Capitalising on the AI Data Centre Gold Rush to Fuel South African Macroeconomic Growth: Benchmarking the Hyperscale Job-Creation Impact

Strategic Context / The Capital MigrationThe Gold Rush Ledger: Where Extraordinary Capital Lands and Who Keeps the Gold

The artificial intelligence boom is a capital migration before it is a technology revolution. The International Energy Agency’s April 2026 report records that global data-centre electricity demand rose 17 per cent in 2025, against 3 per cent growth in total global electricity demand. It projects that data-centre consumption will roughly double by 2030, from 485 terawatt-hours to 950 terawatt-hours. This is not simply an expansion of computing. It is a global contest for the physical infrastructure that makes computing possible: land, electricity, water, fibre connectivity, equipment, and regulatory certainty. Capital gravitates towards jurisdictions that can deliver these inputs at scale, reliably, and competitively. The location of a data centre therefore determines more than where servers operate. It influences where construction contracts are awarded, where technical capabilities develop, where infrastructure is upgraded, and which economy captures the wider investment multiplier. The digital economy may be borderless; its physical wealth is not.

South Africa is already part of this capital migration. The President’s February 2026 State of the Nation Address recorded 55 data centres already built and more than R50 billion in digital infrastructure investment expected over the following three years. Public-facing industry reports place the investment pipeline at approximately R50 billion, concentrated principally around Johannesburg and Cape Town. Microsoft has committed a further R5.4 billion to expand its cloud and AI infrastructure by the end of 2027. Teraco has been associated in industry reporting with four facilities and almost US$900 million in investment. Vantage announced a US$1 billion investment in its Waterfall campus in 2021. The Development Bank of Southern Africa’s 2025 Digital Infrastructure Investment Study, as reported by ITWeb, estimates that investment in South African data-centre infrastructure exceeded R39 billion in 2023 and could reach R68 billion by 2029, including investment commitments. These figures establish the scale of the opportunity. They do not, by themselves, establish how much value South Africa will retain.

Geography, however, is not competitive advantage; infrastructure is. A country can occupy a strategic position on the global digital map and still lose investment to jurisdictions that can provide more reliable power, faster grid connections, and greater regulatory certainty. South Africa’s fibre connectivity and regional market position offer a genuine advantage, but electricity and water constraints can turn that advantage into an expensive illusion. In October 2026, at the Windaba conference, Minister of Electricity and Energy Kgosientsho Ramokgopa said government believed South Africa could secure 0.4 gigawatts of growing global data-centre demand, potentially more. That is a significant ambition, but it is also a bounded one. The strategic question is not simply whether South Africa can attract 400 megawatts of data-centre demand. It is whether the country can supply that capacity without displacing higher-value economic activity, while ensuring that the resulting investment generates domestic procurement, industrial participation, skilled employment, infrastructure improvements, and fiscal returns. Otherwise, South Africa risks hosting the infrastructure of the AI economy while capturing only a fraction of its wealth. The gold rush may arrive; the gold may leave.

Analytical Examination / Capital Against LabourThe Employment Inversion: Why Billions of Capital Employ Hundreds of People

Capital intensity and employment intensity are not interchangeable measures of economic value. Hyperscale data centres expose the difference with unusual clarity: billions of dollars can finance infrastructure that requires relatively few permanent employees once construction ends. Food & Water Watch estimates that Virginia’s data centres generated one permanent job for every US$13 million invested, using Virginia Economic Development Partnership data dating back to 1990. Its analysis of more recent investment is starker still: between 2020 and late 2025, one direct, permanent job was generated for every US$54 million invested. That equates to approximately 19 permanent jobs per US$1 billion. The distinction is consequential. Capital can accumulate at extraordinary speed while employment expands at a fraction of that pace. These are the economics of infrastructure built by many hands, but operated by few.

Virginia’s fiscal-year 2025 disclosures sharpen the question of public value. Data-centre operators reported 1,610 net new jobs alongside approximately US$48.6 billion in investment, while the reported tax benefit associated with the exemption was approximately US$1.94 billion (the state’s 2025 annual financial report records approximately US$1.6 billion of forgone revenue). Dividing that tax-benefit figure by the reported jobs produces approximately US$1.2 million per job, but this is an arithmetic comparison, not a verified cost-per-job calculation: the tax benefit is an aggregate fiscal measure, not a direct payment for each position, and the employment figures are operator-reported. The underlying policy question remains unavoidable: what permanent employment and wider economic returns justify the public incentives attached to capital-intensive investment?

Construction is a surge; operation is a plateau. Georgia’s experience illustrates the distinction, but also the danger of accepting headline figures without examining their methodology. The state’s original 2025 estimates of 28,350 construction jobs and 5,471 operational jobs were subsequently corrected. The revised estimates, which the auditors attributed to the exemption, fell to 8,505 construction jobs and 1,641 operational jobs. Even after that substantial downward revision, construction-related employment remained approximately five times operational employment. The distinction is not merely statistical. Construction creates a temporary employment pulse; operations establish a smaller, longer-term employment base.

The wider evidence is more nuanced than the claim that data centres create few jobs. A Brookings study examined approximately 1,500 US data-centre facilities and 52 announced-but-cancelled projects, linking them to county-level employment and wage data from 2003 to 2024. It found that the arrival of a first large data centre increased employment in data processing by 56 per cent over the following decade, with telecommunications employment rising by 43 per cent. Its estimates imply approximately 100–200 additional jobs in a typical affected county, depending on facility type. Yet the distribution of those benefits matters: hyperscale facilities generate broader telecommunications effects, while colocation facilities, which lease capacity to tenants that may have no local operating presence, produce more limited employment gains. Investment type, therefore, matters as much as investment volume.

