Snowflake: Funding the Next Layer of the Data Platform
Download case PDF (7 pages)Request teacher’s pack
Teaching case | MBA students and working professionals
Decision date: September 30, 2026
Information cutoff: September 30, 2026, 11:59 p.m. ET
A financing window, an operating question
On September 29, Snowflake priced $3.75 billion of zero-coupon convertible senior notes, split between notes due in 2029 and 2031. The company estimated net proceeds of $3.70 billion and expected the transaction to close on October 1, two days after the information cutoff for this case. Snowflake had entered into capped-call transactions at pricing; approximately $383.5 million was expected to be paid from the proceeds at closing. The company also announced a prospective use of about $548.3 million to repurchase $261.8 million principal of its 2027 notes. The balance would be available for general corporate purposes, potentially including additional debt repurchases, share repurchases, acquisitions, or strategic investments. [1]
For a strategy adviser preparing a hypothetical capital-allocation recommendation, this created a question larger than how to spend a financing windfall: how much should Snowflake commit to extending its platform while its AI economics and infrastructure needs remain difficult to isolate? At the cutoff, the offering had been priced but not closed. Neither the proceeds nor the announced repurchase had been received or completed. The company had disclosed no pending board vote or approved allocation of the residual.
The arithmetic offered a starting point, not a budget. Subtracting the planned capped-call and 2027-note repurchase amounts from estimated base net proceeds leaves approximately $2.768 billion. The estimate depends on the offering closing and the eventual transaction amounts; it excludes any effects from the note purchasers’ options and other adjustments. [1]
A consumption business with several moving parts
Snowflake sells a cloud data platform available across Amazon Web Services (AWS), Microsoft Azure, and Google Cloud. Customers pay as they consume services. In its FY2026 annual report, Snowflake said a substantial majority of its business ran on AWS and identified the major cloud providers as both suppliers and competitors. The filing did not quantify cloud cost by provider. [2]
For the quarter ended July 31, 2026, Snowflake reported product revenue of $1.492 billion, up 37% year over year, and total revenue of $1.547 billion, up 35%. Management attributed most of the product-revenue increase to greater consumption by existing customers. Net revenue retention was 126%. The company reported $9.0 billion of remaining performance obligations (RPO), up 30%. RPO included deferred revenue and certain non-cancelable contracts, but excluded on-demand consumption and some time-and-materials arrangements; it did not specify when revenue would be recognized or at what margin. [3]
The customer base showed both breadth and concentration by cohort. Snowflake reported 828 customers with more than $1 million in trailing-twelve-month product revenue, 174 more than a year earlier. This group generated about 68% of product revenue. Separately, the company reported no individual customer at 10% or more of revenue or accounts receivable. These measures use different thresholds and denominators. [3]
Product gross margin was 71% for the quarter, compared with 72% a year earlier. Third-party cloud infrastructure costs, including AI inference and GPUs, increased $113.6 million in the quarter and $199.7 million in the first half compared with the prior year. Management said the increase was mainly driven by higher platform consumption. It also cited new product features that had not yet reached scale, workload mix and timing, pricing and discounts, and product investment as factors affecting margin. The filing did not break out AI revenue, AI-specific margin, or the cost to serve an AI workload. [3]
Snowflake’s reported liquidity was approximately $4.3 billion in cash, cash equivalents, and short- and long-term investments as of July 31, before the September notes. Its second-quarter non-GAAP free cash flow was $83.8 million: $91.4 million of cash from operations less $7.6 million of property and equipment purchases and capitalized software costs under the company’s definition. This was one quarter’s result, not a forecast of annual cash generation. [3]
Commitments and capacity choices
