DeltaStream

The company that wants streaming data to be boring

Here is a fact about streaming data that the industry would rather you not dwell on: the technology mostly works. Apache Flink, the open-source engine that powers a good chunk of the world's real-time pipelines, is genuinely powerful. It can process millions of events a second, join streams, maintain state, and never blink. The catch - and it is a large catch - is that running it has historically required a small team of specialists who think about checkpointing and backpressure for a living. The capability was never the problem. The operational tax was.

DeltaStream, a roughly 15-person company on Sand Hill Road in Menlo Park, is a bet that if you remove that tax, everything else takes care of itself. Its pitch is almost aggressively unglamorous: you get all the power of Apache Flink, you write SQL, and you never see a cluster. That is the whole product. The interesting thing about DeltaStream is not that it does something Flink cannot. It is that it does the same thing while asking much less of you.

This is a company built by someone who has run this experiment before. Founder and CEO Hojjat Jafarpour created ksqlDB at Confluent - the first database purpose-built for stream processing, a project that popularized the radical idea that you could point SQL at a Kafka topic and get answers. Having built the streaming-SQL engine that a lot of people already use, he left to build the serverless platform around it. In a sense, DeltaStream is the sequel. Same author, bigger budget.

What you can actually do with it

The mechanics are refreshingly literal. You connect DeltaStream to a streaming source - Apache Kafka, AWS Kinesis, Confluent Cloud, AWS MSK, Redpanda, Pulsar - and then you write SQL against it as if it were a table that never stops filling up. You can create a materialized view that stays current in real time. You can build a pipeline that transforms one stream into another. You can set up real-time anomaly detection, feed a dashboard, or route enriched events to a downstream sink. And you can do it in minutes, in a language your data team already speaks, which is the actual point.

The part that makes DeltaStream more than a nicer Flink wrapper is governance. It adds role-based access control, hierarchical namespacing, a streaming catalog, and secure data sharing on top of data that is constantly in motion. This is a genuinely hard problem - it is one thing to govern a table that sits still, another to govern a firehose - and it is the piece the company likes to compare to what Snowflake and Databricks did for stored data. The analogy is doing some work, but it points at a real gap: everyone built controls for data at rest, and comparatively few built them for data in flight.

Raw streams in, governed real-time output out - with no cluster in the middle for you to manage.

One platform, three engines, no tool sprawl

The bigger swing is DeltaStream Fusion, which reached general availability in 2025. Fusion's premise is that most data teams do not actually want three separate systems - one for streaming, one for real-time analytical queries, one for batch. They want one. So Fusion orchestrates three engines under a single serverless surface: Apache Flink for streaming, Apache Spark for batch, and ClickHouse for real-time analytics. It integrates with Apache Iceberg and Snowflake for lakehouse use cases, so streaming and stored data show up in the same view.

The commercial logic here is not subtle, and that is a compliment. Every additional system a data team runs is another bill, another on-call rotation, another way for something to break at 2 a.m. Collapsing three tools into one is a story a budget owner can follow. Fusion also leans into the moment: it adds AI-assisted pipeline creation, so you can describe what you want and have it generate the streaming SQL, and it positions the platform as the real-time context layer for GenAI applications and agents. Stale context breaks autonomy; a live data layer is how you keep an agent current. Whether that becomes a category or a feature is an open question, but it is a reasonable place to be standing.

The money, and who is behind it

DeltaStream has raised $25 million total. It started with a $10M seed in 2021 led by New Enterprise Associates, and added a $15M Series A in September 2024, again led by NEA, with Galaxy Interactive and Sanabil joining. That is a tidy, focused cap table for an infrastructure company - the kind of backing that buys runway to build a hard product rather than a rush to spray it at a market. The funding, in the company's telling, goes toward product development, a modestly growing team, and partnerships across the data and AI ecosystem.

It is worth being clear-eyed about the competitive neighborhood, because it is crowded and the neighbors are enormous. DeltaStream is effectively competing with Confluent (Jafarpour's former employer, and the home of Flink-based offerings), with Databricks and Snowflake as they push into streaming, and with a cohort of newer entrants like Decodable, RisingWave, and Materialize. Against that field, a 15-person company is not going to win on breadth. The bet is narrower and sharper: be the layer that makes data in motion as governable and easy as data at rest, and let the integrations - Databricks' UI, Snowflake, Iceberg, Azure - carry you into the stacks people already run.

Why the boring part is the interesting part

The most telling thing about DeltaStream is what its philosophy subtracts rather than adds. Most infrastructure companies sell you more: more knobs, more power, more surface area. DeltaStream's entire proposition is that it takes things away - the cluster, the tuning, the specialist headcount, the second and third tool. In a field that loves complexity as a moat, choosing simplicity as the product is a real position. The best infrastructure is the kind you stop thinking about. DeltaStream is trying to make streaming data that quiet, and if it succeeds, most of its users will never know how much work went into the silence.