South Africa provides an unusually direct illustration. Equinix reported that its facility outside Johannesburg employed approximately 750 people during construction but now retains just 26 permanent employees. The contrast is stark: substantial construction activity does not translate into an equivalent permanent workforce. The facility remains economically significant, but its employment profile demonstrates why temporary job creation and enduring employment must never be conflated.

South Africa’s arithmetic is harsher because its employment challenge is larger. Statistics South Africa recorded 8.5 million unemployed people in the second quarter of 2026, an official unemployment rate of 33.6 per cent, and 16.7 million employed people. 10 data centres employing 100 permanent staff each would create 1,000 jobs, equivalent to approximately 0.012 per cent of the unemployed population. This is an illustration, not a forecast. It exposes the scale mismatch between capital-intensive infrastructure and a labour market requiring employment on a vastly greater scale. The investment is not worthless because it cannot absorb mass unemployment. But neither is it a credible mass-employment strategy merely because its investment value is measured in billions.

The strategic implication is not to demand that data centres become something they are not. It is to ensure that South Africa captures employment beyond the server hall. Construction procurement, electrical engineering, cooling systems, fibre networks, security, maintenance, renewable-energy supply, technical training, and locally delivered digital services can extend the employment dividend beyond the relatively small operational workforce. The distinction is between importing a finished technological ecosystem and developing domestic capabilities around it.

Government must therefore evaluate data-centre proposals against more than capital expenditure, megawatts, and projected GDP contributions. Investment incentives should be conditional on transparent employment baselines, local procurement commitments, skills-transfer obligations, measurable supplier participation, and independently verified economic outcomes. Construction jobs, permanent jobs, indirect jobs, and induced jobs must be reported separately. Announced employment must not be mistaken for realised employment, and projected multipliers must not be presented as verified outcomes.

The ultimate test is not how much capital enters the country, but how much productive capacity, domestic enterprise, and durable employment remain when the construction crews leave. A country can host the infrastructure of the future while importing much of its value and exporting much of its economic opportunity. That is the employment inversion: extraordinary investment, limited direct employment, and a national dividend that remains contingent on what the host economy builds around it.

Analytical Examination / Value RetentionThe Leakage Problem: Gross Activity, Net Retention, and the Rand That Never Stays

Gross economic activity and domestic value retention are different ledgers. A data centre can generate substantial investment, construction expenditure, electricity demand, and reported economic activity without retaining a proportionate share of that value in the host economy. The distinction begins with employment. Direct jobs sit with the operator; indirect jobs arise across suppliers of materials, equipment, electricity, and maintenance; induced jobs emerge when employees and suppliers spend their additional income. All three can be counted or estimated, but they are not evidentially equivalent. Direct employment can be observed. Indirect and induced employment generally require economic modelling, with results shaped by assumptions about domestic sourcing, supply-chain capacity, household spending, and economic linkages. Combine observed jobs and modelled multipliers without disclosing those assumptions, and an estimate becomes an apparent fact. The multiplier may be impressive; the underlying domestic contribution may be considerably less so.

The same distinction applies to capital. A facility may stand on South African soil while much of its equipment, cooling technology, software, intellectual property, and specialist services is sourced internationally. Construction expenditure may flow through local contractors, yet a substantial portion of the underlying value may accrue to foreign equipment manufacturers, technology providers, and specialist suppliers. This is not evidence that every South African project has high import dependence. It is the structural exposure that investment announcements rarely resolve. A domestic address does not guarantee domestic value; local activity does not guarantee local ownership.

Teraco illustrates the distinction between domestic financing and international ownership. Absa confirms a R2.5 billion financing transaction supporting Teraco’s infrastructure development. Of this, R2 billion was allocated to data-centre construction in Johannesburg and Cape Town, while R500 million was intended to provide liquidity and optimise the company’s capital structure. South African banking capital therefore helped finance the expansion of strategically important domestic digital infrastructure. Yet the ownership ledger tells a different story. Digital Realty completed its acquisition of a 61.1 per cent indirect controlling interest in Teraco on 1 August 2022. Its June 2026 filings with the United States Securities and Exchange Commission state that a put option exercised by minority shareholders in May 2026, covering approximately 16 per cent of Teraco, will raise that interest to 77 per cent once settled, which was expected in the second half of 2026. The platform is built and operated in South Africa, supported partly by domestic finance, but controlled by a US-listed multinational. These facts are not contradictory. They expose the distinction between the location of productive assets, the origin of financing, and the ownership of economic returns.

Foreign ownership does not mean that all economic value leaves the country. South African employees receive wages, local suppliers earn revenues, banks earn financing returns, and government collects applicable taxes. Domestic shareholders and empowerment interests may also participate in the returns. Teraco’s 2024–2025 verification certificate records Level 3 B-BBEE status with 110 per cent procurement recognition. These are relevant dimensions of domestic participation, but they do not establish the proportion of total project expenditure retained locally or the ultimate distribution of economic returns. The strategic error would be to treat foreign control as proof of economic leakage, just as it would be to treat local construction and employment as proof of comprehensive domestic value capture.

The decisive metric is therefore the domestic share of total project expenditure, but even that requires disciplined definition. It should distinguish local labour and services from domestically manufactured equipment, local supplier value added from imported inputs resold by local intermediaries, and South African financing from foreign equity and debt. It should also separate capital expenditure from recurring operational expenditure. A locally registered contractor purchasing imported equipment does not create the same domestic value as a South African manufacturer producing that equipment. Nor does a rand spent locally necessarily remain in the domestic economy in its entirety. The relevant measure is not simply where an invoice is issued, but how much South African value is created and retained through the transaction.

