more valuable in those moments.
Execution and reliability. Incidents, security issues, or delayed roadmap items can shake confidence. Keep a lightweight risk register: note uptime status, incident post-mortems, and major feature delivery slippage. If you sell governance or reliability tooling, these become selling moments. But tread carefully and sell value, not FUD (fear, uncertainty, and doubt).
Regulatory and data residency. New privacy rules or localization requirements can complicate centralized data strategies. If you build security, lineage, synthetic data, or anonymization tooling, these shifts are opportunity. If you’re a Marketplace ISV, stay ahead on certifications and document the controls your customers’ compliance officers will ask about.
Correlation mechanics for planning. A simple way to quantify exposure: calculate the share of your revenue and pipeline tied to customers who list Snowflake as a strategic platform, then apply a haircut equal to Snowflake’s last-reported net retention delta if it deteriorates (e.g. if NRR went from 124% to 118%, model a 5–10% drag on your expansion-led growth). It’s not precise, but it forces explicit assumptions and helps set board expectations.
10. Near-term catalysts and practical playbooks for founders + investors
Product catalysts. Snowflake’s continued push into AI (native inference, feature stores, Python ergonomics) and its Postgres move expand the reachable developer base. If these land well, expect more mid-market adoption and more “apps-on-the-warehouse” startups. For ISVs, that’s a bigger total market and more reasons to integrate deeply.
Go-to-market catalysts. Marketplace improvements and stronger co-sell with the hyperscalers can shorten enterprise sales cycles. If you’re an ISV, align your enablement with Snowflake’s field teams and speak their value language: governance, time-to-insight, and cost visibility. If you’re a challenger platform, concentrate on verticals or workloads where you’re dramatically better. You won’t out-distribution Snowflake, but you can out-outcome it in a niche.
M&A and partnerships. Snowflake will keep buying capabilities that help them win developer mindshare and AI workloads. Each acquisition is a signal about adjacencies they consider strategic. If you’re in one of those adjacencies, decide whether to lean in as a partner, differentiate sharply, or pivot up/down-stack.
Fundraising timing. When Snowflake prints clean quarters (steady growth, strong RPO, upbeat guidance), investors re-risk in data infra. If prints wobble, valuations compress first in the “picks and shovels” around the warehouse. If you’re raising, aim to announce on the heels of strong industry prints. If you must raise in a soft patch, show more concrete ROI and clear multi-platform support.
Sales playbooks that work today. Three approaches can land here:
Prove a hard dollar cost reduction on Snowflake spend (e.g. 20–30% savings) with a two-week pilot.
Show a time-to-insight improvement that unblocks a specific recurring workflow (e.g. log analysis or revenue ops dashboards) and quantify the saved hours per month.
Deliver a governance/compliance control that a CISO or risk team needs for an audit this quarter.
If your pitch depends on “eventual” benefits or indefinite “AI readiness”, reframe it into one of the above.
11. What would change our view
If you’re using Snowflake as a market barometer, keep a simple dashboard:
Growth cadence: Is product revenue growth holding in the mid-20s or re-accelerating? A steady 25–30% off a multi-billion base is bullish enough for the ecosystem.
Net dollar retention: Above 120% signals healthy expansion. Sustained drift below that hints at spend discipline or competitive encroachment.
Large customer counts: >$1M-spend customers and Global 2000 penetration are the “quality of demand” metrics.
Bookings: This is forward demand. If this slows, plan a conservative second half.
Free cash flow: Healthy free cash flow keeps ecosystem programs funded. A squeeze here means fewer credits and tighter co-marketing.
Incident/uptime trend and roadmap delivery: Confidence indicators for buyers and for ISVs riding along.
Three positive triggers that would make the case for being more aggressive in backing Snowflake-adjacent companies:
Re-acceleration of large-customer adds and NRR trending back toward mid-120s or higher.
Material Marketplace GMV growth with improved ISV monetization terms.
Clear technical wins in AI features that demonstrably shift workloads from external tools into Snowflake.
Three negative triggers that would lead to more caution:
Two or more quarters of NRR drifting toward ~115% with flat RPO growth.
Hyperscaler bundle pressure that pulls major accounts back to native warehouses.
Reliability or security incidents that cause high-profile churn.
Putting it all together: What this means for infra startups and investors
Snowflake is no longer a scrappy disruptor. It’s part of the core enterprise fabric. That status brings both stability and gravity. Stability, because the platform has proven staying power, deep compliance, and entrenched workflows. Gravity, because it bends the rest of the data stack around it: partners, pricing, and buyer expectations.
If you’re a founder:
Pick your angle consciously. Be the accelerant (cost control, performance, governance), the specialist (vertical apps, domain data, compliance), or the alternative (superior engine for a specific workload). Don’t be a fuzzy middle.
Design for co-sell. Align your messaging with Snowflake’s outcomes. Show how you help customers extract more value from their existing Snowflake investment without surprise costs.
Diversify with intent. Even if Snowflake is your best channel, prioritize a second platform by a revenue milestone so you’re resilient to vendor cycles.
If you’re a VC or corp dev lead:
Expect correlation. A portion of your data-infra portfolio’s growth will correlate with Snowflake’s cycle. Model it. Use Snowflake’s quarterly signals to adjust reserves and pace.
Underwrite distribution. Products that can ride the Marketplace or co-sell motions deserve higher go-to-market credit in your model. Challengers need clearly superior unit outcomes to offset distribution disadvantage.
Watch for adjacencies Snowflake treats as strategic. Those areas will either be great M&A outcomes for portfolio companies or regions where Snowflake’s native products compress startup room. Time your bets accordingly.
The short version: the “data cloud” has matured into a dependable layer of enterprise infrastructure. Snowflake’s growth may not be the rocket ride it once was. But its combination of broad adoption, usage expansion, and cash generation makes it a reliable anchor for the ecosystem. For infra startups, that reliability can be an advantage. If you plug into it smartly, measure your exposure, and keep a second engine humming in case the winds shift.
If you are getting value from this newsletter, consider subscribing for free and sharing it with 1 infra-curious friend: