Temporal Cloud vs Self-Hosted 2026: True Cost and When Each Wins

Overview

Temporal is a durable execution platform that runs long-running, fault-tolerant workflows as code. Founded in 2019 by Maxim Fateev and Samar Abbas (the original authors of Uber Cadence), Temporal is offered as both a managed service (Temporal Cloud) and as an open-source server (Apache 2.0 license) that organizations can self-host. The decision between the two is one of the most common questions in workflow-engine adoption, and the right answer depends on workload scale, operational maturity, compliance requirements, and total cost of ownership.

As of May 2026, Temporal Cloud runs in multiple AWS and Google Cloud regions (US East/West, EU West/Central, AP Southeast/Northeast) with namespace-level isolation. The self-hosted Temporal Server is the same codebase, deployed as a set of services (frontend, history, matching, worker) backed by Cassandra, MySQL, or PostgreSQL.

Temporal Cloud

Temporal Cloud is a fully managed multi-tenant service. Customers create namespaces and connect workers (their workflow code) over mTLS. The Temporal team operates the persistence layer (Cassandra), the history shards, the matching service, and observability. SLA is 99.9% on Standard and 99.95% on the Mission Critical tier.

Pricing as of May 2026 is consumption-based with three components: Actions (workflow events) billed per million ($25/M Actions on the Standard tier), Active Storage (GB-month of running workflow state), and Retained Storage (GB-month of completed workflow history retained for replay). A typical small-medium production deployment lands in the $200-$2,000/month range; large deployments can exceed $50,000/month.

The Mission Critical tier (introduced 2024) adds higher SLA, dedicated support, and priority namespace placement at a roughly 2x consumption multiplier.

Self-Hosted Temporal

Self-hosted Temporal is the same open-source server, deployed on Kubernetes (recommended via the helm-charts repository) or on VMs. The components are stateless except for the database. Cassandra is the recommended persistence layer at scale; PostgreSQL works for smaller deployments. Operators run Temporal frontend, history, matching, and worker services, plus the optional UI and Elasticsearch for advanced visibility.

License cost is zero (Apache 2.0). The cost is entirely infrastructure and operational labor.

True Cost of Self-Hosting

A frequent mistake is treating self-hosted Temporal as "free." A realistic small-production self-host costs include:

• Cassandra cluster: 3-node minimum for production. AWS r6i.2xlarge x3 plus EBS storage runs roughly $1,200-$1,800/month before storage and backups.
• Temporal services: 4-6 EC2/EKS pods across frontend, history, matching, and workers. Roughly $300-$600/month.
• Elasticsearch (for advanced visibility): 3-node cluster, roughly $500-$900/month, or AWS OpenSearch managed.
• Backups, snapshots, monitoring (Prometheus/Grafana), logging (CloudWatch/Loki): $200-$500/month.
• Engineering operational time: realistically 0.25-1.0 FTE depending on how much custom tooling, on-call rotation, and incident response is needed.

A small production self-host with reasonable reliability typically lands in the $2,500-$4,500/month infrastructure range plus operational labor. At higher action volume, this can be cheaper than Cloud; at low-to-moderate volume, Cloud is almost always cheaper once labor is included.

Latency and SLA

Temporal Cloud reports p95 schedule-to-start latency under 50ms for tasks within the same region as the worker. Self-hosted Temporal can match or beat this if the Cassandra cluster is provisioned correctly, but latency degrades quickly under unbalanced shard distribution or undersized history nodes.

Cloud SLA is 99.9% (Standard) or 99.95% (Mission Critical), with credits issued for SLA breaches. Self-hosted SLA is whatever the organization can engineer; achieving 99.9% requires multi-AZ Cassandra, automated failover, and tested disaster recovery runbooks.

When to Choose Temporal Cloud

• Action volume below roughly 50M-100M actions/month
• The team does not have Cassandra operational experience
• Compliance requirements are met by Temporal Cloud (SOC 2 Type II, HIPAA available)
• Engineering capacity is better spent on workflow logic than on operating durable execution infrastructure

When to Choose Self-Hosted

• Compliance/data residency rules out multi-tenant cloud (defense, certain regulated industries, sovereign clouds)
• Action volume is high enough that Cloud consumption pricing exceeds infrastructure cost by a significant margin
• The team already operates Cassandra and Kubernetes at scale and the marginal operational cost is low
• Workloads require deep customization of the Temporal server (custom search attributes, replication topologies, archival pipelines)

Crossover Point

A rough heuristic, current as of May 2026: self-hosted becomes cost-competitive at around 30-50M Actions/month for organizations that already operate Kubernetes and Cassandra; below that threshold, Cloud consumption pricing is almost always cheaper than the all-in cost of self-hosting (infrastructure plus operational labor). Organizations with strict data-residency or sovereignty requirements operate self-hosted regardless of cost.

Editor's Note: We have run both. A 2026 client engagement on a fintech compliance workflow stayed on self-hosted Temporal because their data-residency posture (EU-only, no US cloud egress) ruled out Cloud at the time. Cost was roughly $3,200/month infrastructure plus 0.4 FTE for operations, against an estimated $1,400/month on Cloud Standard at their action volume. The compliance constraint made the decision; without it, Cloud would have been the right call. A second 2026 client (mid-stage SaaS, 8M actions/month) moved from a self-host attempt to Cloud after realizing their on-call rotation could not absorb the operational load. Pattern: choose Cloud unless compliance, scale, or an existing Cassandra operations team makes self-host the cheaper option.