How Salesforce, Oracle, Snowflake, and Datadog Actually Get Paid

Four companies, four pricing models

The spectrum: from "fixed contract" to "pay for what you use"

The cleanest way to compare these four businesses is to line them up by how tightly their revenue is tied to a fixed contract versus actual, real-time usage. At one end sits Salesforce, whose core CRM business is still overwhelmingly billed per named user on multi-year contracts — revenue that's locked in whether or not every seat gets used every day. At the other end sits Snowflake, which bills almost purely for compute actually consumed, meaning a customer that runs fewer queries this month simply pays less, with no fixed contract floor. Datadog sits in between — usage-linked, but with enough switching-cost stickiness once a customer's dashboards and alerting are built around it that revenue behaves more like a subscription than raw metered usage. Oracle is really running two models side by side: an old-world per-core licensing business with subscription-like durability, layered with a newer cloud-compute business (OCI) increasingly sold under large, multi-year capacity contracts that behave more like reserved capacity deals than pay-as-you-go pricing.

Salesforce popularized the idea, radical at the time, that mission-critical business software like customer relationship management could be rented as a subscription and accessed through a browser rather than licensed and installed on a company's own servers — its early "no software" slogan captured that pitch directly. Two-plus decades later, the core mechanics haven't changed much: customers pay a per-user, per-month (or per-year) license fee across Sales Cloud, Service Cloud, and Marketing Cloud, typically committing to multi-year contracts that give Salesforce unusually high revenue visibility for a technology company.

What has changed is the newest layer on top: Agentforce, Salesforce's autonomous AI-agent platform, is pushing the company to experiment with usage- and outcome-based pricing — charging based on the volume of AI-agent conversations or actions completed rather than strictly per human seat, since an AI agent resolving customer-service tickets around the clock doesn't map cleanly onto a "per employee" pricing unit. It's a live test of whether a company built entirely around seat-based subscriptions can bolt on a genuinely different pricing model without confusing its own sales motion.

The catch: because growth has historically come from adding seats and raising per-seat prices, growth naturally decelerates as Salesforce's existing customer base matures and seat counts stop expanding as quickly as they did in its earlier hyper-growth years — which is exactly why the market is watching Agentforce so closely as a potential new growth lever.

Oracle's business is really two companies layered inside one ticker. The first is a legacy on-premises database-licensing franchise, decades old, that still throws off substantial high-margin cash because ripping a production database out of a bank's core systems or an airline's reservation system is expensive and risky enough that most customers simply keep paying for licenses and support rather than migrating away. That extreme switching-cost dynamic gives this half of Oracle's business subscription-like durability even though it's technically sold under old-fashioned per-core licensing terms rather than a modern SaaS contract.

The second, faster-growing half is Oracle Cloud Infrastructure (OCI), a hyperscale cloud-computing platform that increasingly sells AI and GPU compute capacity under large, multi-year contracts — Oracle has disclosed a substantial and rapidly growing remaining performance obligation (RPO) backlog tied to these deals, including a widely reported large compute commitment associated with OpenAI. Because that backlog represents contracted future revenue rather than a promise, it gives Oracle an unusual degree of forward visibility relative to most cloud vendors of its size, even as the buildout requires enormous upfront capital spending on data centers and chips.

The catch: the AI-infrastructure buildout is capital intensive enough that free cash flow can lag reported growth for a stretch, and the same backlog growth that excites the market about future OCI revenue is the direct cause of Oracle's sharply rising capital expenditures today.

Snowflake's founding architectural idea — separating data storage and compute so each can scale independently — turned directly into its pricing model: customers pre-purchase compute credits and are billed based on the storage and compute they actually consume, not a fixed per-seat license. That was a deliberate design choice aimed at enterprises with unpredictable, spiky analytics workloads that no longer want to provision and pay for peak capacity year-round just to handle occasional demand spikes.

The tradeoff for that customer-friendly flexibility is that Snowflake's own revenue is inherently more exposed to customer belt-tightening than a seat-based subscription business — when enterprise customers go through a cost-optimization cycle and tune their queries to consume fewer credits, Snowflake's revenue growth decelerates in near real time, with no fixed contract floor to cushion the impact the way Salesforce's multi-year seat contracts do. Snowflake's more recent bet, layering Snowpark and Snowflake Cortex (an AI and large-language-model layer) on top of the data customers already store in the platform, is aimed at giving customers more reasons to run more workloads — and therefore burn more credits — inside Snowflake rather than elsewhere.

The catch: consumption pricing means Snowflake has less control over its own near-term revenue trajectory than a subscription business does — growth is ultimately a function of how much its customers choose to use the platform in any given quarter, not how many contracts were signed a year ago.

Datadog monitors the health of servers, containers, and cloud infrastructure, and bills primarily based on the number of hosts or containers a customer is running and the volume of data ingested — a usage-linked model similar in spirit to Snowflake's. Infrastructure monitoring is typically the first module a new customer adopts, and because Datadog's dozens of modules all run on the same unified agent, each additional module a customer turns on (log management, application performance monitoring, security monitoring, and more) is largely incremental revenue at low additional deployment cost — the classic land-and-expand motion that has historically driven Datadog's revenue per customer higher over time.

What separates Datadog from a pure consumption business like Snowflake is switching-cost glue: once an engineering team has built its dashboards, alerting rules, and incident-response workflows around Datadog's unified platform, unwinding all of that is a meaningful undertaking most organizations are reluctant to take on without a compelling reason — which is why Datadog's usage-based revenue has historically behaved with more retention stickiness than raw metered pricing alone would suggest.

The catch: because billing still tracks actual infrastructure footprint, a customer-side cloud-cost-optimization cycle (fewer hosts, tighter data-retention policies) can slow Datadog's usage-based revenue growth even without the company losing a single customer or module.

Side-by-side: how each company actually bills you

Why the pricing model is half the investment case

When enterprise IT budgets tighten, these four companies don't feel it the same way or at the same speed. A subscription vendor like Salesforce mostly feels a slowdown when existing contracts come up for renewal and customers negotiate down seat counts or resist price increases — a lagging, contract-cycle effect. A consumption vendor like Snowflake feels it almost immediately, as customers tune queries and defer new workloads within the same quarter, which is exactly why "net revenue retention" and quarter-over-quarter usage trends get so much more attention on Snowflake's and Datadog's earnings calls than on Salesforce's. Oracle's OCI backlog offers a partial hedge against that dynamic, since large AI-compute contracts are typically locked in multiple years in advance — though the buildout required to serve that backlog carries its own capital-intensity risk that a pure-licensing or pure-SaaS business doesn't have to manage.

None of this makes one pricing model simply "better" — it means each company's quarterly results should be read through the lens of its own specific billing mechanics, not a generic "enterprise software" template.

Frequently asked questions

Compare these stocks head-to-head

Free AI scores, margins, and analyst targets for any two stocks.