Feldera in 2025: Building the Future of Incremental Compute

For 50 years, the industry has been doing analytics the same way. Collect data. Run batch jobs. Hours or days later, you get insights. Meanwhile, your data keeps arriving and changing, but your only option is to reprocess everything in big expensive batches.

Feldera broke this 50-year barrier with incremental computation. A better way to create sophisticated analytics at the speed your data actually changes - real-time.

We’ve established ourselves as the only platform on the market that can incrementally evaluate arbitrarily complex SQL at scale, and traverse the entire spectrum from batch to streaming in one engine. All demonstrated in production at category-leading enterprises.

2025 was a year of momentum

The year in numbers:

166 unique releases
1,162 total changes shipped to production, over half of which are major features or improvements
A new release every 2.4 days on average

For our small team, this is a feat in and of itself. We’re shipping you the features and improvements you need at startup speed.

What we accomplished

All SQL welcome.

Today, we’re proud that our customers have ported hundreds of thousands of lines of SQL off of their warehouse vendors to Feldera, and we maintain full parity in terms of the SQL we run. This means all kinds of joins, aggregations, window functions, complex data types, JSON handling and more, organized as deeply nested views.

We want to give a major shoutout to our co-founder Mihai, who is a top contributor to Apache Calcite and will serve as PMC chair for 2026. A significant part of his contribution includes bug fixes and code review that have been essential to improving Calcite’s quality and reliability. Mihai has also added new features like support for UUID data types, support for unsigned integer data types, and support for lateral column aliasing (a feature that allows you to use column names you introduce right away in some SQL dialects). On top of it all, we support user-defined aggregates.

State-of-the-art infrastructure at scale.

We remain the only engine on the market that can uniformly compute over the entire spectrum from batch to streaming, for arbitrarily complex SQL, at scale.

Reliability and efficiency have been a key focus as well. We introduced pipeline fault tolerance to prevent data loss and resilience. We added S3-backend pipelines and checkpointing for operational convenience. We invested heavily into our storage-layer to handle datasets that are orders of magnitude larger than RAM. We shipped category-defining capabilities like transactions for efficient backfills and incrementally modifying pipelines.

We are also proud of our self-managed control plane, with enterprise-grade features that include multi-tenancy, security, and isolation. To compile massive SQL programs, earlier this year we shipped a feature that reduced compile times for some customers from 30 minutes to 2 minutes by splitting generated programs into many small Rust crates. We also shipped parallel compilation, so you can compile your pipelines simultaneously. There is an end-to-end focus on giving customers control of workload isolation boundaries, resource management, authentication and security controls. Rest assured, Feldera can handle your organization's scale and complexity securely.

Your data sources called. They want in.

In order to get the most value out of a technology as powerful as incremental computation, it needs to work seamlessly with your existing data and lakehouse stack. That’s why we shipped hundreds of improvements to our connector infrastructure. We significantly improved our Kafka, Delta Lake, Iceberg, Postgres, S3 and other connectors, invested heavily into hardening different data formats (Avro, Parquet, and more), added features to pass connector metadata to pipelines (like Kafka headers), and making throughput improvements to our existing HTTP ingress.

Many of these improvements have helped our customers migrate to what we call the real-time Medallion Architecture (read our blog post about it).

A focus on user experience.

We added many features to improve the pipeline engineering and operations experience. We introduced a visual in-product profiler, ad-hoc queries, and support bundles for one-click diagnostics, which make it easier to identify and pinpoint issues. We integrate out of the box with your monitoring stacks, like Datadog, Prometheus and Grafana. Developers can now programmatically control Feldera pipelines via hundreds of improvements to the Python SDK and use the improved fda CLI with new commands.

See what’s possible

We’ve had a lot of momentum in 2025 and we’re proud of our engineering team. Companies using our platform have cut cloud spend by 10x or more, reduced 70-node Spark clusters with single digit Feldera instances, and gone from hours-old insights to sub-second latency for customer facing analytics.

If you’re struggling with high cloud spend and slow analytics, book a demo today. We’d love to show you how incremental computation can reduce your cloud spend by 10x.

So what’s next in 2026?

2026 is about making incremental compute inevitable, and we’re not slowing down. You can expect major performance and scale improvements, even deeper connector integrations, more enterprise features, and technical deep dives, video tutorials, and detailed customer stories.