Kong.

The traffic cop you never see.

Somewhere in a Cargill data center, a tractor in Iowa sends a soil-moisture reading toward a forecasting service. Somewhere in São Paulo, a Sky subscriber taps play on a stream. Somewhere in a developer's terminal in Lahore, a Python script asks an LLM to summarize a quarterly report. Three completely different requests, three completely different products. All of them pass through the same kind of door on the way in - a piece of open-source software called Kong.

You have probably never typed the word Kong into a URL bar. That is the point. Kong is plumbing - the kind of plumbing that makes other people's products feel fast and modern. In 2025 the company decided it was tired of being invisible and rebranded itself, with a perfectly straight face, as the AI connectivity company. Which is a polite way of saying: the same gateway that has been frisking your REST calls for a decade now wants to frisk your LLM calls too.

In November 2024 investors agreed loudly enough to be heard from Sand Hill Road. Tiger Global and Balderton co-led a $175 million Series E, valuing Kong at $2 billion, almost 45% above the prior round. The press release was the kind of crisp, slightly smug document that companies issue when the cap table is full and the runway is long.

Software stopped being one thing.

For most of the last twenty years, a "product" was a thing you shipped. Today it is a constellation. A modern company runs hundreds of microservices, dozens of SaaS tools, a handful of clouds, and now a growing zoo of AI models - and somehow every piece needs to talk to every other piece, securely, quickly, and without falling over on a Tuesday.

That conversation is what we now call an API call. Multiply it by a million-per-second and you get the actual operating system of the modern enterprise: a tangle of contracts, keys, quotas, and quietly broken promises. Someone has to direct that traffic. Someone has to authenticate it, rate-limit it, log it, encrypt it, retry it, observe it, and - increasingly - decide whether the AI on the other end should even be allowed to answer.

Kong's bet is straightforward and slightly impudent: the company that owns the gateway owns the connectivity layer, and the connectivity layer is becoming more important than the things connected to it. It is the kind of bet that sounds obvious only after it has paid off.

From a Milan side project to a San Francisco standard.

Augusto Marietti, Marco Palladino, and Michele Zonca did not set out to build infrastructure. They built a marketplace. The original 2009 project, called Mashape, was an attempt to make APIs as easy to buy as apps in an app store. It was charming, ambitious, and - in the way of marketplaces - mostly early.

What they kept noticing, though, was that customers cared less about the marketplace and more about the small gateway sitting in front of it. The gateway was scrappy, fast, written on top of Nginx, and it solved a problem the marketplace barely scratched at: keeping API traffic alive at scale. In 2015 they open-sourced it as Kong. In 2017, they did the thing few founders ever do: they sold the original company's assets, dropped the original name, and became a pure-play infrastructure business.

One gateway, a thousand plugins, several disguises.

Asking what Kong "does" is like asking what a power strip does. The short answer is: it lets other things plug in. The longer answer is a product line.

Boring infrastructure, exciting numbers.

Cargill - 150,000 employees, over a third of the world's food supply on its books - reportedly used Kong on AWS to deploy more than 450 new digital services in six months, with deployment times up to 65x faster. Sky Italia used Kong to lift performance by 70% while holding 99.99% availability, which is the kind of statistic that gets quietly framed in a CTO's office.

The customer roster reads like a tour of the global economy: GE, Honeywell, SkyScanner, Sky Brazil, FabFitFun, Yahoo Japan, plus 600+ other enterprise logos. The ARR is a reported ~$146 million. The OSS gateway has been downloaded somewhere north of "we stopped counting."

Make connectivity invisible. Then keep going.

Read Kong's official mission and you will find the standard phrasing about enabling companies to become "API-first" and accelerating digital innovation. Read between the lines and the actual ambition is simpler. Kong wants the same default-status that databases have. You do not pick a database for a side project - you reach for Postgres. Kong wants to be the Postgres of connectivity.

It is a long bet. There are real competitors: Google's Apigee, Salesforce's MuleSoft, AWS API Gateway, Azure API Management, the open-source Istio mesh, the upstart Tyk. None of them, however, are organized around the same idea - that one connectivity layer should sit cleanly between humans, services, and now models, on every cloud, with the same plugin model.

The AI gateway question.

Here is the bet that justifies the rebrand. Every company experimenting with AI right now is quietly building the same thing in a corner: a tiny middleware layer that decides which model to call, caches responses, enforces a budget, redacts PII, logs prompts, and handles failover when OpenAI has a bad afternoon. It is, by any honest reading, a gateway. The only question is whether each company writes its own, badly, or buys a real one.

Kong's pitch is that they already built the gateway pattern once, for APIs, and they are extremely happy to do it again for LLM traffic. The technology is the same shape. The risks - cost overruns, prompt injection, governance - are the same shape. The customers, in many cases, are already on the phone.

If Kong is right, the AI boom does not need a thousand custom proxies; it needs one boring, reliable, observable connectivity layer that sits in the middle of everything and minds its business. If Kong is wrong, the company still owns the gateway market it has spent fifteen years building. Either way, the door is theirs.

Back to the tractor.

The Iowa tractor sends its soil reading, the São Paulo stream starts, the Lahore developer asks the LLM to summarize a report. Three years ago, exactly one of those requests would have passed through a tool like Kong. Today, all three do. In another three years, if the company gets what it is reaching for, you will not have to think about it - which is exactly the kind of success an infrastructure company is allowed to want.

That is Kong's argument, anyway. The traffic keeps coming. The door is still there. Somebody has to direct it.

Links, channels, primary sources.

Watch & listen.