🤖 LAVA is, more than anything else, a bet on African talent. AI is the lever that lets that talent — systematically undervalued and overlooked — produce world-class products at global scale. We back tools that augment human intelligence, connection, and cultural practice, rather than tools that try to replace them.
The bet underneath the bet
It is not that there is more talent in Africa, or that the talent here is qualitatively different. It is that this talent has been systematically mispriced — undervalued by global capital, overlooked by global hiring markets, and under-served by global tooling. The best work is happening in places most people are not looking.
AI shifts that calculus. A small African team with great taste, real domain knowledge, and frontier tooling can now ship products competitive with much larger global teams — across content, software, finance, and operations. AI is the multiplier that lets African founders convert under-priced talent into globally legible outcomes. (See Great Stories Will Find a Way for the long version.)
This thesis sits alongside Simple Finance and Trust Infrastructure as a multiplier on both. We will occasionally step a little outside the strict boundaries of the first two when we find founders who exhibit the kind of world-class talent we know exists on the continent and who are using AI as genuine leverage.
Augmentation over displacement
We are far more bullish on AI tools that augment human intelligence, connection, and cultural practice than on tools that try to displace the need for thinking, reasoning, or consulting other people.
The most interesting results we have seen — in our portfolio and in our own use — are not “press a button, receive an artifact.” They are humans collaborating with models to refine work they already cared about: cleaning up a recording so it sounds professional, sharpening a piece of analysis, accelerating a diligence cycle, drafting a memo from a real conversation, translating a story across languages and registers without losing its texture.
In a world where capabilities advance every month, models are not a moat. The moats are:
- Data the world doesn’t have. Africa’s linguistic, cultural, financial, and operational data are dramatically under-represented in frontier training sets. Founders who can ethically generate, license, or accumulate this data — and use it to fine-tune, ground, or evaluate models — sit on something genuinely defensible.
- Distribution and embeddedness. Real product surface area in front of real users. Embedded inside payment flows, content platforms, business operations, customer support — places where the AI is invisible because it just works.
- Taste and craft. The judgement to know when a model is right, when it is hallucinating, and when the human still needs to do the hard part. This is downstream of culture and lived experience, and it is exactly where African founders are under-priced.
Why data is the real frontier
The Chinchilla paper made it clear in 2022 that data — not parameters — is the binding constraint on model quality. Since then, the industry has improved programming, mathematical, and scientific reasoning largely by feeding models synthetic training sets generated by previous models. That works well in the “hard” sciences, where verification is cheap.
It does not work as well in the social and cultural domains — storytelling, persuasion, negotiation, taste, craft — which, as E. O. Wilson argued in Consilience, are the hardest things to do well. Our view is that telling great stories — and operating skillfully in human contexts — matters more for widespread, economically valuable AI use than novel mathematical proofs do.
If models are to do more than code and physics, the world needs better data: data that reflects how humans speak, trade, argue, and decide, in the languages and contexts where the bulk of humanity actually lives. We think African founders are uniquely placed to contribute here, and we want to back the ones already doing the work.
Case study: Jamit and AI-native storytelling
Jamit — built by Ikenna Orizu, formerly engineering lead at Roku and NewsCorp — is the cleanest articulation of this thesis we have backed so far.
Jamit is an AI-native storytelling platform powered by a custom model called Magic Producer. With one click, anyone can produce studio-quality audio stories and recordings, and at the same time scale their original IP globally. Think “webtoons for audio,” powered by intuitive tools that understand how a creator speaks and how to make their work captivating to an audience.
Three things make Jamit a clean expression of AI as leverage:
- Augmentation, not replacement. Magic Producer can generate a story from a prompt, but the strongest results come from cleaning up recordings of human speech to make them sound professional. The model does the last 10% of polish that normally takes 90% of the effort. The story remains the human’s.
- Data is the moat. Story-telling has deep roots in African soil — arguably an area where the continent has both qualitatively better and quantitatively more talent than anywhere else on the planet, but where that talent has historically been hard to capture and scale. Jamit converts that latent talent into structured, multi-language audio data that improves the model — and is hard for any global lab to replicate.
- Distribution flywheels via Originals. Ike had a front-row seat as Netflix Originals was built after Disney pulled their content. Jamit Originals — long-form fictional audio with well-known voices — extend reach beyond the app, license well into commute and ambient-listening markets (especially Indian and Asian markets), and pull people back into the platform’s user-generated, studio-quality catalogue, with web3-powered global payments to creators built in.
With over 500M global podcast listeners and double-digit annual ad growth, the opportunity for new storytelling infrastructure is enormous. Walter Ong called the communitarian effects of electronic media secondary orality; Jamit turbocharges those effects with augmentation tools and crypto-abstracted financial rails — making the boring stuff (payments, subscriptions, profit shares, listener loyalty rewards) easy for anyone, anywhere.
Where else this thesis applies
Beyond storytelling, we are interested in AI-as-leverage across:
- Verification — this is where cryptography and AI (and potentially cryptocurrency) actually intersect. We’re interested in tools that go beyond “Authorize my agent to send 100 USDC to 0xdeadbeef”, which can actually verify that an agent acted well under uncertainty and ambiguity, with cryptographically-assured reasoning and execution traces.
- Wisdom — which is not a property of any given LLM, but rather exists at the intersection of state of the art systems and the humans who use them. We’re looking for people that are not optimising for productivity gains, but rather for wise use that builds long-term capability rather than short-term profit.
- Creative and cultural production — music, audio, video, and writing tools that augment African creators and protect the IP they generate.
- Education and training — tutors and copilots tuned to African curricula, languages, and economic realities.
- Verticalised models and evaluation — small, focused models and rigorous evals tied to specific African workflows where general-purpose frontier models underperform.
What we look for in founders
- A specific, defensible relationship to data: original collection, hard-won partnerships, or a product that compounds proprietary signal over time.
- Real product taste — an answer to “why is the AI doing this and not the human?” that is not just “to save cost.”
- Discipline about model dependence: assume the underlying model gets better and cheaper every quarter, and design for it.
- A clear, opinionated view of who the AI is for — and a willingness to deeply embed in that user’s workflow, language, and context.
- Maturity about the social and cultural stakes of deploying AI in African contexts, including labour, language, and data sovereignty.
Out of scope
- Wrappers around frontier models with no proprietary data, distribution, or workflow embedding.
- “AI for Africa” narratives that are really just localisation of off-the-shelf US products.
- Tools that explicitly aim to displace human judgement in high-stakes social or financial decisions without a clear governance and accountability story.
- Speculative “AI + crypto” combinations where neither side meaningfully strengthens the other.
We are looking for founders who can use frontier technologies to amplify what humans on this continent have been doing since time immemorial — and to make that work legible, valuable, and durable to the rest of the world. From Africa, for the world.