Local NVMe performance and an open-source control plane, from the team that has run five managed Postgres services over fifteen years. Available on Ubicloud Metal, AWS, and Google Cloud.
TPC-C (sysbench), published May 2025 · standard-8 ($327.50/mo) vs. tested AWS RDS 8 vCPU ($725.02/mo). Exact configuration and methodology published. Workload-specific. Methodology · Full results
"Ubicloud eliminated our IOPS bottlenecks and is already saving us $300k a year, with $500k projected as we migrate more workloads."
Alexander Belanger, Founder, Hatchet
RDS costs of about $60k per month dropped to the mid-$20k range. The core database moved with the application online the whole time.
[ READ THE STORY → ]"Migration was about as painless as I could possibly have expected. If I was to grade it, I'd give it an A. You can't really beat that."
Sam Ingrassia, CTO, AudienceKey
Multi-terabyte PostgreSQL with complex partitioning, moved without unexpected downtime. The savings opened new budget for software development.
[ READ THE STORY → ]The same control plane, the same standard PostgreSQL, and the same operations on our cloud, on AWS, or on Google Cloud. Pick the infrastructure and keep everything else.
Local NVMe on our own hardware in Germany, Finland, and US-Virginia. The benchmarks below were measured here. From $15.63/mo.
AWS regions with NVMe instance storage, or inside your own AWS account for select customers. 8.5x faster transactions than RDS, measured on AWS.*
See the AWS numbersFirst-class PostgreSQL on GCP, in preview. A database engineer onboards you directly.
*Measured on AWS infrastructure with pg_bench.
Traditional managed Postgres sits behind network storage, where I/O mostly leaves the machine and every IOPS is a line item. Ubicloud puts PostgreSQL on local NVMe. Latency drops. The IOPS bill simply stops existing.
same machine
Postgres and its disk share a host. No hop, and no IOPS meter.
crosses the network · adds latency
Postgres and its disk are separate services. I/O mostly leaves the machine, and it is metered.
1.4x Aurora · 4.6x RDS
Running on AWS doesn't mean paying RDS prices. Ubicloud on AWS uses NVMe instance storage with I/O included, so there is no provisioned-IOPS line item. Available in AWS regions, or inside your own account.
Measured on AWS infrastructure with pg_bench and TPC-H.
max 256K / $0.065 per IOPSmax 256K provisioned IOPS / billed $0.065 per IOPS/mo
1M+ IOPS included — no extra charges
>42× higher IOPS, <59% less cost
Live calculator using the same data as our AWS landing page. RDS pricing follows AWS published rates.
Step through exactly what this platform does on its worst days, including what it can't do. We'd rather show you the machinery than promise you a number.
Everything here has a console page, a CLI command, and an API endpoint, and you hold the actual postgres superuser role. Nothing on this page needs a support ticket.
Every server's role and health, connection details, and current activity. One view per cluster, with the same data in the console.
CPU, memory, disk, connections, cache hit ratio, and deadlocks, plus Prometheus remote-write straight into your monitoring.
Postgres, PgBouncer, and upgrade logs, searchable for up to 7 days, or shipped to your provider via OpenTelemetry.
Any minute of the last 7 days, restored as a new instance you can inspect before cutting over.
In-place major version upgrades: fork first, test, then upgrade in your maintenance window.
Standbys and read replicas on demand. Promote a replica to standalone any time, and rehearse failover before you need it.
386 PostgreSQL parameters on PostgreSQL 18 (368 on 17, 353 on 16) and 64 PgBouncer settings, validated on write; restart-required ones are flagged.
Per-database firewall, private DNS, connected subnets. One allow-list rule covers both the direct port and the pooled one.
Your versions and extensions against the 86 we support, config diff, downtime budget.
Logical replication or dump and restore, timed on a fork instead of production.
Schema, row counts, application behavior, real query performance.
In a window you choose, connection strings staged in advance.
Reverse replication until you're sure. Rollback is a plan, not a hope.
A Postgres engineer works the migration with you. This is the team that has run five managed Postgres services over fifteen years.
Today that means standard PostgreSQL migration methods plus direct engineering help. There is no packaged "zero-downtime migration service", and we won't pretend otherwise. It works, though: Hatchet moved 15 TB this way with the application online the whole time, and AudienceKey graded their multi-terabyte move an A.
Ozgun Erdogan and Umur Cubukcu co-founded Citus Data, the distributed-PostgreSQL company Microsoft acquired in 2019. The team went on to run one of the largest managed Postgres services at Azure.
Co-founder Daniel Farina built Heroku Postgres in its foundational years, operating PostgreSQL as a cloud service before that was a category.
The same people who write the control plane publish how it handles fencing, memory pressure, WAL backlog, and disk exhaustion. In public.
The engineers who designed the failure behavior described on this page are the ones operating it, and the ones you reach when it counts.
Built by the team behind Citus, Heroku Postgres, and Azure Postgres. Open source, and measured in public.
