BantuNomics is 3Mega.ai's commercial infrastructure for the Bantu language family (spoken by ~400 million people): a complete, machine-readable substrate of 459 Released Full Syllable Inventories, native-speaker recordings, and standards documents — licensed to AI labs through four independent doors (a free public test, a free self-serve evaluation, a 75-day validation pilot that is fully creditable, and a full annual subscription), and used the way English-language stacks use UTF-8.
BantuNomics treats Bantu phonotactics the way standards organisations treat UTF-8, ICANN treats DNS, and the Periodic Table treats the elements: as a closed, enumerable, indexed reference that every downstream operation queries but never mutates. The platform ships 459 Released Full Syllable Inventories (FSIs), plus the audio substrate that makes them decidable, plus the standards spine — BTS-S100 for document conformance, BTS-API-100 for API + MCP conformance — that lets a procurement evaluator audit the deliverable without trusting our marketing.
The failure mode is structural, not statistical. Frontier models trained on text scraped from the web are missing the layer of Bantu that decides meaning — tone, vowel length, downstep, phrase-boundary, voice quality. They tokenise Bantu with BPE while every Bantu child has internalised a syllabary. On L26, the operating-alphabet benchmark, every frontier model tested scores 26 of 26 on the English alphabet and not one passes the blind Bantu bar. BantuNomics is the substrate that closes the gap.
Munyambala's long-form essays — written for BantuNomics, canonical at 3Mega — set out the diagnosis ("the Flat Text Problem"), the architectural framing ("FSIs are infrastructure"), the pedagogical reality ("how a Bantu child learns to read"), and the phonology that ties them together ("tone lives in the syllable").
When an AI lab encounters a Full Syllable Inventory for the first time, the question they ask is "what is the licensing fee." That is the wrong frame. FSIs are not a dataset you train on. They are the substrate every downstream operation references — closed, enumerable, standardised. Subscribe accordingly.
The syllable is the unit. Not the letter. By age 6 a Bantu child has internalised the syllabary of their language — the same inventory a frontier LLM in 2026 cannot enumerate. One layer off, and the error propagates through every prediction.
Tone in a Bantu language is not decoration — it is meaning. And tone does not attach to the letter or the morpheme. It attaches to the syllable. If a frontier model cannot enumerate the syllables of a Bantu language, it cannot understand its tone.
Bantu languages have been spoken for thousands of years. The writing system is barely a hundred years old — and it leaves out the part that decides meaning. We named this gap so the AI labs could see it.
BantuNomics ships the substrate frontier AI labs subscribe to. amina is where it comes from — the contributor-facing recording platform where native Bantu-speaking communities contribute paid, consented audio that becomes the licensed corpus. Same operating entity. Same database. Same schema. Same speakers. Subscriptions fund recording; recordings deepen the substrate.