Open source models could light African AI fire

The wall fan cooled hall at ADC's CPT 1 facility in Cape Town, with a graphic of Africa overlaid. Source: Lindsey Schutters.

“Beyond finance, AI-powered platforms like farmer.chat and PROMPTS are empowering farmers and improving maternal healthcare, showcasing the transformative potential of AI across various sectors on the continent,” Smith continues.

Financial services and banks have been quick out the blocks to grab an early technological advantage in the arms race.

Standard Bank took its first strides partnering with AWS and Microsoft Azure to leverage cloud.

Optiva is taking Gemini-powered agentic AI-powered BSS to MWC

3 days

“Our focus on digital and cloud-based solutions has allowed us to reduce investment in on-premises infrastructure and lower capital expenditure,” Standard Bank CIO Jorg Fischer when the bank made its cloud play.

“This transition is crucial for us to remain competitive and meet the evolving needs of our customers.”

Ultimately that was all a bluff as a well-placed source told Bizcommunity that the bank added to its R11-billion technology investment bill to secure high end AI hardware to help build out its services.

Standard Bank declined to comment on the hardware purchase rumour.

The hardware bubble popped

The hardware game has since been shaken by the advent of cheaper AI development methods, epitomised by the stock market impact of China’s DeepSeek.

Nathan Lambert, research scientist at the Allen Institute for AI (Ai2) explained on the Lex Fridman Podcast that DeepSeek’s ability to train large language models (LLMs) with limited access to advanced GPUs by optimising communication between GPUs – even going below the level of Nvidia's proprietary CUDA libraries – shows that innovation in AI doesn't always require the latest and most powerful chips.

"AI has become a bit of a black box for people, much like computers were a black box to users, in the late 80s and early 90s,” says Smith about the misconceptions in the African business market.

“In some cases, AI is being perceived as a magic wand that can solve all problems.”

Smith sees the rise of DeepSeek as a reset button that has levelled the playing field for African AI development.

New rules of engagement

Heath Huxtable, CEO of Microsoft specialist partner Braintree, shares this view and is advocating for corporates to get out of the arms race and be more considered in their approach.

“Technology adoption, even for transformative AI tools, needs to be grounded in tangible ROI. And let’s be clear, ROI is not just about the money. There are also intrinsic benefits that can be measured, such as efficiencies, performance, sales and so forth,” he says.

South Africa has only recently (November 2024) received the data centre upgrades needed to power the AI compute revolution when Africa Data Centres added 1,000 racks of white space to its CPT 1 facility in Cape Town.

The expansion added three state-of-the-art halls and another 6MW of critical IT load alongside a first for Africa fan wall cooling unit.

Electrons to feed the energy hungry “most efficient and sustainable data centres ever built in South Africa” comes courtesy of a 12MW solar farm in the Free State.

An African AI dream

AI hardware reaching our shores, combined with the open source ingenuity suggests that African companies may not be bound by the same technological limitations as before.

Coupled with the understanding of the need for data centres to use resources responsibly, Smith believes there's potential for the development of bespoke AI solutions tailored to the continent's unique challenges and needs.

But there is a catch. “Policymakers need to create regulations that promote responsible AI innovation while safeguarding ethical considerations, data privacy, and security,” she says.