Enza Capital launches dedicated AI fund for African startups, targets constraint advantage
Enza Capital has launched a dedicated fund for AI and AI-enabled startups across Africa, arguing that specialist funds outperform generalists as the continent's venture market matures. The firm has raised $58 million across its first two funds, deployed capital into 47 companies across nine African countries since 2019, and now counts 11 AI-native investments with a 12th in progress.
The fund backs what managing partner Mike Mompi calls the "constraint advantage" — products built for Africa's limits (patchy connectivity, multilingual voice interactions, limited compute) that can scale globally. Portfolio companies include Hakimu AI, which trains legal models on hundreds of thousands of unpublished East African Supreme Court cases; Aethex AI, building Pidgin and code-switching voice models for West Africa; WideBot, serving 14 governments and 400+ businesses with dialectal Arabic models; and Guidewheel, whose factory optimization tech deployed first in Kenya now scales across North America.
Enza screens for three moats at the application layer: proprietary data no one else has, existing distribution into customer workflows, or product integration into daily work. They demand high margins or a clear path to them, noting AI inference costs can double and kill businesses built on cheap-compute assumptions. The firm runs an office of the CTO led by Sir John Lazar (ex-Metaswitch, sold to Microsoft) and sponsors AI Tinkerers communities in Lagos, Nairobi, Cairo, and Johannesburg.
Africa captured just over 5% of global AI venture funding in 2025 versus 60% globally — a gap Enza views as the opportunity. Their first AI exit came in 2025 when Quro Medical, using ML to cut hospital stays from ten days to two, was acquired by South Africa's largest listed hospital group.
If you're building AI for African conditions, does your moat come from unique local data, existing customer relationships, or workflow integration — and can your margins survive if inference costs double?