Vespa Cloud: Let the world's expert on Big Data Serving take care of the serving.

All features, and more! Scroll down for more details, then see Getting Started:

Proactive, automated world-class 24x7 operations of applications: Focus on innovation rather than operations: Reduces typical operation cost from 1-3 FTE to ~zero
Automated, continuous, safe deployments of application changes
Autoscaling: Applications can scale automatically with actual usage
Development Cloud: Lets developers experiment with copies of applications in seconds
Support plan
Vespa performance tune-up program: Typically reduces cost by 50%
Automated continuous upgrades of Vespa platform / OS: Safe roll-out with minimal redundancy: Typically 50% HW reduction compared to self hosting. New features available daily.
Strong security out of the box: OS and VM hardening, all communication over mTLS with per-node certificates

Model Serving

The best teams do not only train models, but build a system with signals to model training, model training automation, automated model deployment and live model updates. This keeps the model current, and routine tasks are automated - all work can go into improving the model.

Vespa Cloud’s role is being a component in this cycle, see model serving. Emit data for next model training cycle using grid logging and feature export. Then accept a new model version using Safe Launches by just deploying a new model URL. Repeat..

Leading Edge

The best teams are global leaders in their business, or working to be. They drive Vespa to implement new features, so teams can innovate together. Vespa Cloud lets teams access features as they are built. Lead time from feature request to ready-to-use is at best days - but can be years, for game-changing features like tensors and approximate nearest-neighbor.

Organizations use Vespa to get a competitive advantage, and enables that. They use Vespa Cloud to let teams use their expertise best, and work closely with Vespa Experts to use the right features the right ways. Follow the Vespa Blog for the latest news.

Cost Optimization: In-place Upgrades

Vespa is built to be always-on, also while upgrading. Vespa Cloud tracks releases and upgrades multiple times a week. Upgrading means taking nodes offline for new software install. Vespa Cloud cloud orchestrates the process, taking nodes offline safely one by one, and data is synced before upgrading the next node. This keeps all data available at all time, with no extra cluster setup.

In-place software and application updates

As all Vespa Cloud applications must be sized to tolerate single node failures, the upgrader stops upgrading during failures, and resumes once new nodes are in place, and data is synced.

In contrast, manually upgrading cluster takes time, and are often hence put off to later. Often a separate application is spun up, data is migrated, manual tests are run, etc. This has high costs in both HW/cloud resources and people time.

Cost Optimization: Simple Pricing Model

Transparent and easy-to-understand pricing model, based on node resources only. No extra modules needed to unlock Vespa features, as these are native to Vespa and cannot be disabled (this also enables high performance and operational simplicity)

<nodes count="16">
    <resources vcpu="8" memory="32Gb" disk="100Gb"/>

Use resource scaling for seasonal peaks - just change resource allocations, and auto data redistribution enables extra capacity. No need to pay for the light-blue resources when not needed:

Increase resources during seasonal peak

Developer Cloud

Using the Vespa Cloud Sandbox and sample applications enables developers to get started in 15 minutes. The first strawman application with test data is normally built in a day. The fastest teams launch to a production environment the first week. This in order to integrate with production data feeders and query sources early, for rapid iteration.

Developers can deploy Sandbox applications to a Docker instance on their laptop, too - that makes it easy to be productive in all environments. Read more.

Performance Cloud

Vespa applications range from applications with 100k documents and 100k query load, to 20B documents and 1k query load - and everything in between. Update rates range from 0 to 10k/sec.

The applications hence have different latency and throughput requirements, and Vespa has more indexing and topology options to optimize for the different use cases. Adding to this, Vespa’s search threads configuration helps balance latency vs. CPU usage. Use attribute memory options to balance read/write latencies vs. memory usage. Read more in Benchmarking.

Safe Launches

The best launches are non-events. As any launch has a set of unknowns, it makes sense to pad node usage, then quickly scale back when in steady state - see Cluster Sizing. This eliminates the need for lengthy and costly pre-launch performance test cycles for both node instance type and number of nodes used.

Safe launches through system and staging tests

Launches are changes to capacity, configuration and/or code. To safeguard any change, Vespa Cloud has Automated Deployments. Automated Deployments are System, Staging and Production tests run at any deployment. This lets developers define quality metrics and test for these a-priori. The best teams build a strong test suite and are hence confident in all changes. These teams deploy changes to production more times a day, and are orders or magnitude more productive than without automated deployment.

Vespa Cloud has built-in support to block potentially destructive configuration changes, see safe schema modifications.


Bugs happen, and quick failure resolution is essential. See Developer Cloud for how developers copy the exact production config into a sandbox for debugging and development.

Using Vespa Cloud enables Vespa Support to quickly assess production serving problems - the application dashboard is already available to them, with the relevant metrics. In many cases, Vespa Support can in advance predict future problems based on trends.

Nodes and data centers can and will fail. Vespa Cloud detects and replaces and failed nodes automatically, and migrates data for replica re-generation.

Catastrophic event handling with multiple availability zones and Auto failover is supported for the data center / network events.

Data Privacy

The General Data Protection Regulation (GDPR) and California Consumer Privacy Act (CCPA) are relevant for applications with personal data. Highlights:

  1. Prepare - primarily application owner - includes reviewing Vespa’s compliance. Use Vespa’s data auto-expiry in application configuration
  2. Protect - see Security. Also relevant is application logging
  3. Processes - including notifications from Vespa to application owner and application data auto deletion at disuse


Using the Auto in-place Upgrader (above), Vespa code and dependencies are upgraded multiple times a week. As new Operating System versions are available (normally once a month), Vespa base hosts are booted with the new release, using the Auto in-place Upgrader. This keeps the clusters safe - all software on the node is up to date

External and internal interfaces are encrypted using mTLS. Data at rest is encrypted.