A San Francisco startup with roots in academic computer science emerged from relative obscurity this week with $21.5 million in funding to challenge how companies manage ever-growing data warehouses.
Feldera, which builds what it calls an "incremental compute engine," announced September 21 it had raised a $15.4 million Series A led by Inovia Capital, adding to an earlier $6.1 million seed round. The company promises something that sounds almost too good for enterprise data engineers: SQL queries that update in milliseconds while slashing warehouse compute bills by 95%.
The pitch comes at a moment when businesses are drowning in data but struggling to keep it fresh. Traditional data warehouses reprocess entire datasets every time information changes, burning through cloud computing credits and leaving analytics stale for hours. Feldera's approach calculates only what's different, updating results as new data arrives.
"We're seeing companies spend tens of thousands a month just keeping dashboards current," said Lalith Suresh, Feldera's CEO and a former senior researcher at VMware. The alternative his team built runs on what they call Incremental View Maintenance—a concept that's been theoretically possible for decades but notoriously difficult to execute at scale.
The technology isn't purely theoretical anymore. Auth0, the identity platform now owned by Okta, has deployed Feldera to manage permission checks across more than 7 billion entries. According to Feldera, the system updates its authorization index in an average of 195 milliseconds while using just 58 gigabytes of memory at peak load. For context, that's handling roughly the population of Earth in permission entries on hardware modest enough to run in a single rack.
The Academic Foundation
What sets Feldera apart from the crowded field of database startups might be its pedigree. The core team built the engine on DBSP—Database Stream Processing—a theoretical framework they developed that won best research paper at VLDB 2023, one of computer science's most prestigious database conferences. A year later, SIGMOD recognized the work as a research highlight.
That academic firepower translates to some unusual capabilities. While competing systems often stumble on complex queries, Feldera claims it can incrementally maintain arbitrary SQL, including recursive queries and sliding windows, with correctness guarantees matching traditional warehouses. In one benchmark the company detailed earlier this year, it processed a 200-gigabyte backfill of 250 million rows from Delta Lake, then updated outputs in roughly 200 milliseconds per subsequent change on a single 16-core machine.
The founding team reads like a who's who of systems research. Chief Scientist Mihai Budiu previously worked at VMware, Barefoot Networks, and Microsoft. CTO Leonid Ryzhyk, who co-created DBSP with Budiu, has stints at Samsung Research, Carnegie Mellon, and the University of Toronto on his resume. Chief Engineer Ben Pfaff founded and led development of Open vSwitch, networking software that powers much of the cloud infrastructure world.
Suresh, who holds a PhD from TU Berlin, is betting that this depth of expertise can solve what investor Taha Mubashir of Inovia Capital called "a 50-year-old fundamental database problem." That might be marketing speak, but the mathematics backing DBSP suggests they've at least made progress where others stalled.
A Crowded Field
Feldera isn't alone in chasing incremental computing. Materialize, which uses Timely and Differential Dataflow to maintain streaming views, has raised significantly more capital and built a larger team. RisingWave, a Rust-based streaming database, targets similar use cases. Even Databricks, the data platform giant co-founded by Ion Stoica—who also participated in Feldera's seed round—published research describing Enzyme, its own incremental view maintenance system.
The startup also positions itself against dbt's incremental models, arguing those aren't true universal engines and often require full rebuilds when schemas change. Whether that distinction matters to customers remains to be seen.
Battery Ventures and Costanoa Ventures, which led the seed round, both joined the Series A. Dealroom pegged the round's valuation at $62 million, though Feldera declined to confirm the figure. The company lists 12 employees on LinkedIn and is hiring across engineering and go-to-market roles in the US, Europe, and India.
What Comes Next
Feldera plans to invest the fresh capital in core engine improvements, deeper integrations with data lakehouses, and building out a fully managed cloud service. That last piece matters—most startups in this space have learned that even brilliant technology needs to be dead simple to deploy if it's going to displace entrenched incumbents.
In a prepared statement, Suresh made the now-obligatory nod to artificial intelligence: "AI agents are only as good as the data they can access." It's the kind of line every database startup includes in fundraising announcements these days, though Feldera's focus seems squarely on the less glamorous work of keeping enterprise data pipelines running efficiently.
The real test will come as the company scales beyond early design partners like Auth0. Incremental computing has always worked beautifully in demos and struggled in production, where messy real-world data and edge cases multiply. Feldera's mathematical proofs and research accolades buy credibility, but customers will ultimately care whether it actually cuts their AWS bills in half without breaking when queries get weird.
For now, the startup has cash and a story that resonates with investors betting on infrastructure. In a market where data volumes keep growing faster than budgets, even a partial solution to the reprocessing problem could carve out a meaningful business.
