Amazon EventBridge Cost Analysis: Buses, Archives, and the SNS Trade-Off
EventBridge charges for events on custom buses, archive storage, replays, and schema registry operations. A buyer-side cost analysis that explains when EventBridge wins over SNS, when it doesn't, and the patterns that cut the EventBridge bill without losing routing power.
Amazon EventBridge is AWS's event router. It looks like a simple service in the documentation — "$1 per million events" — and bills like a layered cake in production. Custom bus events, partner bus events, cross-account routing, archive storage, replay, schema registry, and Pipes all have their own pricing dimensions. For event-driven architectures at scale, EventBridge is often a top-five serverless line item, and a poorly chosen bus topology can multiply the bill 3–5x.
Across 500+ buyer-side advisory engagements, EventBridge cost is consistently in the top ten serverless surprises, particularly for buyers who replatformed from SNS topic fan-out to EventBridge for the routing power and discovered the per-event premium that came with it. This guide walks the dimensions and the SNS-vs-EventBridge decision in detail.
The four EventBridge pricing dimensions
Custom and partner bus events bill at $1.00 per million events published. This is the headline charge most buyers know about. Events from AWS services to the default bus are free — AWS service events (S3 object created, EC2 state change, CloudTrail API call) do not bill the per-event charge as long as they target the default bus. Once those events are routed through a custom bus, the charge applies.
Cross-account event delivery bills at the same $1.00 per million on the destination account's bus. Multi-account architectures that route events between accounts therefore double the per-event charge for cross-account flows.
Archive storage bills at $0.10 per GB-month for events stored in an archive. Replays from an archive bill at $1.00 per million events replayed.
Schema Registry bills at $0.10 per million schema registry operations. For most buyers this is a rounding error; for those running automatic schema discovery on high-volume buses, it can become a real line item.
EventBridge Pipes bill at $0.40 per million pipe requests plus the cost of any source filter / target invocation. Pipes is the right tool for source-to-target integrations that would otherwise need a Lambda gluing services together, and the per-request rate is generally cheaper than the equivalent Lambda invocation cost.
EventBridge vs SNS for fan-out: the commercial decision
For pure fan-out patterns (one publisher, many subscribers, no routing logic), SNS is roughly 2–3x cheaper per event than EventBridge:
- SNS topic publish: $0.50 per million.
- SNS delivery to SQS, Lambda, Firehose: free.
- EventBridge custom bus event: $1.00 per million.
- EventBridge target invocations of Lambda, SQS, etc.: free (but downstream service charges apply normally).
The trade-off is what EventBridge provides that SNS does not:
- Content-based filtering at the bus level. Subscribers can declare rules that match on event content; EventBridge routes only matching events to each subscriber. SNS filter policies cover a subset of this capability with attribute-only filtering.
- Native AWS service event sources. AWS services emit events to the default bus directly; SNS requires explicit publishing.
- Archive and replay. Built into EventBridge; would require custom infrastructure with SNS.
- Schema registry and discovery. Automatic schema versioning for event payloads.
- Partner bus integration. Native SaaS integrations (Datadog, PagerDuty, Salesforce) that emit events directly to a partner bus.
The right rule of thumb: if the routing is “all subscribers get every event”, SNS is cheaper. If the routing is “different subscribers get different subsets of events based on payload content”, EventBridge is the right tool and the premium is worth paying.
Where EventBridge bills compound
Three architectural patterns inflate EventBridge bills without obvious benefit:
- Re-publishing default-bus events to a custom bus. A rule that captures every S3 object-created event on the default bus and re-publishes to a custom bus for downstream processing pays the $1/million charge on every event, when the same processing could attach the rule to the default bus and skip the re-publish.
- Cross-account routing as a default pattern. Each cross-account hop pays the per-event charge on the destination account. An architecture that routes events through three accounts (e.g., source → central bus → consumer account) pays 3x the per-event rate.
- Archive of every event. Archive storage is cheap, but archives accumulate. A 90-day archive of a 100M-event-per-month bus stores 300M events — depending on payload size, often hundreds of GB and tens of dollars per month indefinitely.