The public sources reviewed for this briefing did not disclose a comparable domestic-retention percentage for the South African hyperscale projects examined. That is a disclosure gap, not evidence that their domestic contribution is negligible. But the absence of a disclosed baseline has consequences: without it, government cannot reliably assess local value capture, compare projects, test investment incentives against actual economic returns, or determine whether public concessions are producing commensurate national benefits.

Government should close that gap. Incentives and other material public concessions should be linked to transparent, independently verifiable reporting of domestic procurement, local value added, employment by category, skills transfer, tax contributions, financing, ownership, and reinvestment. Reporting should distinguish commitments from actual expenditure and projections from realised outcomes. Commercial confidentiality can be protected through appropriate aggregation without making the national economic dividend invisible.

The strategic objective is not merely to attract capital, but to negotiate its domestic consequences. A data centre can be physically local, financially international, technologically dependent, and economically only partially integrated. South Africa must ensure that these conditions do not combine to produce an impressive investment headline and a diminished national return. What is not measured cannot be negotiated with precision. What is not negotiated may be surrendered by default. The building may remain; the value may travel.

Analytical Examination / Electricity EconomicsThe Grid Paradox: The Infrastructure Advantage That Could Become a National Liability

Electricity is where the opportunity and the liability converge. Data centres convert digital demand into physical claims on land, power, transmission capacity, and water. The Development Bank of Southern Africa’s South Africa’s Digital Infrastructure Investment Study 2025 forecasts installed data-centre IT load rising from 435 megawatts in 2024 to 829 megawatts by 2029. That is an increase of approximately 91 per cent. Those endpoints imply a compound annual growth rate of approximately 13.8 per cent, which matches the 13.77 per cent reported by Mordor Intelligence, the source of the capacity forecast. The scale of the forecast remains significant.

The International Energy Agency provides the global comparison. A typical AI-focused data centre can consume as much electricity annually as 100,000 households; the largest facilities under construction could consume 20 times as much, equivalent to approximately two million households. This is not an ordinary increase in commercial electricity demand. It is a concentration of extraordinary demand within a single industrial site. Such facilities can attract investment, accelerate renewable-energy procurement, and stimulate grid expansion. But where additional generation and transmission capacity fail to materialise, the same demand can intensify competition for electricity among data centres, mines, smelters, manufacturers, and households. The strategic question is not whether data centres need electricity. It is whether their expansion increases the country’s productive energy capacity or merely reallocates scarcity.

Operators are already responding by securing energy beyond conventional grid supply. Daily Maverick reported in February 2026 that Teraco was developing a 120-megawatt solar photovoltaic facility in the Free State, with electricity wheeled through the national grid to support its data-centre operations. The report also identified other arrangements: Vantage had secured 87 megawatts for Waterfall City through SolarAfrica, while Africa Data Centres had begun developing a 12-megawatt solar farm near Bloemfontein to support its Cape Town expansion and future Johannesburg facilities. These investments can add generation and reduce operators’ dependence on conventional supply. Yet renewable generation and transmission capacity are not interchangeable. Wheeling requires access to the network, and a private power purchase agreement does not, by itself, eliminate the physical constraints of moving electricity from where it is generated to where it is consumed.

This creates a consequential paradox. Private generation can strengthen national electricity supply, but private access to scarce transmission capacity can also sharpen competition for the network. The outcome depends on additionality, location, timing, storage, transmission availability, and the treatment of existing users. If a data centre finances genuinely additional generation and the necessary network investment, it can become a contributor to the system. If it secures preferential access to constrained transmission infrastructure without materially expanding available supply, it risks becoming a competitor for scarce public capacity. The distinction is between adding energy to the economy and securing a larger share of energy that the economy already lacks.

Forecasting risk compounds the problem. In July 2026, Thomson Reuters Foundation reporting carried a warning from Bryce Allan, Teraco’s head of sustainability, that some developers had inflated the electricity requirements of their proposed facilities. He described these exaggerated claims as “bragawatts”, arguing that actual demand could take years to approach even half of the capacity being announced. This is an attributed industry assessment, not an independently established finding about every developer. Its strategic significance, however, is considerable. Announced capacity is not commissioned capacity; reserved power is not necessarily consumed power; and projected demand is not realised demand. If utilities and public authorities plan infrastructure against inflated forecasts, capital can be committed to assets whose eventual utilisation falls materially short of expectations. If they underestimate demand, the opposite risk emerges: insufficient generation and transmission capacity constrain investment, industrial production, and economic growth.

The United States offers a historical warning against confusing projected demand with bankable demand. In July 1983, the Washington Public Power Supply System defaulted on US$2.25 billion in bonds associated with two abandoned nuclear power projects. The failure was not attributable to inaccurate demand forecasts alone. It involved major construction-cost escalation, project failures, financing pressures, and demand assumptions that proved untenable. The lesson is broader than forecasting error: infrastructure built on overconfident assumptions can leave the public carrying the cost long after the original investment rationale has collapsed. A grid expanded for demand that never materialises can become a stranded asset; a grid that cannot accommodate demand that does materialise can become a constraint on national production.

South Africa therefore needs a more exacting electricity compact with the data-centre industry. Government, Eskom, transmission planners, municipalities, and operators should assess projects against independently verified demand forecasts, realistic commissioning schedules, actual utilisation, and the additional generation and transmission capacity they require. Grid connections and transmission access should be assessed transparently against system-wide capacity, not simply the purchasing power of individual investors. Where projects rely on private generation and wheeling, the assessment should disclose how much new generation is added, when it becomes available, what transmission capacity it consumes, and whether storage or network investment is required. Forecasts should be periodically reconciled against actual consumption, with material deviations disclosed.

