Redpanda Data

Kafka won the war. Then it became the war.

For most of the 2010s, Apache Kafka was a quiet revolution. It gave engineers a durable, scalable log they could throw events at and read back in order, anywhere. Banks adopted it. Retailers adopted it. The data-streaming category, as analysts would eventually call it, was effectively a Kafka category. Compatibility with the Kafka API became table stakes for any new piece of infrastructure that wanted to live near production.

The trouble is that Kafka was a 2011 design. It runs on the JVM, leans on a separate consensus service in ZooKeeper, and asks operators to learn an extensive vocabulary of partitions, replicas, brokers, controllers, ISRs and garbage collection tunables. Companies hired entire teams whose job description, more or less, was "keep Kafka alive." Even modest clusters were known to produce remarkable cloud bills.

So a peculiar thing happened: the technology that promised real-time freedom became, in many shops, the slowest thing in the building to operate. Everyone wanted the Kafka API. Almost nobody enjoyed running the Kafka cluster.

Translation: keep the dialect, rip out the engine.

Alex Gallego had already built one of these. He thought he could do better.

Alexander Gallego is a Colombian immigrant who attended NYU Tandon and spent the back half of the 2010s building a distributed stream processing engine called Concord Systems. Akamai acquired Concord in 2016. Most engineers in that situation would take a senior staff role at a large company and call it a decade.

Gallego instead spent the next three years thinking about what he would build if he started over with the benefit of hindsight, modern hardware, and the brutal honesty of having operated his own streaming engine in production. The thesis crystallized in 2019. In January 2020, a then-quiet company called Vectorized raised a $3 million seed round from Lightspeed Venture Partners. The first lines of code were written from an apartment in Miami.

The bet was specific. Speak fluent Kafka at the wire. Throw away everything underneath. Use C++. Use the Seastar-style thread-per-core architecture that powered ScyllaDB. Use Raft for consensus and eliminate ZooKeeper. Treat the JVM as a tax that customers had unknowingly been paying for ten years. And, because branding still matters, give the project a name people actually want to say. The internal name, Vectorized, lost a customer survey to the name of the product itself: Redpanda. The company renamed.

A rare instance of marketing winning a fight against engineering, fair and square.

What you actually get when you install it.

Redpanda ships as a single binary. There is no separate ZooKeeper to deploy and no JVM to tune. Each node speaks the Kafka protocol natively, so the libraries and connectors that already exist in your codebase keep working without modification. Tiered storage lets clusters offload cold data to S3, GCS or Azure Blob, which is the kind of feature operations teams quietly cheer about once the invoice arrives.

Around the core platform there is a small constellation of tools. Redpanda Console is a web UI for browsing topics, debugging messages and managing schemas. Redpanda Connect, the descendant of the Benthos project that Redpanda acquired in 2023, is a declarative stream processor with hundreds of connectors and, increasingly, processors designed for generative-AI pipelines. Redpanda Cloud is the managed offering on AWS, GCP and Azure. For regulated industries that prefer to keep data inside their own VPC, there is BYOC.

Four products. One thesis: keep the Kafka contract, kill the operational drag.

The short version, in chronological order.

Quiet years, loud years. The pattern most infrastructure companies share.

Money in, money up.

Investor enthusiasm is not the same as customer enthusiasm. But in infrastructure, the two tend to rhyme. Redpanda has now raised more than $265 million across five rounds. GV has led every round since the Series B. Lightspeed has been in since the very first cheque. The Series D pulled the company past the $1 billion line.

The customer roster, as far as Redpanda is willing to disclose it, includes Activision Blizzard, Cisco, Moody's, Texas Instruments and Vodafone. Collectively the company says its customers move hundreds of terabytes of real-time data per day across its clusters. None of that is a guarantee of future revenue. It is, however, a strong hint that the operational pitch is landing.

Reading a customer logo wall is a bit like reading tea leaves. Sometimes the leaves agree.

Make real-time the default, not the upgrade.

If you take Redpanda at its word, the company is not trying to be the cheaper Kafka. It is trying to make event-driven architecture cheap enough and simple enough that more software can be built that way in the first place. That is a different claim. It implies that for the past decade, latency, cost and operational drag have been quietly limiting how event-driven the world's software actually became.

The Agentic Data Plane announcement in April 2025 made the bet explicit. As AI agents move from chat demos into production workflows, they need a low-latency substrate to read shared context, fire off actions, and react to one another. That substrate looks suspiciously like a streaming platform. Redpanda's view, naturally, is that it should be theirs.

The company is fully remote, with engineering hubs of gravity in San Francisco and across Europe. Around 180 people. Open source by gravity rather than by slogan - the main repository sits in public on GitHub, alongside more than a hundred sibling projects.

Why a faster Kafka matters more in 2026 than it did in 2019.

Two trends are squeezing the data plane from opposite sides. AI agents are sending a torrent of small, low-latency events between models, tools and human inboxes. And FinOps teams, finally awake to the bill, are asking pointed questions about cloud spend in a way they were not five years ago. Streaming infrastructure sits exactly where those trends collide.

Redpanda's pitch happens to align with both. Less hardware, less staff, less ceremony, more throughput. Agentic AI workloads that would have buckled a traditional Kafka deployment now have somewhere to land. It is not a guarantee of category dominance - Confluent is a public company, AWS sells MSK at scale, AutoMQ and WarpStream are circling the same problem from different angles. But it is a clear lane.

Back to the opening scene. Somewhere right now, a panda is moving your money. It is not making a fuss about it. That is the point. Redpanda Data spent six years building a piece of infrastructure that, when it works, you forget is there. Most of its largest customers will keep declining to name it publicly. That, too, is part of the brand. The quiet ones are usually the ones moving the most.

In infrastructure, fame is failure. Anonymity is uptime.

Where to go from here.