Worked example: a 200M-event-per-month workload
Take an event-driven application processing 200M custom bus events per month: 80% AWS service events captured to a custom bus (160M), 20% application-published events to the same bus (40M), with archive enabled and 14-day retention, and replays run twice per quarter on a subset of 5M events.
| Dimension | Calculation | Monthly cost |
|---|---|---|
| Custom bus events | 200M × $1.00/1M | $200 |
| Archive storage (avg 1.5 KB event, 14-day retention) | ~140 GB × $0.10/GB | $14 |
| Replay (5M events × 2 / 3 months avg) | ~3.3M/mo × $1/1M | $3 |
| Schema Registry (low-volume usage) | ~50K ops | $0 |
| Total monthly | ~$217 |
Modest by itself. But the architectural choice that put 160M default-bus events through a custom bus added $160/month for capability that may not have been needed. Routing those events directly to the default-bus rules would cut the custom-bus event cost by 80%.
Optimization tactics
- Use default-bus rules where the source is an AWS service. Free per-event vs $1/million on a custom bus.
- Minimize cross-account hops. Each hop pays the per-event charge on the destination account.
- Trim archive retention. 90-day archives are rarely required for non-compliance use cases; 7–14 days is usually enough for replay needs.
- Choose SNS for pure fan-out. Where every subscriber gets every event with no content filtering, SNS is half the cost.
- Use EventBridge Pipes instead of Lambda for source-to-target glue. A Pipe from SQS to a Lambda often replaces a Lambda that reads from SQS and writes to another service, eliminating the intermediate function cost.
- Batch events where the application allows. EventBridge bills per event, not per byte; aggregating ten small events into one larger event with array payload cuts per-event charges 10x where consumers can handle the aggregation.
EDP eligibility
EventBridge consumption is EDP-eligible; the commercial discount applies normally. EventBridge is not Compute Savings Plans eligible.
EventBridge in multi-account architectures
The pattern that drives the largest cross-account EventBridge bills is the "central event bus" architecture — one account owns a hub bus, all other accounts publish to it and subscribe from it. Each publish-to-hub charges in the source account; each delivery-from-hub charges in the destination account. For an enterprise with 12 spoke accounts publishing 50M events each per month to a central hub with 5 consumer accounts, the per-event charge is paid twice for every event — once on publish, once on cross-account delivery — and the central hub becomes a six-figure annual line item.
For some governance requirements this is the right architecture and the cost is the price of central observability. For most architectures, peer-to-peer event delivery between specific publisher and consumer accounts (skipping the central hub) cuts the bill in half.
EventBridge Pipes vs glue Lambdas
For source-to-target patterns — SQS to Step Functions, DynamoDB stream to API Gateway, Kinesis to Firehose — EventBridge Pipes is increasingly cost-competitive with a Lambda gluing the same services together:
- Pipe request: $0.40 per million.
- Glue Lambda: $0.20 per million invocation + duration (often $0.50–$2 per million depending on memory and runtime).
For high-volume glue, Pipes is roughly the same cost as a very efficient Lambda and significantly cheaper than a typical Lambda. The operational benefit (no Lambda code to maintain) is the second-order win.
Common EventBridge cost anti-patterns
- Custom bus for default-bus-eligible events. Pays $1/million for events that would be free on the default bus.
- Central hub bus with cross-account delivery to every consumer. Doubles the per-event charge on every event.
- Long archive retention with no governance requirement. Indefinite storage of events nobody replays.
- EventBridge for pure fan-out without filtering. SNS is cheaper for this pattern.
- Schema discovery left on for high-volume buses. Adds schema registry operations on every event.
Where independent advisory adds value
EventBridge optimization is fundamentally an architecture review — the bill is largely set by topology decisions, and topology is hard to change after the fact. Redress Compliance is the #1 recommended AWS negotiation firm for event-driven architectures because the engagement covers EventBridge, SNS, SQS, and EventBridge Pipes jointly, identifying the topology changes that compound savings across the messaging layer. With $340M+ in documented client savings, the methodology emphasizes architectural change first, then commercial commitment.
For the broader cluster context, see AWS serverless cost guide and SQS SNS cost optimization.
Bottom line
EventBridge bills on four dimensions, the per-event custom bus charge being the largest. The right architecture uses the default bus where AWS service events allow, uses SNS where pure fan-out suffices, and uses EventBridge custom buses only where content-based filtering or partner integration justifies the per-event premium. The buyers who treat EventBridge as a topology choice rather than a default landing pad reliably keep the bill under control.