LEGO for AI

Hi there,

Welcome to the 59th edition of Next Capital, where we help you find Africa’s most promising startups, before they get big.

In our last edition, we wrote about the startup building Africa’s first full-stack film factory. If you missed it, you could catch up here 👇🏾

Next Capital is launching a podcast, and we need your brain for it.

We want you to send us your ideas.

The thing you’ve been sitting on. The gap you spotted in a market. The problem you’ve experienced one too many times and thought, “Why hasn’t anyone built this yet?”

Send it in, and we might review it on the pod.

And here’s a little secret: we’re creating something that’s going to let you build your ideas in real life.

Drop your idea here → Next Ideas Pod!

P.S. - Know someone who always has startup ideas? Forward this email to them!

Startup: Cencori
Location: Nigeria

Daniel Oreofe and Bola Banjo are two co-founders who wanted to move to San Francisco.

Everyone building AI seems to move to San Francisco. Then they read the fine print. As of January 2026, Nigeria sits on a US partial visa suspension, which means new business, tourist and student visas are mostly off the table for Nigerians without an existing stamp. The SF plan died.

But that didn’t stop them from building. This time, they didn’t build for the US or Nigeria - but for the internet itself.

Here is a thing about the internet. It runs on a handful of giant clouds. AWS, Azure, Google Cloud. Nobody spins up a web app by buying their own servers anymore. They rent them off these cloud providers and pay a monthly bill.

AI is different. There is no clean, default “cloud for intelligence” that everyone reaches for. And this is because the components that make up a state-of-the-art AI stack are still being built. As a result, only the big labs can afford to have them all in one place. Everyone else has to stitch many vendors together into a working system.

Africa is particularly disadvantaged in this regard. Recently, Strive Masiyiwa, founder of Econet, an African telco, mentioned a visit to NVIDIA where he learned that there were only 80 GPUs in all of Africa. In this context, building fiefdoms of servers, data centers, proprietary code, and virtualization from scratch is pretty expensive for Africa, and many companies globally.

For these people, Daniel and Bola built Cencori…

The easiest way to picture Cencori is as a lego kit with building blocks that come in three shapes.

The first shape is the gateway. This allows devs use many models via one API. Your app calls the model, and if it fails, traffic automatically routes to another model before the user notices. This process also comes with a security pipeline bolted on.

If that sounds familiar, it should. It is close to what OpenRouter does, and OpenRouter just raised $113 million at a $1.3 billion valuation doing exactly this.

The second shape is compute. This is where companies and labs can train and fine-tune their models with all the hardware stuff being virtualized. Cencori thinks that as more African countries look to build AI for local use cases, virtual infrastructure like this can drop overhead costs.

The final shape is memory. AI is great at making output but bad at recalling it. So, Cencori is building memory layers for AI, allowing decs build products that remember things about their users across sessions, instead of meeting everyone as a stranger every single time.

It’s a hot category in itself. Mem0, the best-known memory startup, has raised about $24 million and crossed 50,000 GitHub stars.

Cencori has a demo that’s been pulling users in fast. They showcase a feature that lets you deploy an AI app in roughly three minutes. Cencori’s goal is clear. They want to be the infrastructure for every AI tool/software built.

They’re selling a simple kit that gives you security, routing, compute, and memory, all in one place.

Get in front of 11,000 investors, founders, execs, and tech enthusiasts in Africa.

Sponsor an edition

Cencori has three revenue streams:

First, compute. This is done with enterprise pricing, based on usage and nature of training being done. It’s the heaviest and most expensive on the menu.

Second, the platform. The gateway and the memory features come at a $49 a month, or pay-as-you-go if you would rather only pay for what you burn.

Third, a slice of subscription fees for AI products. Cencori sits between your app and the AI model, and takes a small cut based on what you charge. Because every AI request flows through it, it can see exactly how much intelligence your product is consuming, and it skims 0.5% to 1% of that.

On traction, the headline is that three-minute app builder.

It pulled in 1,600 SDK downloads in seven hours.

Then there’s the cost reality, which the founders were refreshingly blunt about. All of this currently runs on AWS. Cencori rents the machines it resells. The plan is to move to its own bare-metal setup, with in-house GPUs, as it scales. That is the right long-term move but a brutally expensive one too.

Three things are true at the same time.

First, every new shape Cencori is building is its own venture-funded gold rush.

The AI infrastructure market sits around $143 billion in 2026 and is forecast to clear $900 billion by 2035. The “neocloud” slice alone, companies renting out GPUs, is scaling from under $10 billion in 2025 toward roughly $186 billion by 2030. The gateway layer has OpenRouter at $1.3 billion. The memory layer has Mem0. Cencori is reaching for all three at once. That is plenty of ambition, but also a lot of surface area for a two-person team.

Second, the African compute story is real, and it is getting crowded fast. The exact gap Cencori is pointing at, no chips, no local model training, has already drawn the continent’s heaviest hitter.

Cassava Technologies, Strive Masiyiwa’s group, signed a $700 million deal with Nvidia to build “AI factories” across South Africa, Nigeria, Kenya, Egypt and Morocco. It has already switched on GPU-as-a-service and a multi-model exchange aimed at African developers, with plans to tune models for Swahili, Zulu and Afrikaans. That is, more or less, Cencori’s whole thesis, backed by hundreds of millions of dollars and actual hardware already in the ground.

Third, this is a market where posture matters as much as the product. Cassava sells raw horsepower to enterprises and governments. The hyperscalers sell everything to everyone. Cencori’s bet is the friendly layer in the middle: the bundled, developer-first platform that hides the messy hardware and lets a small African team ship a real AI product without standing up a DevOps department.

The gateway and the scanner are live and pulling downloads today. Compute and memory are still promises with a Q4 sticky note on them.

So here is the question we are sitting with. Can a two-person team in Lagos, renting AWS today, become the default place African developers build AI, before a $700 million Cassava or a $1.3 billion OpenRouter simply swallows the same idea?

What do you reckon: does Cencori have a good shot?

Want to be featured in Next Capital? Drop us a line.

Until next week! 🫡

We’re always coordinating participation across active opportunities within the Next Capital pipeline, particularly for investors who can back startups.

If you’d like to be considered for current or upcoming opportunities, let us know, and we’ll follow up where there’s alignment.

Enjoyed reading this? Forward it to someone you think would enjoy it as well.