Technology

From left, Ahmed El Beheiry, CEO of Cassava AI, Ifeoma Jibunoh, CMO, Cassava Technologies, Emmanuel Lubanzadio, Africa lead, Open AI (Delicious Bokeh)

In 2025, there was little evidence of many AI workloads being run in datacentres anywhere in Africa. Cassava Technologies is betting that there is pent-up demand to run these workloads on the continent, and announced in March that it would deploy Nvidia accelerated computing and AI software using Nvidia Cloud Partner architecture at its datacentres in South Africa. The company has now fleshed out these plans, and said in November that it would be offering a GPU-as-a-Service to customers in Africa. It also announced a slew of partnerships with many of the world’s major AI companies in its ambition to become the gateway to AI services, and sees opportunities in agriculture, healthcare, academia, and for entrepreneurs. It has not yet announced pricing of workloads. According to a survey by Zindi, a professional network of 11 000 data scientists in Africa, only 1% has on-premises access to GPUs. Four percent of those polled can afford cloud GPU use, but can only pay $1 000 a month, which buys them two hours of daily usage on outdated Nvidia A100 GPUs. The remaining 95% of Zindi members use laptops without GPUs, or use Google Colab in the cloud.

Speaking at the Africa Tech Festival in Cape Town in November, Cassava founder and chairperson Strive Masiyiwa said the company had bought 12 000 Nvidia GPUs after meeting with Nvidia founder Jensen Huang. He added that the GPUs would be distributed to five sites on the continent. The Cape Town facility, which will apparently have 3 000 of the GPUs, will serve the southern part of the continent, and work will start soon on one in Egypt, followed by one each in Kenya, Morrocco and Nigeria. Masiyiwa said the GPUs were already being installed at its datacentre in Atlantic Hills outside Cape Town, and that it was onboarding customers. Huang is credited with coining the term AI factory, and Cassava says this is what it will be rolling out in Africa. An AI factory is distinct from a normal datacentre in that it is optimised for AI workloads, particularly around AI inference performance and energy efficiency.