06-reference/research

snowflake marketplace provider economics

2026-10-16·research-brief·source: deep-research·by Ray Data Co (deep-research synthesis)
snowflakemarketplacedata-marketplaceprovider-economicsphdata

Snowflake Marketplace Provider Economics: Revenue Share, Listing Requirements, and Buyer Discovery

Headline finding: the take rate is real but login-gated, the infrastructure tax on a small listing is roughly 2% of bookings, and the binding cost is human hours.

The question

Verbatim: "What are Snowflake Data Marketplace's actual provider economics (revenue-share terms, listing requirements, buyer discovery mechanics) — concretely enough to evaluate as a data-marketplace entry point?"

Context: the [[01-projects/data-marketplace/index]] What's-Needed list names "Understand Snowflake marketplace economics" as an explicit open item. Filed from the 2026-08-11 curiosity sweep. Double relevance: the dormant data-marketplace bet, and phData Snowflake domain knowledge.

What we already know (from the vault)

What the web says

Naming precision, as of 2026-10-16. "Snowflake Data Marketplace" is now just Snowflake Marketplace. Everything in it is a "Data Product," covering three listing types: data shares, Snowflake Native Apps, and Connected Apps. Managed applications and professional services are not eligible for listing (Provider and Consumer Policies). Cortex Knowledge Extensions are now a first-class marketplace category, not a separate surface. Everything below describes the public Snowflake Marketplace provider path unless stated otherwise.

1. Revenue share: a fee exists, the number is not public

Snowflake does not publish a take rate, percentage, or transaction fee anywhere publicly reachable. It confirms the fee exists and puts the number behind a Snowsight login. This is the single most important finding and it is a negative one.

The confirming sentence, provider-facing: "Snowflake Marketplace charges a transaction fee, which is deducted from the Provider's payout," with the instruction to "Find the fee schedule by navigating in Snowsight: Admin → Billing & Terms → Fee" (provider-transactions-invoicing-collections).

The contract does the same thing. The Snowflake Provider and Consumer Terms (effective 2026-03-31) say the customer "is responsible for the Snowflake fees set forth in the relevant Monetization Offering Fee Schedule" (§3.5 Exh. A), and §12.30 defines that schedule as one "made available in Snowsight or as described in the Documentation." The fee is incorporated by reference into a document Snowflake maintains unilaterally. It is not an exhibit to the terms. Terms changes take effect 30 days after posting (§11.8); no notice period is stated for fee-schedule changes.

The net-payment formula is published even though the inputs are not: payout "shall equal the Product Cost minus: (i) the Fees; (ii) any Taxes" plus pre-noticed expenses (§3.5 Exh. A).

Do not quote a percentage. A third-party site surfaced in search claiming "25-30% commission" with no sourcing; that is unverified and is not cited here. The only percentage published anywhere in the corpus is a 1.5% per month late charge (§2.2 Exh. A), which is not a take rate. The single action that produces the real number is an organization-administrator (ORGADMIN) login to Snowsight → Admin → Billing & Terms → Fee.

Does it differ by listing type? Partially answerable:

Other commercial terms that matter and are public:

2. Listing requirements and cost to become a provider

Sources: provider-becoming, provider-listings-creating-publishing, provider-profiles-managing, provider-listings-workflows.

Cross-cloud and cross-region replication, who pays (provider-listings-auto-fulfillment, provider-understand-cost-auto-fulfillment). A listing visible in another region means the data is physically replicated there, into a Snowflake-managed secure share area (SSA), one per region per organization. The provider pays: "When you make a data product available in other regions, you incur additional costs" covering replication compute, duplicated storage, and cross-region egress. Crucially, "these usage costs are attributed to one Snowflake-managed secure share area for each region" under the provider's organization, not to the consumer's account. That is the opposite of manual replication, which bills the target account. Refresh intervals run from one minute to eight days and drive the compute bill. The Egress Cost Optimizer caches data for 15 days so that post-seed egress is charged only on updates. Business Critical is not required; Tri-Secret Secure on SSAs requires a support ticket. Auto-fulfillment is unavailable on trial accounts. Monitoring views: ORGANIZATION_USAGE.LISTING_AUTO_FULFILLMENT_USAGE_HISTORY and LISTING_AUTO_FULFILLMENT_REFRESH_DAILY.

3. Buyer discovery mechanics

Sources: consumer-listings-exploring, consumer-listings-access, provider-listings-managing, provider-listings-monitor-studio, plus the live public catalog measured 2026-10-16.

4. The honest unit economics

This subsection is Ray's own model built on the sourced facts above, not web content. Every number below is an assumption, labeled as one. The take rate is unknown, so it is modeled as a sensitivity rather than asserted.

Assumed provider and product

Assumed infrastructure costs. List prices are third-party-reported consensus for AWS US East and are not verified against Snowflake's own rate card — Snowflake's pricing page renders prices client-side and the authoritative Snowflake Service Consumption Table did not resolve to a fetchable document. Treat these as order-of-magnitude.

Assumed cash infrastructure total: roughly $90 to $125 per month, call it $1,100 to $1,500 per year. Against $72,000 of gross bookings that is 1.5% to 2%. The infrastructure tax on a small Snowflake Marketplace listing is not the problem.

Net to the provider, by take rate (gross $72,000/yr, infra $1,300/yr):

Snowflake take rate Snowflake's cut Provider infra Provider net cash
0% (hypothetical floor) $0 $1,300 $70,700
10% $7,200 $1,300 $63,500
20% $14,400 $1,300 $56,300
30% $21,600 $1,300 $49,100

The spread between a 10% and a 30% take rate is $14,400 per year on this model, larger than every infrastructure line combined by a factor of ten. The one number Snowflake withholds is the one number that moves the answer most. Any plan built on this without logging in to read the fee schedule is built on a guess.

