Time to course-correct for AI sustainability, says sustainability lead Claire Bradbury

That’s equivalent to the energy used by hundreds of millions of homes.

More troubling still, the water required to cool data centres especially in regions that already face water scarcity could reach crisis levels. In South Africa, we are already living with the twin pressures of unreliable electricity and severe water stress. And yet, local enterprises are racing to adopt AI without asking the crucial question: how sustainable is this growth?

The uncomfortable reality is that if we don’t course-correct now, AI will push us closer to climate instability even as it helps us solve other problems. It’s the ultimate contradiction using a future-forward tool with a 20th-century energy model. That contradiction must be resolved. And it starts with accountability. Every South African organisation that embraces AI must do so with full visibility into the environmental cost and a commitment to minimising it.

The Sustainable AI Quotient (SAIQ) is a performance framework that allows companies to evaluate AI investments against four critical thresholds: financial return, energy usage, water dependency, and carbon emissions.

The hidden cost of AI is fast becoming the next frontier in corporate accountability.

Fortunately, the solutions are within reach. Start with smarter silicon. New computer architectures like Processing-In-Memory (PIM) and Compute-In-Memory (CIM) can dramatically reduce the energy intensity of AI operations. Instead of constantly shuttling data between memory and processing units, an energy-hungry exercise, these chips perform computation directly within memory, cutting power consumption significantly. That’s not a tech detail. That’s a sustainability breakthrough.

Then consider the geography of your data. AI workloads must be located where clean, affordable energy is available. That might mean shifting some operations to regions with high solar penetration or hydro capacity. The next generation of competitive advantage will come from clean computers.

There’s also the matter of design discipline. It’s time to apply restraint. Use AI where it matters. Train models with purpose. Avoid redundant data cycles. This is about thoughtful innovation, not performative digitalism.

And finally, governance. Sustainability in AI must be written into the code. That means deploying governance-as-code frameworks that automate sustainability guardrails, monitor energy thresholds in real-time, and flag violations before they spiral.

South African companies have a responsibility to lead in the design of responsible AI.The AI decisions we make in the next 24 months will determine whether we lock in a high-carbon future or build the foundation for sustainable digital transformation. That’s the fork in the road.