Snowflake’s July filing described a five-year cloud-infrastructure arrangement with a $6.0 billion cumulative minimum through March 2031 and annual minimums ranging from $0.9 billion to $1.25 billion. Shortfall payments applied if minimums were not met. A June amendment added an estimated $58 million minimum for incremental storage through May 2036, subject to actual use. The filing did not name the infrastructure counterparty. A May Snowflake announcement described a matching five-year, $6 billion AWS commitment; the two disclosures appear to refer to the same arrangement and should not be added together. [3, 4]
The filing also disclosed a separate, unnamed AI-service-provider agreement with a $390 million minimum over three years through November 2028, of which $270 million remained at July 31. Some qualifying services accessed through public clouds could count toward the agreement. The public disclosures did not establish all possible overlap between commitments. These were minimum purchase obligations with specified shortfall remedies, not amounts all payable immediately. [3]
Snowflake’s own engineers had studied the operational side of such commitments. A 2025 paper analyzed three years of normalized resource demand across multiple cloud providers. One example resource pool reached 3.9 times its initial demand by the end of the trace, while aggregate weekly demand declined in 37% of weeks. The authors described balancing the cost of unused committed capacity against more expensive on-demand capacity; workload timing and commitment expiration schedules could affect that balance. The study covered Snowflake’s supplier-side infrastructure operations. It did not measure customer willingness to pay, AI-specific contribution margins, or the return on a new product investment. [5]
Competitors move into adjacent territory
AWS and Google were adding analytics and AI features to their own platforms. AWS announced AI-driven scaling for Redshift Serverless workgroups, while Google described BigQuery AI functions, graph tools, and a cross-cloud lakehouse preview. Snowflake announced dynamic model routing and cost-management features for its AI products. These announcements showed strategic activity, but each was a vendor’s own description; they did not provide comparable independent performance or cost benchmarks. [6]
The competitive question was not simply whether to spend more on AI. Snowflake could invest in platform capabilities and complements that encourage customers to bring additional workloads. It could also preserve the ability to serve those workloads across clouds and manage the cost of underlying infrastructure. At the same time, scale in consumption did not by itself establish attractive incremental economics: revenue, cloud usage, customer mix, discounts, and software efficiency could move differently.
The recommendation
The adviser’s task was to recommend how Snowflake should prioritize the estimated residual proceeds if the September offering closed on the announced terms. The recommendation could describe a staged allocation, strategic priorities, or a decision rule tied to evidence that management could monitor. It should address the company’s existing commitments and financing obligations, explain how the proposal would preserve or spend strategic flexibility, and state what evidence would lead to accelerating, pausing, or redirecting investment.
Plausible uses included:
- Accelerating platform, AI, and ecosystem investment to build capabilities and compete for workloads.
- Staging investment while retaining liquidity and requiring evidence of paid usage, customer retention, or improving unit economics.
- Applying more capital to debt retirement, share repurchases, or cash preservation.
- Keeping capacity for selected acquisitions or strategic investments that complement the platform.
No public source at the cutoff supplied AI-specific contribution margins, customer-level AI returns, provider-specific infrastructure spending, project return estimates, or an approved allocation of the expected residual. A recommendation therefore needed to distinguish what was known from what remained uncertain.
Discussion questions
- What strategic problem should the adviser treat as the central capital-allocation question: speed, workload economics, infrastructure flexibility, financial risk, or a combination?
- Which uses of the expected residual are most defensible given Snowflake’s growth, margin, provider commitments, and competitive position? What should be staged, and what should be funded immediately?
- What indicators and decision thresholds would make an investment tranche earn the next one? Which measures could be misleading if used alone?
- How should the adviser account for the 0% notes, announced 2027-note repurchase, potential dilution, and existing liquidity?
- What new information would most change the recommendation? Identify the limits of the public evidence rather than filling gaps with unsupported assumptions.
