SingleStore

What SingleStore actually is

SingleStore is a distributed SQL database that refuses the usual split between transactional and analytical systems. One engine handles rowstore writes, columnstore analytics, JSON, full-text, vector, geospatial and time-series data, and it does so with ordinary SQL. The company started life in 2011 as MemSQL, an in-memory rowstore with ambitions toward massively scalable transactional processing, and rebranded to SingleStore in 2020 as the product grew past that original shape. The managed offering, SingleStore Helios, runs on AWS, Google Cloud and Azure, and the same engine is available self-managed for teams that want to run it on their own iron.

The pitch is HTAP — hybrid transactional/analytical processing — and the practical meaning is that you can ingest a stream of events, transact against it, and run aggregations across billions of rows without building an ETL pipeline into a separate warehouse. We see this pattern most often in financial services, adtech, gaming backends, and operational analytics dashboards where a warehouse that refreshes hourly is simply too stale. If your workload is a plain CRUD app with modest read traffic, most of what makes SingleStore interesting is wasted on you, and cheaper options in /category/database-hosting will serve you better.

Pricing: what you really pay

Helios is metered usage billing, not per-seat licensing. Compute is sold in credits at a list price of $3.96 per credit, and each workspace size consumes a fixed number of credits per hour. On the Standard edition in a Tier-1 AWS region, the smallest workspace (S-00: 2 vCPU, 16 GB RAM) runs at 0.25 credits/hour, which the pricing page prices at $0.99/hour. The ladder scales linearly — S-1 (8 vCPU, 64 GB) is 1 credit/hour at $3.96/hour, and the largest published size, S-20 (160 vCPU, 1,280 GB), burns 20 credits/hour, or $79.20/hour. Compute is metered by the second and billed monthly on actual usage.

Storage is separate: $0.023 per average GB per month in Tier-1 regions, $0.025 in Tier-2, $0.026 in Tier-3, $0.040 in Tier-4. SingleStore does not charge for data ingress, though your cloud provider may bill egress. Several multipliers stack on top of the base rate: Enterprise edition consumes credits at 1.5x, Multi-AZ deployments at 1.3x, region tiers from 1.0x up to 1.375x, and cache or compute scaling above the bundled ratio adds further multipliers (2x cache scaling costs 1.3x credits, 4x compute scaling costs 4x). That stacking is where budgets surprise people — a small workspace can look like pocket change while an always-on Multi-AZ Enterprise deployment in a Tier-4 region bills several multiples of the headline rate.

There is a genuine free on-ramp. The Shared tier gives one free workspace for evaluation, development and non-production testing, with all SQL and programmability features and monitoring, backed by self-support. Standard plans also start with $500 of free credits, and Enterprise starts at $1.49/hour on the smallest workspace. BYOC (bring your own cloud) and Dedicated region pricing require talking to sales, as do volume discounts and committed-use rates. In our view the published rate card is unusually complete for this category — you can model a real monthly bill from the public pages without a sales call — but the multipliers mean you should model your actual configuration, not the $0.99 headline.

Performance & infrastructure

Helios provisions vCPU, memory and disk cache automatically from the deployment size you pick, and compute and cache can be scaled independently. High availability is built into the architecture: when a cloud instance fails, Helios replaces it and keeps the cluster available. The service is a fully automated SaaS — provisioning, upgrades and patching are SingleStore's problem — with 24x7 support on paid tiers. The published SLA is 99.9% for single-AZ deployments and 99.99% when you run across two availability zones. Security posture is documented: ISO 27001, SOC 2 Type 2, and compliance with HIPAA, CCPA and GDPR.

On the database itself, the strengths are real. Columnstore tables with universal storage let row and column data live in the same table; query planning is competent at mixing point lookups with large scans; and the wire protocol is MySQL-compatible, which shortens migration from MySQL-backed analytics stacks more than you would expect. Database branching on Standard is genuinely useful for test environments, and Enterprise adds online point-in-time recovery, smart disaster recovery, audit logging and customer-managed encryption keys. There is also "Flow" for data movement from source databases into a SingleStore workspace, priced in separate credits (0.123 to 18.731 credits/hour by size, currently preview pricing).

