$3M+ · AI + Telecom

What they're building

AethexAI is not wrapping GPT-4 and calling it a voice agent. They built their own model stack from scratch. The reason is technical and specific: most voice AI platforms rely on large language models hosted on high-end GPU infrastructure in North America or Europe, and for users in Africa and the Middle East, that geographic distance introduces noticeable latency and jitter. Instead of using existing orchestration tools like Vapi or LiveKit, the company built its own small models and orchestration layer from scratch. The Kora series, ranging from 300 million to 1.7 billion parameters, is designed to run efficiently on local infrastructure while maintaining accuracy across English, French, and Arabic dialects spoken in the region. The platform has three layers. The model layer (Kora 1) covers 100+ voices across markets, handles dialect-aware routing, code-switching, and performs at under 500ms streaming latency. The enterprise platform layer gives businesses an Agent Studio (no-code conversation flow designer), a simulation environment for testing against real call scenarios, a workflow engine that reads and writes data inside existing systems, and call-level analytics. The infrastructure layer handles telephony via channel partnerships with telecom providers across the region, keeping calls on local networks rather than routing through US or EU infrastructure. Common applications include debt collection, customer activation, and Know Your Customer (KYC) verification for banks and telecoms.

Why this matters

Voice remains a primary channel for enterprise customer interaction across emerging markets, and while many companies have already experimented with voice AI, most solutions have failed to perform reliably in production. The specific failure modes are documented from the founders' fieldwork. In Egypt, a call center automated a significant share of its calls, but rolled the system back because of poor results. Several support centers in Africa told them that finding and hiring engineers to automate calls at the right cost was a persistent headache. The problem was not demand. It was infrastructure. Walter Badoo, co-founder and managing partner of lead investor 4DX Ventures, argued that enterprises in Africa and the Middle East process roughly three times the call volume of their Western counterparts, with voice remaining the dominant customer interaction channel. ElevenLabs raised $500M at an $11 billion valuation in February 2026. Its focus is speech synthesis and conversational AI for Western enterprise markets with premium pricing. AethexAI is initially targeting a market of 1.5 billion people across Africa and the Middle East, where global providers have yet to deliver at scale. The data strategy is the most operationally interesting detail in the business. Rather than chasing the largest possible models, they decided small models are enough to tackle the latency problem while maintaining accuracy. To train these models, the startup used anonymized recordings from a call center partner. It also shipped hard drives to radio stations across Africa to collect more audio data. To keep costs down, it built a contributor network of university students to annotate data and pronounce local names.