What the model leaves out, and why it dominates. None of the above prices the human. Curating and maintaining a dataset good enough to pass review and keep four enterprise consumers renewing is the real cost: source acquisition, daily pipeline babysitting, documentation that satisfies the "accurately describe the Products" policy, a support inbox on a 3-business-day clock, and a manual negotiation for every inbound Request. Add the accounts-receivable reality: the provider is seller of record, carries tax and collections risk, and is paid 30 days after Snowflake collects, on provider-set net terms. Assumption: 5 to 15 hours per month at steady state, more during the first year. That figure is a judgment call, not a sourced benchmark.

And the demand-side distribution is unknown. $100M of gross bookings across more than 1,700 transactions over five and a half months is a ~$59,000 mean transaction. Across the 2,679 public listings that can monetize (791 Paid plus 1,888 Free to Try) that is ~$37,300 per monetizable listing per half-year. Both figures are rough: the bookings window is Jan-Jun 2026 and also captures private listings, apps, and capacity-drawdown deals, while the listing counts are public listings measured on 2026-10-16. Different scopes, different dates. More importantly, no per-listing earnings distribution is published anywhere, and a mean cannot show a typical outcome. Financial alone holds 1,274 of 4,315 listings and the curated homepage carousel features institutional vendors such as LSEG and The Washington Post, which is suggestive of concentration but is not measurement. Do not read $37,300 or $59,000 as what a new small provider would earn.

Convergences and contradictions

Synthesis for RDCO

On the data-marketplace bet: the answer is no, and the reason is not the infrastructure cost. That cost turns out to be trivially small. A 200 GB dataset replicated to three extra regions runs on the order of $100 a month to maintain, roughly 2% of a plausible $72,000 of annual bookings. If the question had been "does Snowflake eat the margin on a small listing," the answer would be a clean no. The binding constraints are elsewhere: a take rate nobody outside a Snowsight login can see, a per-listing earnings distribution that is entirely unpublished, and a workload shape that is poorly matched to the hours actually available.

The workload shape is decisive. Snowflake Marketplace is not a storefront you can leave unattended. Self-serve paths exist, and Get and Buy do transact without the provider in the loop, but custom or negotiated deals all arrive as a Request that emails a company name and a human's address and expects a reply, policy sets a 3-business-day response clock on every inquiry, paid listings require contacting a Snowflake business-development partner before approval, and view telemetry is anonymous and two days stale so the only actionable signal arrives already attached to a person who wants an answer. That is a recurring, inbound, time-bounded obligation. For a founder whose main bet is a full-time phData role, it collides with the same calendar phData owns. This is not the teaching-and-consulting conflict from 2026-09-23; a dataset is a product, not a service. It is a calendar conflict, and the three-day clock does not care which.

One carve-out is worth naming rather than burying. A free listing has materially different physics: no transaction fee, no Stripe account, no business-development gate, no collections or tax exposure, and if private, not even the Provider and Consumer Terms. The only residual obligations are compute and the support-response duty. If Snowflake Marketplace presence were ever wanted as a credibility artifact rather than a revenue line, that is the version that costs almost nothing. This is explicitly not a recommendation; it is the only door this research leaves open, and whether it conflicts with the phData no-selling rule is an open question listed below.

On phData domain knowledge: this is the more valuable half of the brief, and it is immediately usable. A Snowflake client asking "should we publish our data on the Marketplace" is a conversation a Data Solutions Architect (DSA) or Technical Account Lead (TAL) will have, and most of the answer is counterintuitive. Five facts carry weight. First, auto-fulfillment bills the provider, not the consumer: cross-region costs land in a Snowflake-managed secure share area under the provider's organization, the opposite of manual replication, and it is the cost surprise that will land in a client's bill. Second, no premium edition is required to be a provider; a full non-trial account is the whole prerequisite, and only Virtual Private Snowflake is blocked from paid listings. Third, the review gates are fast for data shares (about 1 business day on metadata) and slow for Native Apps (up to 14 days, with a 3-day manual security review if the automated scan fails), which should shape any accelerator packaging decision alongside [[2026-09-24-cortex-knowledge-extensions-vs-agent-publish]]. Fourth, the telemetry-versus-events identity split is an easy misreading: marketplace analytics is not a web-analytics funnel, it is anonymous above the fold and a full contact record below it, with two days of lag and a warehouse bill to read it. Fifth, the 100-listing cap and 10 TB auto-fulfillment default are limits worth knowing before an architecture is drawn. Carry the CKE number alongside these: 17 Cortex Knowledge Extension listings on a 4,315-listing catalog means supply is thin and buyer demand is unmeasured, so anyone proposing a CKE-based accelerator as a distribution play should be shown that denominator first.

Why this is in the vault

Closes the "Understand Snowflake marketplace economics" line item on [[01-projects/data-marketplace/index]] with a documented negative. Separately, it is the reference a phData DSA or TAL needs when a Snowflake client asks who pays for cross-region listing replication, which editions can publish, how long listing review takes, and what a provider can actually see about buyers.

Open follow-ups

Related

Sources

Vault

Snowflake primary — legal and commercial

Snowflake primary — provider requirements

Snowflake primary — discovery and analytics

Snowflake blogs (dated, company-authored)

Live measurement

Secondary, labeled as such — not verified against primary filings