We have not benchmarked SingleStore ourselves — we will not invent numbers we do not have — so treat vendor throughput claims as vendor claims. What we can say editorially: the architecture is designed for concurrent mixed workloads, and that is exactly where single-node Postgres or MySQL deployments start to bend. Our general-purpose comparison numbers live at /benchmarks for engines we did test.

Where it falls short

Cost complexity is the headline problem. Four multipliers (edition, AZ, region tier, scaling configuration) plus separate storage billing make the final invoice hard to predict before you run it. Teams new to credit-based database pricing routinely underestimate always-on production workloads: the pricing page's own worked example bills a modest multi-tenant setup at over $35,000 per month.

Second, SingleStore is not the right shape for simple applications. If your workload is a small web app with relational CRUD, you pay for a distributed engine you do not need, and you inherit its operational concepts — shard keys, distribution, reference tables. The learning curve is real even though the SQL surface looks familiar. Third, the free Shared tier is explicitly non-production, and self-support on it means an evaluation that hits a wall may stall. Fourth, the self-managed edition has a separate commercial path with licensing that is not published on the Helios rate card, so self-hosted cost is a sales conversation. And while the MySQL wire compatibility helps, it is not full MySQL compatibility — edge-case SQL, certain storage engine behaviors and replication semantics differ, and migrations deserve a proper test pass.

Fifth, the platform's newer surface area is moving parts you should treat as unfinished until it is not. Flow, the data-movement service for pulling from source databases into a workspace, is priced on the public rate card but labelled preview pricing subject to change. Container and AI services sit in a similar position. Betting a production architecture on preview-priced components is a decision to revisit at every invoice. Finally, the support model is tiered in a way that matters for a database: the free tier is self-support only, Standard includes standard support, and the response expectations you actually get in a production incident are part of the enterprise conversation. For a system that may sit in the critical path of payments or trading, we would want the escalation path in writing before go-live, not after the first bad night.

Who should pick this

Our pick when: you have genuinely mixed transactional and analytical load on the same data, sub-second freshness matters, and you would otherwise run a database plus a warehouse plus a sync pipeline between them. Teams in fraud detection, risk, adtech, IoT backends and live gaming economies fit this profile. The BYOC option also makes it viable for organizations that need data to stay inside their own VPC while still wanting a managed control plane.

We would skip it when: you are building standard web CRUD, your analytics can tolerate minutes of latency, or your team has no appetite for credit-metered pricing. In those cases a straightforward managed Postgres, or Postgres plus a lightweight warehouse, will cost less in both money and attention. Budget-wise, plan from the multipliers down: decide edition and AZ count first, then pick the workspace size, then add storage — not the other way around.

Key takeaways

• HTAP in one engine: transactions, analytics, JSON and vector data under one SQL surface, without ETL between an operational store and a warehouse.
• Transparent published rate card: compute from $0.99/hour (S-00) to $79.20/hour (S-20) at $3.96/credit, storage $0.023–$0.040 per average GB/month by region tier, no ingress charge.
• Multipliers stack: Enterprise 1.5x, Multi-AZ 1.3x, region tier up to 1.375x — headline prices understate real production bills.
• Real free tier for evaluation (one shared workspace, non-production) plus $500 in credits on Standard; production SLA is 99.9% single-AZ or 99.99% dual-AZ.
• Wrong tool for simple CRUD apps; the distributed-engine concepts and cost model only pay off under genuinely mixed, high-throughput workloads.
• ISO 27001, SOC 2 Type 2, HIPAA/CCPA/GDPR compliance documented; BYOC keeps data in your own VPC.

Sources

• https://www.singlestore.com/cloud-pricing
• https://www.singlestore.com/pricing/
• https://en.wikipedia.org/wiki/SingleStore
• https://singlestore.com/blog/singlestoredb-on-prem-to-singlestoredb-cloud
• https://www.modern-datatools.com/tools/singlestore/pricing