Altinity × Nirvana: The joint stack behind the hot data for ClickHouse®

Who runs what, from your dashboard down to the disk, and why the answer keeps ClickHouse® data always hot.

TL;DR:

• Nirvana is now a BYOC option on Altinity.Cloud: Altinity's managed service for ClickHouse® running on Nirvana's high-performance storage cloud, GA now
• One boundary organizes the stack: Altinity runs everything on NKS (clusters, operator, console, support); Nirvana runs NKS and everything beneath it (compute, ABS, networking)
• ABS ships 20,000 baseline sustained IOPS included, vs AWS gp3's default 3,000, with no per-IOP surcharges
• Best case: your workload also runs on Nirvana, putting database and application on the same rack with $0 internal egress
• Proven before GA: SSB selective queries at 34-63 ms against 3 billion rows, ~70M rows/sec sustained scans, stable run to run

ClickHouse® performance depends on two broad layers: how the database is designed and operated, and the infrastructure beneath it. This partnership brings together specialists in each: Altinity and Nirvana.

Altinity is the largest independent managed service provider for ClickHouse®, working with the database since 2017. It already supports BYOC deployments across AWS, GCP, Azure, and Hetzner to allow users the flexibility of bringing their own cloud. Nirvana, a high-performance storage cloud, serves production blockchain, AI, database, and other stateful workloads, with Accelerated Block Storage (ABS) and Nirvana Kubernetes Service (NKS), both launched in 2026.

The partnership places Altinity's managed service for ClickHouse layer directly on Nirvana's infrastructure. Here is how the joint stack works together to provide always hot ClickHouse data.

Joint stack architecture overview

Nirvana is now a BYOC option on Altinity.Cloud, meaning Altinity runs its managed service for ClickHouse service on Nirvana cloud, backed by ABS block storage: 20,000 baseline sustained IOPS included (vs AWS gp3 default 3000 IOPS), no over-provisioning needed. The best scenario is when the customer's application also runs on Nirvana: the database and workload can then be placed on the same rack, with minimal distance and hence the fastest execution possible with $0 internal egress between them. The stack splits into three domains of responsibility.

You control:

• Your applications, workloads, and data: the pipelines, dashboards, APIs, and business logic above the database
• Your Altinity & Nirvana account, project-scoped API-key permissions, access policies, and operational requirements
• Your topology choices: compute shape, storage size, shards and replicas, and the Altinity support tier
• The backup, availability, and disaster-recovery outcomes the deployment must meet

Altinity manages (the managed service for ClickHouse® layer):

• ClickHouse® clusters: automated cluster creation, upgrades, rescaling, and monitoring
• The NKS environment required by Altinity.Cloud: provisioning and configuration of the service resources, deployment of the ClickHouse stack, and monitoring from the service layer
• Service-side networking: provisioning and configuration of the network resources and database access paths within the customer's Nirvana project, according to customer-defined policies
• The Altinity Kubernetes Operator and Altinity Cloud Manager console, with monitoring through the open-source stack of Prometheus, Grafana, and Loki
• Database access controls: users, roles, permissions, allowlists, and access tokens, administered according to customer-defined policies
• Schema and query support: Altinity's ClickHouse® engineers assist with cluster design, schema design, query optimization, and advice on scaling production systems
• Backups and operational support: automatic backups with on-demand restore of a full cluster or single tables, and support from 8×5 to 24/7 coverage with SLA-backed response times, according to the selected service plan. High availability and disaster recovery remain shared responsibilities: Altinity provides the mechanisms, while the customer defines the required outcomes and settings

Nirvana operates (the infrastructure layer):

• The managed NKS platform and control plane: Kubernetes lifecycle operations, upgrades and maintenance, plus Karpenter-based worker-node autoscaling
• New compute model: dedicated VMs on high-clock AMD EPYC processors
• ABS block storage: 20,000 baseline sustained IOPS included, with burst performance up to 600,000 IOPS and sub-millisecond latency. No burst credits, separate performance metering, or per-IOP surcharges
• Networking: private VPCs with firewall rules, same-rack placement when the customer's workload also runs on Nirvana

Altinity runs everything on NKS; Nirvana runs NKS and everything beneath it. That way, the infrastructure under ClickHouse comes from a vertically integrated provider whose storage, compute, and networking are designed and benchmarked together for stateful, performance-sensitive workloads. And because the two teams built and benchmarked this integration together, a performance question that crosses the boundary lands in an engineering relationship, not a support ticket between strangers.

Proven before GA

Before this partnership was announced, Altinity benchmarked the stack on Nirvana to make sure the product held up, then proved it with a production client. Only then did Nirvana become a provisioning option in Altinity's documentation, alongside AWS, GCP, Azure, and Hetzner.

The results of that proving-out are public: on the joint stack, SSB selective queries answered in 34-63 ms against a 3-billion-row table, full scans sustained ~70M rows/sec from disk, and repeat runs landed within a few percent of each other (full report).

And the integration itself is evidence: getting on the joint stack is five steps in the Altinity console, with one project-scoped API key. When two companies wire their products together properly, the seam is small enough to fit in a quickstart guide.

Try it

Pick Nirvana Labs as your BYOC option in Altinity.Cloud, free trial live. Or bring your dataset and queries, and we'll benchmark them on the same infrastructure the reports were measured on: Talk to us →