YugabyteDB

What YugabyteDB actually is

YugabyteDB is a distributed SQL database: PostgreSQL-compatible query layer (YSQL) on top of a distributed storage engine (DocDB), built for workloads that need ACID transactions across failure zones and, when you want it, across regions. It is the Spanner-inspired architecture without Spanner's lock-in — a consistent/partition-tolerant system with automatic sharding, built-in replication, and failover that happens below the SQL layer. There is also a Cassandra-compatible API (YCQL) for legacy workloads, and change data capture for streaming out.

The company behind it is Yugabyte, Inc., founded in February 2016 by three ex-Facebook engineers — Kannan Muthukkaruppan, Karthik Ranganathan and Mikhail Bautin — who had built and operated Cassandra and HBase at Facebook. Headquartered in Sunnyvale, California. The database went fully open source under Apache 2.0 in July 2019, and that is one of the sharper contrasts in this category: unlike CockroachDB's source-available licence with revenue caps and telemetry, you can run YugabyteDB's core without restrictions and without reporting anything. The commercial angle is the managed service, YugabyteDB Aeon (the renamed YugabyteDB Managed / Yugabyte Cloud), plus enterprise support and self-managed subscriptions.

The company is well capitalized — a $188M Series C in October 2021 at a $1.3B valuation, roughly $290M disclosed across all rounds — which matters when you are betting a production database on a vendor. A decade of existence, a real enterprise sales motion, and the freedom to walk away to the open-source build at any time is a healthier posture than most DBaaS buyers get.

What is the product actually for? The honest pitch is: Postgres semantics and tooling where a single-node Postgres stops being enough — multi-zone or multi-region deployments, horizontal write scaling, planned failover between regions, geo-partitioned data for residency rules. If one well-tuned Postgres instance serves you, this is not your problem and YugabyteDB is not your bill.

Pricing: what you really pay

YugabyteDB Aeon now uses one pricing model across all deployment options (fully managed or bring-your-own-cloud): tiers priced per vCPU per month, with storage and data transfer billed additionally.

• Standard — from $125/vCPU/month. For startups and scale-ups needing scalability and zone-level fault tolerance. Includes the core database (YSQL, YCQL, CDC), core security and networking (RBAC, encryption in transit and at rest, VPC peering and Private Link), migration tooling (YugabyteDB Voyager), automation (REST API, CLI, Terraform, periodic full backups, automated upgrades and patching), core observability, and 24/7/365 support.
• Professional — from $167/vCPU/month. Adds global deployment topologies (multi-region stretched clusters, read replicas, geo-partitioning for data residency) and enhanced observability (Performance Advisor, metric and audit-log export, webhook alerts). Enterprise security (SSO/OIDC, customer-managed keys) and business continuity (PITR, disaster-recovery workflows, blue-green upgrades) are paid add-ons at $25/vCPU/month each.
• Enterprise — quote-based. Includes the add-on capabilities, plus bring-your-own-cloud, partner integrations (Commvault, Delphix, Nutanix) and enterprise commercial terms.

A free Sandbox cluster exists for evaluation: single node, up to 2 vCPU / 4 GB RAM / 10 GB storage, capped at 15 simultaneous connections. It is explicitly not for production or performance testing, pauses after 10 days of inactivity, and is deleted after 15. A full-featured free trial of Aeon is available on top of that. The open source database itself is free without restrictions.

How does the pricing land? A production-shaped cluster — three nodes across three zones at 2 vCPU each — is 6 vCPU, so $750/month at the Standard rate before storage. That is the number to hold in your head. It is roughly in line with CockroachDB's recently restructured cloud pricing and a world away from self-managed Postgres on a €20 VPS; you are paying for distribution, failover and operational offloading, not for compute. On the transparency side, we give credit where it is due: public per-vCPU rates, a public feature matrix per tier, published add-on prices, and a legal SLA page. The soft spots are "starting at" phrasing (real clusters size above the entry), the Enterprise tier behind a contact form, and storage/transfer costs spelled out in the billing docs rather than on the pricing page itself.

Performance & infrastructure

Aeon runs on AWS, Microsoft Azure and Google Cloud — single- and multi-region clusters across all three, which puts US, EU and APAC deployments on the same footing and covers data-residency needs via geo-partitioning in the Professional tier. Bring-your-own-cloud deploys the same database on standard infrastructure in your own environment with Yugabyte managing it, at the same per-vCPU price.

