What comes next.

The first two issues covered what exists. This one covers what does not exist yet, which is a harder thing to write honestly, because the temptation is to describe ambitions as though they were milestones.

So here is the rule we are applying: everything below is labelled by where it actually sits — built, funded, or contingent.

The ladder

Omnis progresses in deliberate stages. We do not skip them, because every failure at the smallest size costs hours and the same failure at the largest costs six figures.

Omnis Small 1.0 is the rung that matters. It is the first model we expect to put in front of real users on its own merits, and — unlike the stages above it — it is reachable with a funded round rather than a nine-figure one. Everything in our current plan is organised around getting there properly rather than getting there loudly.

The arithmetic, stated plainly

Training compute scales roughly with parameters multiplied by tokens. Run that honestly and the conclusion is uncomfortable but clarifying: matching a frontier general-purpose model head-on is a nine-figure problem before you count the research team. That door is closed to us. It is closed to almost everyone.

We would rather tell you which door is shut than let you assume we are walking through it.

The open door is narrower. A specialised model, trained on data that is not on the open web, serving a task where the frontier is structurally weak. Yorùbá, Igbo, Hausa and Pidgin. Local curricula, local law, local business norms. In-region latency. Data that never leaves the jurisdiction. Cost per token roughly two orders of magnitude below a frontier API. Offline operation on a mid-range phone.

None of those are marketing angles. Each is a measurable property that a well-scoped small model can hold and a general-purpose large one cannot.

Opening the infrastructure

Today our AI infrastructure serves our own products. The intention has always been that it serves other people's too.

The first piece is already out: the AfriKDP Publishing API is live, and external applications can upload books, run checkout and pay authors programmatically. It is the proof that we can expose our infrastructure cleanly to someone who does not work here.

What follows, in order:

• Omnis inference services — our own models behind an API, starting with the capabilities where our adapters already clear the bar.
• Translation and language services — the African-language capability, offered on its own, because it is the thing most often asked for and least often available.
• Automation and agent services — the orchestration layer underneath AfriAssist, made available to businesses that need the workflow rather than the interface.
• Platform APIs — marketplace, escrow and commerce primitives, so other builders extend our infrastructure instead of starting from zero.

Because every product already speaks to the gateway rather than to a model, exposing these externally is a packaging and access-control problem rather than an architectural one. That is the dividend of the decision described in Issue 01.

Global structure

AfriStack was founded in Africa because exceptional technology can originate anywhere. Our market was never limited to Africa, and the corporate structure is now catching up with that.

We are preparing for an international structure and expansion strategy designed to support global investment, international partnerships, payment infrastructure, enterprise relationships, U.S. market expansion, global talent and international customers.

Africa is where we started. The global market is where we are going. Both halves of that sentence are load-bearing — the African-language capability is not a stepping stone we intend to abandon once we have international customers. It is the thing we expect to be best in the world at.

What capital accelerates

AfriStack is preparing for its next stage of growth and strategic investment. To be specific about where it goes:

• Compute infrastructure — GPUs, servers, storage and networking. Our published model results were produced on a single consumer laptop GPU. This is the binding constraint, and it is the first one capital removes.
• AI research and development — larger models, specialised intelligence, training and evaluation.
• Engineering — AI researchers, software engineers, infrastructure engineers and product teams.
• Data — high-quality datasets, multilingual data and proprietary task data. This is the asset that cannot be bought off the open web, and it compounds.
• Product development — AfriKDP, Skilluxo, AfriAssist and the wider ecosystem.
• Global expansion — international markets, partnerships, enterprise customers and distribution.

Longer term, our vision includes dedicated AfriStack technology and AI infrastructure capable of supporting increasingly sophisticated model development — owned rather than rented.

The investment is structured as a revenue-participation programme with published terms, regular reporting, a full legal and financial data room, and every material risk disclosed alongside the opportunity. The complete Investor Package runs to nine volumes.

Who we want to hear from

Being specific is more useful than being welcoming, so:

• Compute partners. Training and inference capacity is the single biggest multiplier on this roadmap.
• Data and language partners. Publishers, universities, broadcasters, linguists and cultural institutions holding high-quality African-language material.
• AI researchers and engineers. Efficient training, tokenization for low-resource languages, evaluation, inference optimisation — real problems, real infrastructure.
• Enterprises and institutions that cannot let data leave the jurisdiction.
• Developers who would build on our APIs before they are generally available.
• Investors who want the measured version of this story rather than the pitch version.

We have already started. Now we are building the infrastructure to go much further.

That is the inaugural briefing. Future issues will be shorter and more frequent: what shipped, what the numbers said, and what changed our minds.