How Companies Like AWS, Snowflake and Twilio Run Usage-Based Pricing: What They Meter

Introduction

Usage-based pricing looks simple from the outside: pick a unit, count it, charge for it. The companies that run it at scale show how much sits behind that sentence. The hard decisions are the billable metric (what to count), the pricing shape (tiers, credits, commitments) and the metering infrastructure (how to count it exactly once).

This article walks through 17 well-known companies and what each one charges for. It is not a claim that any of them “switched” billing platforms to save money; most of their billing stacks are proprietary and their internal architecture is not public. The value is in the pattern: the metric each company chose, and what that choice asks of the billing system underneath.

It then covers the economics that push some teams to own their billing layer, and the options available now. Where we cite a number, we link to its source. Where a figure could not be verified, we left it out.

What Is Usage-Based Billing Software?

Usage-based billing software is infrastructure that meters customer consumption of a product and generates invoices from actual usage, not a fixed monthly fee. It is also called consumption-based billing or metered billing. A complete system typically includes:

• Event ingestion and metering: capturing usage events from application code or infrastructure, exactly once
• Pricing engine: applying rate cards, tiers and volume discounts to aggregated usage
• Invoice generation: producing invoices that finance and customers can verify
• Customer visibility: a way for customers to see their own usage and spend
• Audit and reporting: tracing every invoice line back to raw events

The distinction from a payment processor is scope. A processor collects payment. Billing software decides what to charge.

ABAXUS: open-core usage billing you can run yourself

A free, modular, self-hosted metering and billing engine with idempotent ingestion and versioned pricing. Use ABAXUS Cloud if you would rather we run it.

See how it works

Why the Billing Layer Becomes a Decision

Fees that scale with revenue

Some billing products charge a percentage of billing volume. Stripe Billing’s pay-as-you-go rate is listed at 0.7% of billing volume, with monthly-plan options at higher volume. At $5M a month in invoices, 0.7% is $35,000 a month, or $420,000 a year, before payment processing fees. At $1M ARR a percentage fee is small; at larger volumes it becomes a budget line you can compare against the cost of running your own.

Data control

A hosted billing platform holds your usage events and billing history. That is fine for many teams. For others, such as regulated industries or companies with customer contracts about data handling, it is a boundary they need to think about. Self-hosted billing keeps usage data inside your own environment.

Vendor ownership is no longer stable

Between September 2025 and July 2026 the independent metering layer was acquired: OpenMeter by Kong, Metronome by Stripe, and Orb by Adyen (announced June 2026). The vendor you pick today may change owners, so the data contract, meaning your event schema, idempotency keys and ledger, is worth owning regardless of whose software runs.

The customization ceiling

Off-the-shelf platforms handle per-seat and simple tiered pricing well. Multi-dimensional rates, credits, hierarchical accounts and enterprise commitments push teams toward more flexible engines or custom builds.

Photo by Jakub Żerdzicki on Unsplash

17 Companies and What They Meter

Cloud infrastructure and developer platforms

Amazon Web Services prices dozens of services on usage dimensions: compute time, storage per GB-month, data transfer and API requests. A single customer bill aggregates many independent meters, which is why AWS-style billing is the reference case for multi-dimensional metering.

Snowflake abstracts compute into credits, a proprietary unit that decouples price from underlying hardware. Customers budget in credits; Snowflake keeps freedom to change what a credit buys. See its annual reports for its own financial disclosures.

DigitalOcean bills compute and related resources by usage with a predictable cap, so a customer can always work out the worst-case bill. Legibility is the product feature.

Heroku made the dyno-hour its billing unit: a standard compute container billed by time used. A familiar unit made cost easy to reason about.

Netlify has historically metered bandwidth and build minutes, with free allowances that let developers adopt the product before paying.

Vercel meters function execution, edge requests and bandwidth, with a free tier for individuals and usage-based charges as workloads grow. In September 2025 it announced a $300 million Series F at a $9.3 billion valuation.