The objective is not to ration technological progress. It is to prevent a private infrastructure boom from becoming a public electricity liability. South Africa must attract the investment without sacrificing the energy security of its existing productive economy. The country cannot afford to build a digital economy that consumes the electricity required to sustain its industrial economy. The ultimate test is not how many megawatts data centres secure, but how many additional, reliable megawatts the country gains.

Analytical Examination / Water EconomicsThe Cooling Ledger: Why the Scarcest Input May Be the One Nobody Priced

Electricity is the visible constraint; water is the concealed liability. Artificial intelligence converts computational demand into physical resource consumption, but the resulting water burden depends heavily on how facilities reject heat. According to the Uptime Institute figure cited by IOL, a relatively small one-megawatt data centre can consume approximately 26 million litres of water annually for cooling alone. The objectors in the Cape Town dispute applied a separate estimate of approximately 25.5 million litres per megawatt for a small facility using traditional evaporative cooling. These are indicative benchmarks, not universal consumption rates. Cooling design, climate, operating conditions, and technology can produce radically different outcomes. The distinction matters: a water-intensive facility and a water-efficient facility may deliver comparable computing capacity while imposing vastly different demands on the host community.

The proposed King Air Industria development near Cape Town International Airport exposes the stakes. Opponents, represented by the Legal Resources Centre, estimated that two proposed hyperscale facilities could consume approximately 4.4 billion litres of water annually if conventional evaporative cooling were used. Their calculation applied an estimated 25.5 million litres per megawatt to a reported electrical demand of approximately 174 megawatts. Accounts of the project’s electrical rating, however, vary between approximately 160 and 174 megavolt-amperes (MVA), a measure of apparent power rather than megawatts, which prevents the calculation from being treated as a technically exact project forecast without further electrical specifications. The 4.4 billion litres is therefore a conditional estimate, not a confirmed consumption figure. The same development has been reported as requiring electricity equivalent to approximately 130,000 homes. The comparison conveys the scale of the proposed electrical demand; it does not establish the project’s water requirements.

The distinction between potential consumption and disclosed consumption is the central issue. Teraco has said that it uses closed-loop cooling, which recirculates water rather than continuously relying on evaporative cooling. In Thomson Reuters Foundation reporting, republished by Engineering News and TimesLIVE, Bryce Allan, Teraco’s head of sustainability, said a 30-megawatt facility using closed-loop cooling would consume approximately as much water annually as an average restaurant. This is an industry executive’s comparison, not an independently audited consumption figure. Nevertheless, it demonstrates why applying a single water-intensity benchmark to every data centre is misleading. The industry cannot claim that all facilities are water-intensive; objectors cannot assume that every facility will consume water at the highest estimated rate. The answer lies in verified, facility-specific disclosure, not competing assurances.

King Air illustrates what happens when the evidence arrives too late. The land-use application did not disclose substantive figures for water consumption or detail the proposed cooling system. The landowner indicated that final resource requirements would be provided at the subsequent site-development-planning stage. That sequencing reverses the logic of responsible planning: permission is sought before the public can fully assess the resource demands of the development it is being asked to accommodate. The question is not whether technical specifications may legitimately evolve during design. It is whether the principal resource implications should remain unresolved when decisions establish the land-use rights for infrastructure of this scale. A resource that cannot be quantified at the point of decision cannot be properly weighed against competing public needs.

The regulatory position compounds the problem. South Africa’s National Data and Cloud Policy, published in 2024, recognises that data centres require substantial water and electricity. It requires compliance with environmental legislation and building by-laws, prioritises self-provision of water and electricity, and calls for backup supplies to reduce dependence on national grids. Yet a broad national policy is not the same as a detailed, consistently applied framework for assessing hyperscale facilities, their water intensity, and their cumulative demands on stressed catchments. Pública’s September 2026 investigation found that Cape Town had no separate land-use category for hyperscale data centres, treating them instead within existing commercial or industrial classifications. It also reported that South Africa lacked a dedicated national data-centre policy. The distinction is important: policy exists, but the regulatory specificity required by a new class of industrial-scale digital infrastructure remains incomplete.

The King Air dispute has since moved beyond the original planning decision. Following an appeal lodged by the Housing Assembly and Foxglove, the July 2026 land-use approval was suspended pending the appeal process. In September, the City of Cape Town confirmed that it was developing refined guidelines for large data-centre applications. Equinix subsequently described the site as a long-term land bank and said it had no immediate development plans. These developments do not resolve the underlying water question. They demonstrate that the regulatory process is being forced to confront questions that should have been settled before approval was granted.

Water disclosure should therefore become a condition of planning approval, not a promise deferred to a later stage. Every hyperscale application should disclose:

• Annual water demand: projected withdrawal and actual consumption, reported separately, in litres and megalitres.
• Cooling technology: the proposed system, its water requirements, and its expected performance under local climatic conditions.
• Water source: potable supply, reclaimed wastewater, alternative sources, and the proportion drawn from each.
• Water intensity: consumption relative to IT load and actual utilisation, rather than theoretical maximum capacity alone.
• Drought resilience: peak-season requirements, contingency arrangements, and the effect of water restrictions on surrounding users.
• Cumulative impact: the development’s additional demand in the context of catchment availability, municipal infrastructure, and existing community needs.

Reporting should continue after commissioning, with independently verified annual disclosures. A design-stage forecast is not an operating result; a commitment to conservation is not proof of conservation. Incentives and approvals should reward demonstrable reductions in potable-water consumption, not merely the adoption of technologies described as efficient.

The strategic question is not whether South Africa should accommodate data centres. It is whether the country can attract digital infrastructure without transferring its resource costs to communities and other productive users. A facility can be technologically advanced yet environmentally inefficient; economically valuable yet locally burdensome. The digital economy may run on data, but the communities around it still depend on water. South Africa must not exchange one infrastructure deficit for another. A resource left unmeasured is a resource surrendered; a resource surrendered today can become tomorrow’s national liability.

