Actionable insights — Snowflake: AI Consumption Wins

1. Classify a software model as consumption or seat-based before judging AI exposure

1. Ask the first-order question: does this company get paid more when AI does more work (consumption / usage-based) or does it sell seats (per-user licenses) that AI could shrink?
2. For a consumption name, treat rising AI agent activity as a revenue tailwind; for a seat-based name, treat it as a structural threat to the unit being sold.
3. Re-rate the multiple accordingly — the market is now paying a premium for consumption models and a discount for seat-based ones.

• Management talking up "Agentic Work Units" or token counts without a clear consumption revenue line — a tell that AI activity isn't yet wired to the meter (CRM), versus a name where usage is the bill (SNOW).

2. Use "do agents need governed data here?" to locate the durable moat

1. Ask whether the proliferation of AI agents increases demand for the company's core service — governed data, secure workflows, identity, scalable compute.
2. Check whether the company is buying the missing piece (e.g. agent governance) rather than just describing the opportunity — an acquisition that lets it govern what agents do is more than a feature.
3. Note that agents which act (not just read) consume far more compute — so a governance layer and a consumption meter reinforce each other.

• An acquisition or product that governs agent actions (not just access) paired with rising consumption — the combination that turns the AI-agent trend into billable demand.

3. Read a cloud-provider commitment as a gross-margin lever, not just a hosting bill

1. When a software company signs a multi-year spend commitment with its cloud provider, check whether it buys cheaper, more efficient compute (custom silicon) that protects gross margin.
2. Tie the deal back to the reported product gross margin — a "handshake" that lowers cost of revenue is a margin defense, not just capacity.
3. Distinguish a strategic supply deal (locks in input cost) from a vanity capacity announcement.

• A multi-year cloud commitment framed around efficient/custom silicon and explicitly tied to gross-margin support — the signal it's a cost lever, not just a bigger bill.

4. Strip out acquisitions to find the real organic growth rate

1. When reported revenue growth includes a recent acquisition, back it out to find the organic rate — the true engine of the business.
2. Compare organic growth to the leading indicator (cRPO / contracted future revenue): if both are soft, the headline beat is hollow.
3. Watch for a coarser revenue disclosure introduced at the same time — less granularity often hides where the softness is.

• Reported growth propped up by an acquisition, a missed cRPO, and a newly vaguer segment disclosure arriving together — the trio that says "look past the headline beat."

5. Flag "value-abstraction" risk when a platform sits behind someone else's agent

1. When a company opens its data to external AI agents, ask who captures the value the agent creates — the data owner or the agent vendor.
2. If the company becomes the commodity data store behind a third-party agent with unclear pricing, treat that usage growth as a risk, not a win.
3. Wait for a pricing model that meters the value before crediting the integration to revenue.

• Surging third-party-agent usage of a platform's data with no disclosed pricing model — the setup where the platform does the work but the agent vendor captures the margin.