Data centre expansion requires transparent national resource accounts tracking electricity and water consumption across regions. Unifying infrastructure screening ensures digital ambition aligns with long-term host community value and grid resilience.
South Africa needs the physical infrastructure of the AI economy. But it cannot build highly resilient private AI infrastructure on top of fragile public systems and assume legitimacy will follow.
Ireland shows why this matters. Data centres consumed five percent of its metered electricity in 2015. By 2024, they consumed 22 percent, more than all urban households combined at 18percent. At that scale, compute stopped being an IT issue and became an infrastructure-policy issue.
The DBSA/National Planning Commission's 2025 study cites a 2023 industry estimate that South African data-centre IT-load capacity would reach 435MW in 2024. At full utilisation and an illustrative PUE of 1.3, that equates to roughly 5TWh a year, about 2.1 percent of South Africa's 2025 electricity generation. It is a capacity-equivalent estimate, not measured consumption.
At 1GW of IT load, the same assumption takes annual demand to roughly 11.4TWh, close to 5percent of 2025 generation and about a year's electricity for 1.2 million mid-range suburban homes using an illustrative 800kWh a month. Again, a scenario, not a forecast.
The first policy warning is basic: South Africa cannot readily answer, from one public national source, what share of the country's electricity and water the data-centre sector consumes today. We know installed capacity, operator disclosures and the development pipeline. We do not have a consolidated national resource account.
That matters because permission can disappear quickly once communities believe the bargain is unfair. In the second quarter of 2026, Data Center Watch recorded 45 US data-centre projects worth about US$68 billion blocked or delayed by local opposition.
In Sydney, a developer withdrew a proposed 90MW, A$1.2 billion facility after acknowledging that it had failed to secure a social licence. In Cape Town, a major rezoning was approved and then suspended after an appeal.
The danger is not that South Africa builds too much AI infrastructure. It is that we build highly resilient private facilities with guaranteed power, connectivity, water resilience and security while surrounding communities experience scarcity, then assume planning approval equals enduring legitimacy.
South Africa must build the AI infrastructure
Social licence is becoming an infrastructure risk. Waiting until contestation reaches scale will be too late.
Let me be clear: South Africa must build the AI infrastructure.
A serious AI economy needs domestic compute for regulated workloads, cloud services and any credible ambition around digital sovereignty. The choice is not between data centres and no data centres. It is between building them deliberately or discovering their consequences after they have become too important to question.
AI is not weightless. Behind the models sit substations, fibre routes, cooling systems, batteries, generators and land. These disputes are warnings about what happens when digital ambition collides with physical infrastructure and local legitimacy.
The obvious objection is that South Africa cannot afford another approval layer around infrastructure it urgently needs. I agree. The answer is not another veto point, but to surface power, water, land-use and host-community questions before they become appeals and redesigns after capital is committed.
The uncomfortable inequality question
What happens when a private AI facility has more reliable electricity, water resilience, fibre and security than the community surrounding it?
The facility should not be made less resilient. The problem is the contrast. South Africa cannot allow highly protected private AI infrastructure to become more resilient than the public systems and communities on which it depends.
Large capital investment does not automatically produce large permanent employment. A Brookings study of about 1,500 US facilities found roughly 100 to 200 local jobs in a typical host county. It is US evidence, but it shows why construction announcements and permanent economic value should not be confused.
Data centres can still generate substantial national value through power investment, fibre, suppliers, cloud availability, skills and productivity. But national economic value is not the same thing as host-community value.
Who pays for the substation? What remains locally when construction ends? What does the municipality receive? Those questions determine whether people living alongside a twenty-year asset can see it as part of their own economic future.
Resource questions require the same discipline. Calling all data centres water guzzlers is no more useful than waving the issue away because a modern facility is efficient. The answer is facility-level disclosure: electricity and water required, their source and who pays for reinforcement.
No slogans. Show the numbers.
South Africa's fragile underbelly
By a fragile public underbelly, I mean the shared electricity networks, fibre corridors, municipal systems, security institutions and planning processes underneath the data-centre fence, and the legitimacy of how scarce resources are allocated.
PRASA, Transnet and Eskom are not evidence that anybody will attack a data centre. They show what important infrastructure costs when the wider system becomes fragile. In 2025/26, Transnet reported R3.7 billion in security expenditure, while PRASA reported a R3.4 billion security allocation. Together, that is more than R7 billion devoted to protecting two national transport systems. Eskom reported R191 million in physical-security crime losses, 18percent lower than the previous year.
South Africa already pays a resilience premium when critical infrastructure depends on fragile physical, institutional and social systems. Data centres will not be exempt simply because the buildings themselves are highly secure.
The same principle applies technically. In November 2025, two unrelated fibre breaks hit primary and backup routes between Cape Town data-centre locations. A customer that had bought diverse routes discovered that both ultimately depended on the same vendor.
Redundancy is not independence. The fence around a data centre is not the boundary of its resilience.
So what should public policy do?
Social licence cannot be left to the goodwill of individual developers. It has to be designed into the public-policy framework within which large data centres are approved, connected and allowed to expand.
First, South Africa needs a national resource account: installed IT-load capacity, actual electricity consumption where measurable, water use, location and the forward pipeline.
Second, strategically material projects need one coordinated infrastructure screen. Power, additional generation, water, land use, environmental effects, fibre, municipal capacity, security and host-community economics should meet before approval rather than move through separate processes where nobody owns the combined consequence.
Third, policy should establish a minimum social-licence framework. Not prescribed corporate social investment, but transparency before decisions are locked in, participation that can still influence the project, durable local economic value and commitments measurable over the life of the asset.
A community living next to a twenty-year data-centre asset should be able to point to something durable and say: that infrastructure strengthened our future too.
Developers and operators should publish the bargain before approval: megawatts, water source, grid reinforcement, construction versus permanent jobs, local procurement, skills commitments and municipal benefit.
CIOs, CEOs and boards should treat social licence and systemic resilience as part of capacity risk. Where do supposedly diverse routes converge? Does capacity depend on approvals still under challenge? What happens if the surrounding grid, fibre network or municipality fails? What commitments has the provider made to the host environment?
Social licence will not stop a criminal syndicate. Security cannot manufacture legitimacy, and legitimacy cannot replace security. South Africa needs both.
We are early
That is South Africa's advantage. Ireland tightened its rules after data centres had already become a major electricity load. South Africa can act before reaching that point.
The scarce resource may eventually not be electricity, water or land. It may be public permission to allocate those resources to private digital infrastructure.
Power tells us whether the facility can run. Protection tells us whether it can survive. Permission tells us whether it can remain.
South Africa cannot build highly resilient AI infrastructure while ignoring the fragility of the public systems and communities on which it depends.
The countries that succeed in the AI infrastructure race will not simply be those that build the most compute. They will be those whose compute still has power, protection and permission twenty years from now.