Strategic Context / Competitive PositionThe Contest for Capacity: Why a Strong Address Is Not a Durable Advantage

South Africa leads Africa’s data-centre market, but leadership is not immunity. Fortren & Company reported in July 2026 that the continent had more than 500 megawatts of operational data-centre capacity, with a further 890 megawatts in the development pipeline. South Africa accounted for more than 60 per cent of the operational total. McKinsey identifies South Africa, Egypt, Kenya, Morocco, and Nigeria as the continent’s five leading data-centre markets. The incumbent advantage is real. So is its vulnerability. A dominant market share today does not guarantee a dominant share of tomorrow’s investment, particularly when competitors are building capacity, attracting international operators, and improving their investment propositions. South Africa has the lead; it does not own the future.

The challengers are no longer merely expressing ambition; they are assembling capacity. Estate Intel projects that Lagos’s data-centre capacity will rise from 78.6 megawatts to more than 218 megawatts by 2030, an expansion of approximately 2.8 times. The pipeline includes major developments by 21st Century Technologies, Airtel’s Nxtra, and Open Access Data Centres. Yet Estate Intel also warns that installed capacity is expanding faster than utilisation. This introduces a more complicated competitive dynamic: markets can build ahead of immediate demand to secure future customers, but excess capacity can also produce underutilised assets and weaker investment returns. The contest is not simply to build more; it is to build capacity that arrives at the right cost, in the right market, at the right time.

Morocco is advancing another competitive proposition. ResearchAndMarkets reported in December 2025 that the country accounted for more than half of the upcoming power capacity among the African data-centre projects tracked in its portfolio database. This is a reported share of a particular project pipeline, not proof that Morocco already possesses more than half of Africa’s operational capacity. Nevertheless, it signals the geographical broadening of investment beyond the established southern and eastern African markets. Kenya, meanwhile, promotes its geothermal resources as an energy advantage for power-intensive digital infrastructure. The competitive proposition is increasingly differentiated: South Africa offers an established operator ecosystem and connectivity; other markets seek to compete through energy resources, new capacity, and targeted investment conditions.

The competition is not confined to Africa. India’s 2026–27 Budget proposed, and the Finance Act, 2026 has since enacted, a tax holiday through 2047 for qualifying foreign cloud-service providers using specified Indian data-centre facilities to serve global customers, subject to defined conditions. Domestic customers must be served through an Indian reseller entity, while related Indian data-centre providers may qualify for a proposed 15 per cent safe-harbour margin on cost. This is more than a tax concession. It is a long-horizon investment signal designed to anchor digital infrastructure and global cloud activity within India. The strategic contrast is stark: while some jurisdictions are offering investors visibility extending to 2047, South Africa must still resolve uncertainty around grid access, municipal planning, resource requirements, and the terms on which public infrastructure will support private digital investment.

Geographic advantage is not competitive advantage. Investors assess delivered electricity costs, power reliability, water availability, fibre connectivity, land, regulatory certainty, construction timelines, and technical skills as an interconnected investment proposition. A country may possess excellent connectivity but lose ground through unreliable power; it may offer abundant energy but lack the regulatory clarity or skilled workforce required to deliver projects on schedule. No single advantage compensates indefinitely for a structural weakness elsewhere. South Africa’s established operators, international connectivity, and concentration of digital infrastructure provide a formidable starting position. Yet grid constraints, fragmented municipal processes, competing demands on water and land, and insufficiently defined national investment conditions can convert an incumbent advantage into an increasingly expensive one.

The second-order risk is more serious than losing individual projects. Once competing markets attract operators, suppliers, specialist contractors, technical talent, and successive rounds of infrastructure investment, they begin to strengthen the very conditions that made them attractive in the first place. Capital attracts capability; capability improves execution; execution reduces investment risk; and lower perceived risk attracts further capital. The reverse can also occur. Delayed approvals, constrained electricity connections, and uncertain resource requirements raise delivery costs, weaken investor confidence, and redirect successive investment decisions elsewhere. Competitive advantage compounds; competitive neglect compounds too.

South Africa therefore needs to move from investment promotion to investment competitiveness. That requires a nationally coordinated data-centre investment framework, aligned across national departments, municipalities, utilities, and infrastructure agencies. It should establish predictable approval processes, transparent grid-connection criteria, clear land-use provisions, water and environmental disclosure requirements, and defined obligations for domestic procurement, skills development, and local industrial participation. Strategic sites should be assessed against their actual electricity, water, fibre, land, and transport capacity before they are marketed to investors. Incentives should reward additional productive capacity and measurable domestic value creation, rather than capital announcements alone. Government should also publish a regularly updated national pipeline distinguishing operational facilities, projects under construction, approved developments, and uncommitted announcements. Investors need certainty; the public needs evidence.

South Africa must compete not merely to host the next generation of data centres, but to shape the economic terms on which that infrastructure enters the country. Its current leadership is an asset, not a guarantee. Its geographical position is an opportunity, not a strategy. The decisive question is whether South Africa will define a competitive proposition strong enough to retain its leadership, or allow better-prepared rivals to convert today’s opportunity into tomorrow’s advantage. The market leader can become the market left behind.

Analytical Examination / Fiscal ReturnsThe Fiscal Dividend: Incentives Granted, Revenues Booked, and the Net Left for the Public

A country can win the investment and lose the arithmetic. The arrival of capital is visible; the public cost of attracting it is often less so. Virginia illustrates the scale of the fiscal question. In fiscal year 2025, data-centre operators reported US$1.94 billion in tax benefits and 1,610 net new jobs. Dividing the former by the latter produces approximately US$1.21 million per added job. This is an indicative ratio, not a verified subsidy cost per job: the tax benefit covers 56 participating operators, the employment figure is an aggregate, and the underlying company data were self-reported rather than independently validated. The distinction matters. A striking ratio can expose a fiscal risk, but it cannot substitute for a properly constructed cost-benefit analysis.

