Dear Minister Malatsi,
I write not as a corporate spokesperson but as someone who has spent ten years investing cheques into South Africa’s AI sector and living with the outcomes. My experience includes mentoring entrepreneurs, injecting millions of rand into AI start‑ups, and seeing two exits while three ventures failed. Those failures taught me that the true killers of AI businesses here are unaffordable compute, unreliable electricity, a thin venture‑fund pipeline, and a regulatory landscape that talks loudly about ethics yet remains silent on the fundamentals.
In parallel, I am financing grid‑scale energy projects that support digital infrastructure and hyperscaler compute. This involves navigating national grid connections, municipal supply contracts, environmental permits and water‑use licences. I also advise many of South Africa’s largest corporations on AI strategy and am a regular speaker on global stages. My perspective, therefore, is not theoretical—it is built on capital at risk, infrastructure under construction, and start‑ups that either survived or folded.
South Africa AI Policy
The draft South Africa National Artificial Intelligence Policy published earlier this month misses the mark on several critical fronts. First, AI should be framed as a national security issue, not merely a governance exercise. With a Gini coefficient of 0.67 and youth unemployment above 57 %, the country already faces stark inequality. AI‑driven automation is already cutting jobs in retail, finance, logistics, manufacturing and call centres—the very sectors that employ the most vulnerable workers. If South Africa does not become a builder of AI rather than a passive consumer, the resulting mass displacement could ignite a social crisis far beyond an economic adjustment.
The draft mentions job losses in passing and proposes reskilling programmes, yet that approach underestimates the urgency. The real remedy is to ensure that AI companies are built, trained and deployed locally, creating jobs and retaining talent. This requires robust infrastructure, clear incentives and rapid execution—not the creation of seven new government bodies that add compliance costs without delivering tangible support.
The policy’s institutional blueprint—seven new entities including a National AI Commission, an AI Ethics Board, an AI Regulatory Authority and others—appears premature. No public funds have yet been earmarked for compute infrastructure, nor have any regulatory sandboxes processed applications. Start‑ups I have backed often struggle because they cannot access affordable GPU time, prompting founders to relocate to the Netherlands, the UAE, the United States or even Rwanda for better compute credits and clear investment signals. Governance cannot precede the ecosystem it is meant to oversee.
Energy is the linchpin of any AI strategy. Section 9.1.2 treats “energy preparedness for the AI age” as a single bullet point, when in reality it is the core issue. As of April 2026, South Africa has enjoyed more than 300 consecutive days without load‑shedding, with Eskom reporting an excess of 4 GW of capacity over demand. This surplus coincides with a global shortage of power for data‑centre expansion; major hyperscalers in the United States and Europe are scrambling for electricity‑rich locations. If South Africa leverages its excess generation, renewable potential, strategic timezone and subsea cable connectivity, it could become a magnet for hyperscaler compute. The current draft offers no roadmap for dedicated energy allocation, wheeling agreements or renewable procurement for AI data centres.
Equally absent are concrete incentive mechanisms. The policy references tax breaks, grants and an “AI Innovation Fund” without detailing how they will operate. There is no R&D tax credit schedule, no match‑fund ratios, no special economic zones for AI clusters, and no compute‑credit programmes for start‑ups or universities. Without these tools, South African firms cannot compete with jurisdictions like Singapore or Canada that have already deployed capital with clear eligibility criteria.
Talent retention is already a crisis. More than 70 skilled South Africans leave the country each day, and only 31 % of African universities offer dedicated AI programmes. The draft’s vague “strategic talent retention” plan does not match the speed of the brain‑drain. Immediate measures—competitive research grants, compute access for labs, diaspora return incentives and a clear signal that building an AI company locally is economically viable—are essential.
The continent‑wide opportunity is also slipping away. Under the African Continental Free Trade Area, South Africa could serve as the AI services hub for a market of 1.4 billion people. Nations such as Kenya, Nigeria, Egypt and Rwanda are already investing billions in AI infrastructure. South Africa’s advantage lies in its financial system, legal framework and connectivity, but without a compute substrate, the window will close.
Open‑source AI models represent a strategic asset that the draft scarcely mentions. Open‑weight models like Llama or Mistral enable local fine‑tuning, preserve data sovereignty and reduce reliance on expensive foreign APIs. Policy should actively subsidise the deployment of these models on domestic hardware, fund language‑specific model training, and treat open‑source AI as a sovereign capability.
Nonetheless, the draft has merits worth preserving. The sector‑specific working groups, focus on local data collection, commitment to digitising indigenous languages, and the Ubuntu philosophy embedded in the document provide a solid cultural foundation. The acknowledgement that the policy is a “work in progress” is also encouraging.
To turn intent into impact, I propose seven actionable steps drawn from field experience:
- Declare AI infrastructure a national strategic priority with measurable targets for compute capacity and megawatts reserved for data centres.
- Capture the current energy surplus by establishing a data‑centre energy allocation framework within the first year.
- Introduce concrete incentive instruments—R&D tax credits, compute subsidies, accelerated depreciation, and AI‑focused special economic zones.
- Consolidate institutions into a single National AI Office reporting directly to the minister, staffed by practitioners.
- Launch an emergency talent‑retention programme offering grants, compute access and diaspora incentives.
- Elevate open‑source AI through subsidies for local deployment and fine‑tuning of models for African languages and public services.
- Set quantifiable targets for start‑up support, venture capital deployment, sandbox approvals, AI literacy and data‑centre capacity by 2028.
Minister Malatsi, the global AI infrastructure race is already underway. While other countries spend months perfecting ethics boards, South Africa has the power, location and talent to become a central hub—provided we act now. The next 24 months will determine whether we build an AI economy or merely regulate one that never materialises. The choice is clear: build first, incentivise second, govern third.