Confluent

Right now, somewhere, a river of events is being read.

A bank approves a card swipe before the customer pockets their wallet. A retailer reroutes a warehouse order while the shopper is still on the page. A streaming service notices you paused on episode three and queues a nudge. None of these things wait for a nightly report. They happen in the moment, on data that is moving, and a large share of them move through software stamped with one name: Confluent.

Confluent is, in the plainest terms, the data streaming company. It sells the commercial version of Apache Kafka - the open-source system that quietly became the nervous system of modern software - plus the tools to process, govern, and connect everything flowing through it. The company is headquartered in Mountain View, employs roughly 3,300 people, trades publicly as CFLT, and in December 2025 agreed to be acquired by IBM in a deal valuing it near $11 billion. Not bad for a project that started as an internal headache.

Most software treats data like a warehouse. The world stopped behaving that way.

For decades, the deal was simple. You collected data, you put it somewhere, and later - at a polite hour, in a batch - you asked it questions. This worked beautifully for a world that ran on yesterday. It works poorly for a world that runs on the next four seconds.

The trouble is that a business is not a filing cabinet. It is a constant stream of things happening: clicks, payments, shipments, sensor pings, logins, cancellations. Storing all of that and querying it the next morning is a bit like reading the newspaper to find out whether your house is currently on fire. Technically informative. Tactically late.

The engineers who would go on to found Confluent saw this gap from the inside. At LinkedIn in the early 2010s, they were drowning in exactly this problem: dozens of systems that all needed to talk to each other in real time, and no clean way to let them. Every new connection was a custom pipe. The pipes multiplied. The whole thing threatened to become an unmaintainable knot.

Three engineers wrote Kafka. Then they bet a company that the world would need it.

Jay Kreps, Neha Narkhede, and Jun Rao built Apache Kafka at LinkedIn to solve their own knot - a single, durable, high-throughput log that any system could write to and any system could read from. Instead of every service wiring directly to every other, they all spoke to one stream. They open-sourced it. It spread. Other companies discovered they had the same knot.

In 2014 the three left to start Confluent. The bet was specific and slightly contrarian: that streaming was not a niche tool for a few data-heavy giants, but a default way that all software would eventually be built. The open-source project would stay free. The company would make its money making Kafka something a normal enterprise could actually run without staffing a small monastery of specialists.

There is a small literary footnote here. Kafka the system is named after Kafka the writer, because Kreps liked the idea of a platform optimized for writing. It is the rare piece of infrastructure with a joke buried in its name - and the rare joke that turned into a multibillion-dollar business.

One platform for data that refuses to hold still

Confluent's offering is best understood as Kafka plus everything you wish came with Kafka. The raw open-source project is powerful and famously demanding to operate. Confluent's job is to remove the parts that make engineers reach for the aspirin.

FIG. 1 — Four products, one stubborn idea: the data should keep moving and you should still know exactly what's in it.

The numbers that turned a side project into a public company

Skepticism is fair. Plenty of open-source companies struggle to convince anyone to pay for what they can download for free. Confluent's answer is its customer base: thousands of organizations, including a large share of the Fortune 500, across financial services, retail, healthcare, manufacturing, and telecom.

FIG. 2 — The receipts. Approximate figures drawn from public filings and announcements through late 2025.

Behind those bars is the strategic shift that mattered most. Confluent started as a software vendor. It became a cloud company. The managed service - the part where Confluent runs the hard infrastructure so customers don't - now grows faster than everything else, with managed Apache Flink revenue tripling over two recent quarters.

It also explains why IBM came knocking. As AI systems grow hungrier for fresh, governed, real-time data, the plumbing that delivers it stops being a back-office concern and starts being strategic. IBM's December 2025 agreement frames the logic directly: combine Confluent's streaming backbone with IBM's AI, automation, data, and consulting reach.

Put streaming at the center, not the edge

Confluent's stated aim is to put a data streaming platform at the heart of every company - not as an optional add-on bolted onto a batch world, but as the central nervous system the rest of the architecture organizes around. It is a bigger claim than "we sell Kafka support." It is a claim about how software gets built.

The culture grew out of open source, which tends to produce companies that are engineering-led and allergic to gatekeeping. Confluent keeps the core project free and bets that being genuinely useful to developers is the most durable moat there is. So far the developers have agreed.

The next four seconds keep getting more expensive to miss

Every trend pointing forward - real-time fraud detection, event-driven microservices, AI agents that need to react to the world as it changes - increases the cost of being late. Batch was fine when "this morning's data" was good enough. Increasingly, it isn't. The systems that win are the ones that notice and respond while the moment is still open.

That is the bet Confluent placed in 2014 and rode to an IPO and an $11 billion exit. Whether it lives inside IBM or stands alone, the underlying idea has already won: software now assumes data is in motion.

So go back to the opening. The bank approving your card swipe before you pocket your wallet. The retailer rerouting your order mid-session. The streaming service catching your hesitation on episode three. A decade ago, most of those would have been a report you read tomorrow. Today they are decisions made while you blink - on data that never stopped moving, through software a few LinkedIn engineers built because they were tired of the knot. The river was always there. Confluent just decided someone should read it in real time.