Virginia’s own return-on-incentive model complicates the picture further. It estimates US$3.07 billion in state and local tax revenues over five years against the US$1.94 billion tax benefit associated with fiscal-year 2025 investment, producing an estimated net fiscal return of approximately US$1.13 billion and a revenue-to-cost ratio of 1.6. These are modelled estimates, not independently audited net cash returns. The model incorporates direct and additional employment, property taxes, and other revenue streams, while relying on assumptions about investment that would not have occurred without the exemption. It also excludes local discretionary incentives. The lesson is not that incentives necessarily destroy public value, nor that Virginia has conclusively proved their success. It is that the fiscal verdict depends on the counterfactual: what investment, employment, and revenue would have materialised without the concession?

The employment arithmetic nevertheless demands scrutiny. Good Jobs First’s 2016 study of 11 major US data-centre incentive deals found average subsidies of US$1.95 million per permanent job promised. More recently, a proposed data-centre campus in Genesee County, New York, attracted an incentive package reported in March 2026 at US$1.46 billion in exchange for 125 jobs, approximately US$11.7 million per job. These are US examples, not South African benchmarks, and the New York package is proposed rather than a completed public expenditure. Yet the underlying tension is unmistakable: capital commitments can reach billions while permanent employment remains comparatively small. The public may subsidise the scale of the investment without securing a proportionate employment dividend.

The broader research also warns against treating every incentive as either wasteful or indispensable. Brookings estimates that incentives in hyperscale counties represent approximately 2 per cent of total construction investment. In colocation counties, however, they represent 62 per cent. Its updated analysis finds that hyperscale investment decisions are driven principally by power availability, land, and fibre connectivity, rather than tax concessions. The implication is strategically important: incentives may be too small to determine the location of some major investments, yet large enough to transfer substantial public value in other cases. A concession that does not change the investment decision is not an inducement; it is a windfall. A concession that does change it must still earn its public cost.

South Africa has not yet announced a finalised, data-centre-specific tax incentive in the material reviewed. In his Budget Speech on 25 February 2026, Finance Minister Enoch Godongwana said data infrastructure should be regarded as critical in the same way as electricity, ports, and transport networks, and confirmed that National Treasury would explore options to support the expansion of data centres and related infrastructure. That was a policy signal, not an approved incentive. The summaries of the 21 October 2026 Medium-Term Budget Policy Statement reviewed for this briefing contain no reference to a data-centre incentive. Separately, Communications and Digital Technologies Minister Solly Malatsi has supported extending the Real Estate Investment Trust regime under section 25BB of the Income Tax Act to digital infrastructure, including data centres. This is a further policy option under consideration, not evidence of an enacted concession. The design window remains open, and that is an opportunity to establish the public-interest conditions before fiscal commitments become difficult to reverse.

India demonstrates how far a competing investment proposition can extend. Its 2026–27 Budget proposed, and the Finance Act, 2026 enacted, a tax holiday through 2047 for eligible foreign companies providing global cloud services through qualifying Indian data-centre infrastructure. The exemption is conditional: qualifying foreign providers must meet specified requirements, including using notified Indian data-centre facilities, while services to Indian customers must be routed through an Indian reseller. The Budget also proposes a 15 per cent safe-harbour margin for qualifying Indian data-centre service providers supplying related foreign cloud companies. This is a long-horizon fiscal commitment intended to anchor global cloud activity in India. But South Africa should not mistake the duration of India’s concession for proof of its economic superiority. A tax holiday can attract activity; it does not, by itself, establish the net domestic value retained.

South Africa’s design principle must therefore be net public return, not gross investment; additional economic value, not subsidised activity. Any incentive should be assessed against a transparent counterfactual and a defined public-value test. Would the investment proceed without the concession? How much additional domestic procurement, local value added, and permanent employment would it generate? What taxes would be paid, what public revenue would be forgone, and what additional electricity, water, transport, and municipal infrastructure costs would the state or consumers carry? The calculation must distinguish temporary construction employment from permanent operational jobs, direct employment from modelled indirect and induced effects, and announced investment from capital actually deployed. It must also account for the opportunity cost of public resources that could support other sectors or infrastructure with higher employment and economic returns.

The policy response should be contractual, measurable, and enforceable. Incentives should be conditional on verified capital expenditure, additional investment, permanent jobs, local skills development, domestic procurement, and measurable fiscal contributions. Agreements should specify baselines, delivery dates, reporting obligations, and independent verification. Where commitments are missed, proportionate clawbacks should recover the public benefit granted. Tax concessions should also be time-limited and periodically reviewed, with no automatic renewal. Government should publish project-level commitments and realised outcomes, subject only to legitimate commercial-confidentiality protections. Above all, incentives should reward outcomes that would not otherwise have occurred, rather than reimburse investors for decisions they had already made.

The second-order consequence is decisive. An incentive regime designed around investment announcements can produce impressive capital headlines while quietly weakening the public balance sheet. It can also establish a precedent: once one investor receives preferential treatment, competitors may demand equivalent concessions, turning a targeted intervention into a subsidy race. The state then competes against its own revenue base, while investors capture the benefits of jurisdictional rivalry. The danger is not merely paying too much to attract investment. It is paying repeatedly for investment that no longer requires payment to arrive.

