South Africa is ready to scale AI. Access and infrastructure could still split the winners

More than 2,900 government and industry leaders attended Huawei South Africa Connect 2026 at the Sandton Convention Centre on 23 July.

The programme focused on moving artificial intelligence beyond isolated pilots and into public services, energy, finance, telecommunications and education.

That shift is necessary. AI creates little public value while trapped in demonstrations and innovation labs.

Scaling also increases the cost of mistakes. A flawed pilot disappoints a team. A flawed system running across a bank, power grid or government service can affect millions.

What you need to know

• Huawei says more than 2,900 leaders attended the event.
• Communications minister Solly Malatsi opened the conference.
• Sessions focused on industrial deployment, infrastructure and inclusion.
• Eskom and Standard Bank were among the organisations represented.
• South Africa has strong regional AI readiness and local cloud capacity.
• Connectivity, affordable devices, data quality and skills remain major constraints.

What does moving beyond pilots mean?

An AI pilot tests whether a use case works in a limited environment.

Scaling means integrating it into staff workflows, databases, security systems, procurement, customer support, physical infrastructure, monitoring, compliance and budgets.

The model is only one component.

A useful prediction system for electricity maintenance needs accurate sensor data, integration with work orders, trained technicians and a process for handling wrong predictions.

The demo can be impressive while the organisation remains unprepared.

Where South Africa could use industrial AI

In energy, AI can forecast demand, detect equipment anomalies and prioritise maintenance. In finance, it can support fraud detection, document processing and operational risk. Telecom networks can use it to predict faults and manage congestion. Public services can improve search, translation and case routing. Mining can apply computer vision and predictive maintenance.

Every use comes with a governance question.

Historical banking data can reproduce exclusion. Automated surveillance can expand beyond its original purpose. Critical infrastructure can become dependent on software that attackers target.

Why infrastructure comes first

AI at scale requires reliable electricity, fibre, mobile networks, cloud or local compute, storage, cybersecurity, high-quality data, identity systems and skilled operators.

South Africa has stronger infrastructure than many regional peers. It also has deep inequality.

A company in Sandton and a clinic in a poorly connected municipality do not enter the AI era with the same equipment.

National readiness averages can hide local exclusion.

The access problem

Malatsi emphasised connectivity and affordable devices as immediate priorities.

AI-enhanced education is not inclusive when students share a slow phone and expensive data. Digital public services fail when citizens cannot authenticate or reach them. Small businesses cannot adopt cloud AI when monthly costs are priced against stronger currencies.

The risk is a two-speed economy in which large firms automate and smaller organisations fall further behind.

What industrial AI governance requires

Before scaling, organisations should define the decision the system supports, the data it uses, the acceptable error rate, who remains accountable, human override, drift monitoring, incident reporting and shutdown conditions.

Critical infrastructure also needs resilience against model failure, data poisoning, cyberattack and connectivity loss.

An intelligent grid should remain a grid when the intelligence is offline.

Huawei's role deserves scrutiny

Huawei supplies networks, cloud systems, storage and AI infrastructure. Its integrated stack can simplify deployment. It can also deepen vendor dependence.

South African organisations should examine data location, security, interoperability, local support, procurement competition, export controls, update dependence, exit costs and independent auditing.

The question is not whether Huawei is uniquely risky. It is whether any supplier should become too difficult to replace.

The workforce question cannot be postponed

Industrial AI changes jobs even when it does not remove them. Maintenance teams may receive machine-generated priorities. Bank employees may review automated fraud alerts. Call-centre staff may handle only the cases an assistant could not resolve. Engineers may supervise systems that once required larger operational teams.

That can make work safer and more productive. It can also remove the routine tasks through which junior employees learned. Organisations should therefore map which skills disappear, which new responsibilities appear and how workers will move between them.

Training should happen before deployment. Staff need to understand what the system measures, when it can be wrong, how to challenge it and who remains accountable. Managers should not punish workers for overriding a model when evidence supports the decision.

South Africa also needs pathways for smaller suppliers, universities and local researchers to participate. If industrial AI is purchased only as a complete foreign platform, local workers may become operators rather than builders. Procurement can require skills transfer, local testing, open interfaces and partnerships that create lasting capability.

A national AI strategy should count new expertise and good jobs alongside efficiency. Productivity gains that weaken the talent pipeline can become expensive later.

Public procurement can reinforce that goal by favouring measurable local participation rather than vague partnership language. A supplier should identify which systems local teams will design, which skills will remain after implementation and how universities or smaller firms can access the platform. Without that discipline, South Africa may scale AI consumption faster than AI capability, leaving the most valuable engineering and intellectual property elsewhere.

The tecMAMBO take

South Africa has the infrastructure, companies and skills to become a serious industrial AI market.

Its main challenge is no longer proving that AI can work. It is deciding where it should work, who benefits and how to stop it when it fails.

Scaling technology before scaling accountability is merely a faster route to a larger problem.