A data engineering team was spending $18K/month on ClickHouse Cloud for 50TB of analytics data with 500B daily queries. After a JusDB cost analysis, we moved them to a self-hosted 3-node cluster on AWS — their monthly bill dropped to $4.2K while query latency improved 20% due to local NVMe storage. ClickHouse Cloud is excellent for teams without DBA resources, but the premium is real and quantifiable.
This guide breaks down the real total cost of ownership for ClickHouse Cloud versus self-hosted, including the hidden costs that change the math at scale.
- ClickHouse Cloud costs 3-5x more than self-hosted at equivalent scale for sustained workloads.
- Self-hosted requires DBA time (~5-10 hrs/month for tuning, upgrades, monitoring) — factor this into TCO.
- ClickHouse Cloud is worth it for teams under 10TB, early-stage startups, or teams without dedicated DBA resources.
- At 50TB+ with predictable query patterns, self-hosted ROI is typically positive within 3 months.
Cost Comparison at Different Scales
| Scale | ClickHouse Cloud | Self-Hosted (AWS) | Break-even |
|---|---|---|---|
| 5TB, 100M queries/day | ~$800/mo | ~$600/mo + DBA time | Self-hosted marginal |
| 20TB, 1B queries/day | ~$4,500/mo | ~$1,200/mo + DBA time | Self-hosted wins ~mo 2 |
| 50TB, 500B queries/day | ~$18,000/mo | ~$4,200/mo + DBA time | Self-hosted wins month 1 |
| 100TB+ | Custom pricing | ~$8,000-12,000/mo | Self-hosted wins clearly |
ClickHouse Cloud Pricing Model
ClickHouse Cloud uses compute + storage pricing. Compute scales automatically (a strength) but costs add up quickly for always-on workloads:
# ClickHouse Cloud approximate pricing (2025)
Compute: ~$0.33/compute-hour (scales with query load)
Storage: ~$0.023/GB/month (compressed)
Egress: ~$0.09/GB (significant for large result sets)
# Example: 50TB compressed, 100 compute-hours/day
Monthly = (100 * 30 * $0.33) + (50,000 * $0.023)
= $990 compute + $1,150 storage = $2,140/month base
+ egress if exporting dataSelf-Hosted Architecture on AWS
# Production self-hosted ClickHouse cluster
# 3x r6gd.4xlarge (128GB RAM, 950GB NVMe, 16 vCPU)
# Cost: ~$1.50/hr each = $3,240/month for 3 nodes
# ClickHouse config for NVMe storage
cat /etc/clickhouse-server/config.xml
# /mnt/nvme/clickhouse/data/
# Enable async inserts for high-throughput ingestion
SET async_insert = 1;
SET wait_for_async_insert = 0;
SET async_insert_max_data_size = 10485760;NVMe local storage delivers 3-5x better query performance than EBS for ClickHouse. If you go self-hosted, use instance types with local NVMe (r6gd, i3en) rather than EBS-backed instances.
Hidden Costs in Self-Hosted
Self-hosted is cheaper on compute and storage, but factor in these costs:
- DBA time: ~5-10 hours/month for tuning, upgrades, monitoring, and incident response
- Monitoring: Grafana + Prometheus setup, alerting configuration (~$50-200/month for hosted Grafana)
- Backup: S3 backup storage (~$0.023/GB) + transfer costs
- High availability: 3-node minimum for HA adds ~2x hardware cost vs single node
-- Monitor ClickHouse cluster health (self-hosted)
SELECT hostName(), is_leader, total_rows_approx, total_bytes
FROM system.replicas LIMIT 20;
SELECT query_kind, count(), avg(query_duration_ms) AS avg_ms
FROM system.query_log
WHERE event_time > now() - INTERVAL 1 HOUR AND type = 'QueryFinish'
GROUP BY query_kind ORDER BY avg_ms DESC;When ClickHouse Cloud Makes Sense
ClickHouse Cloud is worth the premium when:
- Your team has no DBA or data infrastructure expertise
- You are under 10TB and still experimenting with your analytics architecture
- Your query load is spiky and unpredictable (auto-scaling saves cost vs over-provisioned self-hosted)
- You need rapid provisioning for a proof-of-concept or new product feature
ClickHouse Cloud data egress costs can be significant if you run large result exports. A 100GB daily export at $0.09/GB adds $270/month that is easy to miss in planning.
- ClickHouse Cloud costs 3-5x more than self-hosted at scale — calculate your specific break-even before committing.
- Self-hosted requires 5-10 hrs/month DBA time — include this in your TCO calculation honestly.
- NVMe instance storage is essential for self-hosted performance — avoid EBS-only instance types.
- ClickHouse Cloud auto-scaling is a genuine advantage for spiky, unpredictable workloads.
Working with JusDB on ClickHouse
JusDB designs and manages ClickHouse deployments — both self-hosted and ClickHouse Cloud — and handles migrations between them. We have migrated teams from ClickHouse Cloud to self-hosted with 3-4x cost reduction and improved query performance.