South Africa should compete on reliable electricity, efficient approvals, connectivity, skills, and regulatory certainty before competing through tax concessions. Where an incentive is necessary, it must purchase demonstrable additionality and a measurable national dividend. The objective is not to maximise the volume of capital attracted at any fiscal price. It is to maximise the public value retained after every concession, cost, and obligation has been accounted for. A country should not confuse the price of attracting investment with the value of securing it.

Evidence and Case Study / The Announcement GapThe Headline Multiplier: What an Indian Gigawatt Reveals About Announced Jobs and Economic Reality

Theory needs a test. Google’s US$15 billion Visakhapatnam AI hub provides one. It is a major infrastructure commitment, but its employment projections expose a fundamental problem in measuring the economic dividend of hyperscale investment: several numbers can describe the same project without measuring the same thing. Direct jobs, total employment, annual employment supported, and permanent technical positions are not interchangeable. Yet, once aggregated into a single headline, they can create the impression of an employment dividend far larger than the number of people the facility will ultimately employ. The investment is singular; the employment narratives are multiple. The difference is where economic analysis must begin.

Evidence and Case Study / South African EvidenceThe Bredell Test: What Teraco’s JB4 Reveals About Rands, Megawatts, and Permanent Jobs

South Africa’s data-centre investment debate needs more than capital announcements and projected economic multipliers. It needs identifiable projects against which employment commitments can eventually be tested. Teraco’s JB4 provides an unusually useful local benchmark: a substantial infrastructure investment, a specified critical power load, and a publicly stated forecast of permanent employment. The numbers expose a structural tension. Capital requirements are substantial; permanent staffing requirements are comparatively modest. The wider economic dividend may be considerably larger, but it cannot simply be assumed.

Teraco is a Johannesburg-based, South African-operated data-centre platform, majority-owned by Digital Realty since August 2022. Its JB4 facility in Bredell, Ekurhuleni, was announced as the largest standalone data centre to be built in Africa at the time. Its significance extends beyond its physical scale: it offers a rare opportunity to examine the relationship between capital expenditure, electricity capacity, and permanent employment using figures disclosed by the operator itself.

Strategic Synthesis / The Bargain TermsThe National Bargain: Converting Hyperscale Capital into Domestic Capability on Measurable Terms

South Africa’s strategy should rest on one exchange: competitive terms for capital, measurable terms for national participation. Investors need reliable electricity, predictable grid access, regulatory certainty, efficient approvals, and, where justified, targeted incentives. In return, the country must secure disclosed employment, domestic procurement, skills transfer, additional energy capacity, fiscal contributions, and meaningful access to computing for South African enterprises. This is not an argument against commercial returns. It is an argument against granting public advantages without securing corresponding public value.

The distinction is fundamental. Investment attraction measures what enters the country; economic development measures what remains, what improves, and what becomes possible as a result. A data centre can generate substantial capital expenditure without creating a proportionate permanent workforce. It can expand computing capacity without making that capacity affordable to domestic firms. It can procure renewable electricity without necessarily increasing the energy available to the wider economy. The physical infrastructure may be domestic; the economic dividend is not automatically so.

The national bargain must therefore be negotiated against measurable outcomes, not investment announcements. Public incentives, grid connections, access to strategic land, and supporting infrastructure should be assessed against the additional economic value they secure. Commitments should be distinguished from realised expenditure, permanent employment from temporary construction work, and observed outcomes from modelled multipliers. Where investors receive public concessions, reporting obligations, independent verification, and proportionate clawbacks should form part of the agreement. The state must offer certainty without surrendering accountability.

A bargain, however, is only as credible as its measurement. International comparisons become misleading when projects use different definitions, employment categories, investment periods, and denominators. The Bandzishe Group Hyperscale Economic Dividend Framework addresses this problem through 12 measurement dimensions. Each is assessed separately, with consistent definitions and disclosed methodologies. Counted jobs must never be blended with modelled employment; announced capital must never be mistaken for realised expenditure; and gross economic activity must never be presented as domestic value retained. Where evidence is unavailable, the framework records the gap rather than manufacturing an estimate. An honest blank is more useful to a board than a confident fiction.

The framework’s twelfth dimension may ultimately prove the most consequential. Computing capacity is an input, not an economic outcome. Its productive value depends on whether businesses, universities, research institutions, and public bodies can access it, afford it, and apply it to improve productivity. A country can host world-class infrastructure while its domestic firms remain priced out of the computing resources that infrastructure provides. That is the contradiction South Africa must avoid: exporting digital capacity while importing the productivity gains it enables.

Domestic access should therefore form part of the investment bargain. This does not mean compelling private operators to provide uneconomic or indiscriminate subsidised access. It means negotiating commercially viable arrangements through which South African enterprises, start-ups, universities, and research institutions can access appropriate computing capacity, develop AI applications, and build technical capabilities. The objective is to expand domestic demand and productive participation, not merely the physical supply of infrastructure.

The second-order opportunity is considerably larger than the data-centre sector itself. If South African firms can use affordable computing to improve manufacturing, financial services, logistics, healthcare, agriculture, mining, and public administration, the productivity dividend can extend well beyond the facilities that host the infrastructure. Local suppliers can move into higher-value services; technical skills can deepen; domestic enterprises can develop exportable capabilities. Conversely, if computing capacity remains inaccessible, the country risks becoming an infrastructure host rather than a technology producer, and a consumer of AI productivity rather than a participant in its creation.

The framework must consequently operate as more than a reporting exercise. Government should establish common definitions, require consistent project-level disclosure, and publish a national register distinguishing proposed, approved, under-construction, and operational facilities. Reporting should be periodic, independently verifiable, and sufficiently granular to expose the difference between commitments and outcomes. Incentives and public infrastructure support should be assessed against the resulting evidence, with periodic reviews and corrective action where agreed outcomes are not delivered.