Exhibits
Exhibit 1. Selected operating and financial evidence
| Measure | Reported value | Period / comparison |
|---|---|---|
| Product revenue | $1,491.9m; +37% | Q2 FY2027, year over year |
| Total revenue | $1,546.8m; +35% | Q2 FY2027, year over year |
| Net revenue retention | 126% | Q2 FY2027; company-defined |
| Remaining performance obligations | ~$9.0bn; +30% | July 31, 2026 |
| Customers above $1m trailing-twelve-month product revenue | 828; group generated ~68% of product revenue | July 31, 2026 |
| Product gross margin | 71% vs. 72% | Q2 FY2027 vs. prior-year quarter |
| Increase in cloud infrastructure costs | $113.6m Q2; $199.7m first half | Year over year; includes AI inference and GPUs |
| Cash, cash equivalents, and short- and long-term investments | ~$4.3bn | July 31, 2026; before September notes |
| Non-GAAP free cash flow | $83.8m | Q2 FY2027 |
All figures are company-reported and rounded here where appropriate. RPO, retention, the customer cohort, and non-GAAP free cash flow have issuer-specific definitions. None isolates AI economics. [3]
Exhibit 2. September 29 financing, expected base case
| Item | Amount | Status at cutoff |
|---|---|---|
| New 0% convertible notes | $3.75bn principal | Priced; expected close October 1 |
| Estimated net proceeds | $3.70bn | Estimate; not received |
| Planned capped-call cost | ~$383.5m | Capped calls entered into at pricing; cost expected to be paid at closing |
| Planned 2027-note repurchase | ~$548.3m cash for ~$261.8m principal | Announced concurrent transaction; not completed |
| Estimated residual for other general corporate purposes | ~$2.7682bn | Derived estimate; conditional, not an approved budget |
Calculation: $3.700bn − $0.3835bn − $0.5483bn = $2.7682bn. Estimated net proceeds could be higher if purchasers exercised their options; final amounts and uses could differ. The new notes had 0% coupons, but remained obligations with future repayment or conversion and potential dilution. [1]
Exhibit 3. Disclosed purchase commitments
| Agreement | Disclosed minimum | Timing and limitations |
|---|---|---|
| Cloud infrastructure | $6.0bn cumulative | Five years through March 2031; annual minimums $0.9bn–$1.25bn; shortfall payments apply |
| Incremental storage amendment | Estimated $58m aggregate minimum | Through May 2036; subject to actual-use adjustment |
| AI service provider | $390m across term; $270m remained at July 31 | Three years through November 2028; provider unnamed; some qualifying cloud-accessed services may count |
These obligations are not all due immediately. The 10-Q does not name counterparties or disclose all cross-crediting mechanics. The matching AWS announcement and cloud minimum should not be counted twice. [3, 4]
Exhibit 4. Competitive announcements through the cutoff
| Company | Announced capability | Evidence limit |
|---|---|---|
| AWS | AI-driven scaling for Redshift Serverless new workgroups | Vendor announcement; no neutral comparison |
| Google Cloud | BigQuery AI functions, graph capabilities, and cross-cloud lakehouse preview | Vendor announcements; preview scope and no comparable economics |
| Snowflake | Dynamic model routing and AI cost-management capabilities | Company announcement; not independent evidence of realized savings |
[6]
Source notes
- Snowflake, September 29, 2026 convertible-note pricing announcement, terms and anticipated closing (paras. 2–4), capped calls entered into at pricing and expected payment from proceeds at closing (paras. 4–5), general-purpose uses (paras. 7–8), and prospective concurrent repurchase (paras. 9–10). The capped calls were entered into at pricing, but their cost and the note proceeds were expected to settle at closing; the debt repurchase remained prospective at the cutoff.
- Snowflake, FY2026 Form 10-K, cloud deployment and competition disclosures (HTML 628–30; printed p. 104 for customer concentration).
- Snowflake, Q2 FY2027 Form 10-Q, revenue, retention, and customer cohort (printed p. 56); current no-10%-customer revenue/receivables disclosure (printed p. 17; HTML 793–95); RPO definition, exclusions, and consumption/timing caveats (printed p. 17; HTML 802–11; MD&A pp. 49–50; HTML 1963–67); margin and infrastructure cost (pp. 57–58); RPO and commitments (pp. 60–61); liquidity (p. 66); free cash flow definition and reconciliation (p. 49). Filed September 4, 2026.
- Snowflake, AWS collaboration announcement, May 27, 2026. Issuer announcement; not contract text.
- Murray Stokely, Neel Nadgir, Jack Peele & Orestis Kostakis, “Shaved Ice: Optimal Compute Resource Commitments for Dynamic Multi-Cloud Workloads”, ICPE 2025, v2, §§2.2–3.3 and 6. All authors were Snowflake employees; normalized supplier-side resource evidence, not customer AI economics.
- Official vendor announcements: AWS Redshift Serverless scaling (April 27, 2026); Google BigQuery AI functions (January 26, 2026); Google cross-cloud and agentic capabilities (April 22, 2026); Google BigQuery Graph (September 1, 2026); Snowflake, dynamic model routing (August 18, 2026).
This case is based on public information available by the stated cutoff. The allocation decision is hypothetical. No subsequent transaction outcome is presented.