Twilio charges per message, minute or API call, so each SMS or voice call is a discrete billing event. High event volume plus carrier cost pass-through is the classic metering workload.

Algolia bills on search requests, a metric that tracks directly with the value customers get from search.

Data and observability platforms

Databricks prices compute in Databricks Units (DBUs), a normalized measure of processing power billed per second, with the rate varying by workload type, tier and cloud. The cloud provider bills infrastructure separately.

Elastic prices Elastic Cloud on deployment size and data volume, a shift from a self-managed product to a consumption-linked managed service.

New Relic bills on data ingested and number of billable users, according to its pricing documentation, replacing the older product-based model.

Datadog meters hosts, custom metrics, log volume and APM traces. These dimensions scale independently for each customer, which makes account-level aggregation and volume discounts a real engineering problem. Datadog’s filings note that it recognizes revenue for usage-based subscriptions as the product is consumed, which can cause results to fluctuate.

MongoDB Atlas charges on cluster tier, storage and data transfer. Atlas made up 73% of MongoDB’s total revenue in fiscal 2026, up from 70% the year before.

Communications, data pipelines and commerce

Stripe charges per transaction, 2.9% plus 30 cents on standard US card payments, a model that is simple to state and complex to operate across disputes, refunds and currencies.

SendGrid built email-volume tiers with overage pricing. Twilio acquired it in 2019 in a deal reported at around $2 billion.

Segment priced on Monthly Tracked Users (MTUs), a metric aligned with customer value that required counting unique users across event streams rather than counting raw events. Twilio acquired Segment in 2020 for $3.2 billion.

Shopify charges a variable platform fee on merchant sales volume for its Plus plans, which requires tracking transaction volume, applying thresholds and calculating fees per merchant every month.

What These Companies Have in Common

1. The metric tracks customer value. Searches, tracked users, compute-seconds and ingested gigabytes all grow as the customer gets more out of the product. A metric that does not move with value creates arguments.

2. The unit is legible. Credits, dyno-hours, DBUs and MTUs are all ways of giving customers a stable unit to budget against while the vendor keeps flexibility underneath.

3. Billing depends on exact counting. Every example above is only as trustworthy as its meter. Duplicate events, late events and clock skew turn into disputed invoices, which is why idempotency keys, versioned meters and an audit trail matter more than any pricing feature.

When to Own the Billing Layer

Not every team should build or self-host billing infrastructure. These are rules of thumb from our own experience, not measured thresholds:

Volume. Below roughly $2M ARR, a hosted platform is usually the right choice: the engineering time costs more than the fee. As billing volume grows, compare the fee line against the cost of an engineer owning the deployment.

Pricing complexity. If your pricing fits per-seat or simple tiers, standard platforms handle it. Multi-dimensional usage, real-time budget caps, hierarchical accounts and custom commitments reach the platform ceiling faster than expected.

Data control. If usage events carry regulated data or your customer contracts restrict where billing data may live, a self-hosted engine keeps it inside your perimeter.

Vendor independence. If the acquisitions above made you uneasy, prefer an open engine whose schema you own.

Running into billing infrastructure limits?

The ABAXUS core is free and self-hosted, with storage and messaging behind adapters. ABAXUS Cloud is the same engine hosted by us. Cloud starts at $12K/yr; early adopters get 65% off.

See Pricing

Conclusion

The common thread across these companies is not that they avoided billing vendors. It is that they treated the billable metric and the meter as product decisions. If you are at that point, the options have historically been to build it yourself or accept a vendor’s constraints.

ABAXUS offers a third path: an open-core engine you can self-host in your own cluster, with idempotent ingestion, versioned meters and plans, and SHA-256 traceable invoice lines, plus ABAXUS Cloud if you would rather not run it. It is in beta with limited seats, so weigh that honestly against longer-established options.

If your billing architecture is constraining your pricing or your unit economics, book an architecture review to map your event sources and data contract.