On availability, Yugabyte publishes what many DBaaS vendors don't: a real SLA. 99.99% monthly availability for Aeon, measured per cluster, with a precise downtime definition (no established connections and all connection attempts failing) and a service-credit table — 5% credit at 99.5-99.99% uptime for zone-level fault-tolerant clusters, 10% at 99-99.5%, 25% below 99%, and lower credits for single-zone clusters. Read the fine print: credits are granted "at our reasonable discretion," cloud-provider failures are excused outages, and you must file a support ticket within 24 hours. Still, this is a measurable commitment, and the multi-zone credit structure quietly states the engineering truth — a three-node cluster across three zones gets a materially better deal than one crammed into a single zone.

Backup posture: every paid tier includes adjustable periodic full backups; point-in-time recovery and DR orchestration (failover, fail-forward, fail-back) are Professional add-ons or Enterprise inclusions. Managed upgrades and security patching are automated. We have not benchmarked YugabyteDB ourselves, so we will not quote throughput numbers — for our own measured results on other platforms see /benchmarks, and treat any vendor's performance charts (including Yugabyte's) as marketing until reproduced.

Where it falls short

Cost, plainly. Distributed SQL is priced as a premium product: $125/vCPU/month on the entry tier means a serious multi-zone cluster runs into four figures a month quickly, and PITR — a feature many teams consider table stakes for a production database — is a $25/vCPU/month add-on unless you are on Enterprise. If your availability model is "restore from last night's backup," a managed single-node Postgres will serve you at a fraction of the price.

Operational complexity is real even in the managed form. Distributed databases have more moving parts than single-node Postgres: consistency modes, tablet splitting, region placement, replica lag. Yugabyte's docs are good and the managed service absorbs most of it, but teams still need people who understand why a cross-region write is slow and what xCluster semantics mean. The common failure mode we see is adoption by teams that wanted "Postgres but bigger" and got a distributed systems problem.

Compatibility is PostgreSQL-flavoured, not PostgreSQL-complete: most extensions and tooling work, some do not, and behaviour differences surface at the edges. Voyager, Yugabyte's migration tool, exists precisely because migrations from Oracle/Postgres need assessments, live migration and fallback paths — budget for that project, not a pg_dump afternoon.

Finally, the free Sandbox is a demo, not a tier: one node, 15 connections, auto-deleted after two inactive weeks. Hobby projects live elsewhere.

Who should pick this

Our pick for teams with a concrete distribution requirement — multi-zone writes, regional failover, geo-partitioned data for residency — who want Postgres semantics and a genuinely open-source core they can self-host if the relationship ends. Regulated industries like the combination of data-residency controls (Professional tier), customer-managed keys and a published SLA with credits.

We would also shortlist it for organizations standardizing on one distributed SQL across clouds: Aeon spans AWS, Azure and GCP with a single price model, and BYOC at the same rate is unusual and welcome.

We would skip it for: anyone whose database fits on one node, teams without distributed-systems experience and no budget for the learning curve, workloads that need cheap read-heavy scaling (read replicas on a single-node Postgres or a caching layer may do), and buyers who need PITR but balk at the add-on.

Key takeaways

• Distributed SQL with Postgres semantics: YSQL on DocDB, ACID across failure zones, automatic sharding and replication, plus a Cassandra API (YCQL) and CDC.
• Truly open core (Apache 2.0 since 2019) — no revenue caps, no telemetry requirement — with the managed service (Aeon) as the commercial layer.
• Pricing: Standard from $125/vCPU/month, Professional from $167/vCPU/month, Enterprise by quote; PITR and enterprise security are $25/vCPU/month add-ons on Professional. A 3-node/6-vCPU production cluster starts around $750/month plus storage.
• Published 99.99% SLA with a concrete credit table (5-25%), though credits are discretionary and cloud-provider outages are excused.
• Runs on AWS, Azure and GCP with US, EU and APAC coverage; BYOC at the same price as fully managed.
• Free Sandbox (2 vCPU/4 GB/10 GB, 15 connections) is for evaluation only; a single-node Postgres is still the right answer for most databases.

Sources

• https://www.yugabyte.com/pricing/
• https://docs.yugabyte.com/stable/faq/yugabytedb-managed-faq
• https://docs.yugabyte.com/stable/yugabyte-cloud
• https://trust.yugabyte.com/legal/aeon-service-level-agreement/
• https://trust.yugabyte.com/legal/aeon-terms-of-service/
• https://docs.yugabyte.com/stable/yugabyte-cloud/cloud-admin/cloud-billing-costs-classic
• https://en.wikipedia.org/wiki/YugabyteDB