The strategic objective is not simply to attract the infrastructure of the AI economy. It is to capture the economic capabilities that infrastructure makes possible. South Africa should welcome capital, but it should negotiate participation; facilitate investment, but measure its return; and expand computing capacity, but ensure that domestic enterprises can convert it into productivity. The bargain must work for investors and the country, not merely appear to do so.

Capital must enter. Capability must remain. Productivity must spread. And the public must be able to measure the difference.

Executive Implications / The Board AgendaThe Board Agenda: Eight Decisions That Convert Investment Announcements into National Returns

The decisions that follow are sequenced from measurement to economic leverage. A concession cannot be priced until its public value is counted; an investment cannot be judged until its domestic contribution is established. South African organisations and global corporations have different responsibilities, reflecting their respective institutional positions, commercial interests, and capacity to influence outcomes. Yet both face the same strategic test: whether the expansion of data-centre infrastructure produces durable economic value beyond the facilities themselves.

These decisions do not all require new legislation. Many can begin through board oversight, commercial agreements, procurement policy, investment conditions, and public-sector coordination. What they require is a change in the measure of success. Investment attracted is an input. Economic value retained is the return.

1. Demand the denominator. Every employment claim presented to a board must disclose the category of employment, the period measured, the investment against which it is calculated, and whether the figure represents observed jobs or a modelled estimate. South African boards should require project-level figures that can be reconciled to actual payrolls; global corporations should apply the same evidentiary discipline to employment claims made in their host markets. Construction jobs, permanent operational positions, and wider employment effects are not interchangeable. A number without a defined denominator is not a performance measure. It is a headline. The first reform costs nothing.

2. Maintain four separate employment ledgers. Construction, permanent operational, indirect, and induced employment must be reported independently, with indirect and induced figures explicitly identified as modelled where they have not been directly observed. Operators should disclose planned and realised permanent employment per facility and per megawatt of critical IT load. Governments should reject incentive applications that rely on a single blended employment total without disclosing its constituent categories and methodology. The distinction matters because temporary construction activity can create an impressive employment headline while the permanent workforce remains comparatively small. Aggregation creates scale; disaggregation reveals substance.

3. Contract domestic value creation. Establish measurable domestic procurement and value-added targets, supported by annual disclosure of project expenditure and independently verifiable supplier participation. South African engineering, electrical, construction, cooling, and technical-services firms should form consortia to compete for hyperscale contracts that individual businesses may lack the capacity to secure independently. Global operators should develop credible local supplier programmes as part of their long-term commercial presence, rather than treating domestic participation as a regulatory inconvenience. Measurement must distinguish genuine South African production and services from imported equipment routed through local intermediaries. Where disclosure is commercially restricted, the limitation should be identified rather than concealed. A local address is not proof of local value; procurement expenditure is not necessarily domestic value added.

4. Price the electricity bargain. Every major project should disclose its expected and actual electricity demand, grid-connection requirements, network costs, renewable procurement arrangements, and contribution to additional generation. South African operators should demonstrate, megawatt by megawatt, how their energy arrangements affect supply available to the wider economy. Wheeling and private power procurement can support additional generation and improve supply arrangements, but neither automatically creates new national capacity nor removes transmission constraints. Eskom, transmission authorities, and municipalities should disclose the network capacity allocated to major developments and the infrastructure costs associated with connecting them. Global operators should expect their access to scarce electricity to be assessed against transparent, consistent criteria. Private energy security must not become a public capacity constraint.

5. Tie public incentives to realised outcomes. Every tax concession, public infrastructure commitment, or other material investment incentive should have defined performance conditions, reporting obligations, independent verification, and proportionate clawbacks where agreed commitments are not delivered. Employment, domestic procurement, skills development, and additional infrastructure should be measured against specified baselines and delivery periods. Virginia’s reported US$1.94 billion in data-centre tax benefits alongside 1,610 net new jobs illustrates why incentives require closer scrutiny. These figures do not, by themselves, establish the cost of creating each job or the net fiscal return. They do demonstrate the need to assess concessions against realised economic and fiscal outcomes, rather than investment announcements alone. National Treasury should incorporate these tests into any future data-centre incentive framework. Public concessions should purchase measurable public value, not merely the promise of private investment.

6. Build the domestic supplier and skills base before demand peaks. Training investments should be tied to documented occupational requirements, including electricians, mechanical and cooling technicians, network engineers, cybersecurity specialists, and other technical roles required to construct and operate these facilities. South African universities, technical colleges, training institutions, and industry bodies should co-design programmes with operators before major construction phases begin. Global corporations should support structured knowledge transfer, specialist secondments, supplier development, and pathways through which local employees progress into higher-value technical and managerial positions. Training volumes alone are insufficient: outcomes should include occupational competence, employment, retention, and progression. Skills developed after the investment cycle has passed cannot capture the opportunity that preceded them.

7. Make domestic computing access part of the economic bargain. South Africa should pursue commercially viable arrangements through which domestic enterprises, start-ups, universities, research institutions, and public bodies can access appropriate AI computing capacity at competitive, transparent rates. Operators should be encouraged to offer defined capacity allocations and pricing arrangements, supported by partnerships that expand domestic adoption and technical capability. Global corporations should recognise that developing local demand strengthens the wider market in which their infrastructure operates. The objective is not to compel uneconomic access or subsidise consumption indiscriminately. It is to ensure that domestic firms can use the infrastructure being established in their economy to improve productivity, develop applications, and compete in higher-value markets. A country that hosts computing capacity but cannot afford to use it risks exporting the productivity gains while importing the finished products.

8. Publish an independently verifiable national scorecard. Adopt the Bandzishe Group Hyperscale Economic Dividend Framework as a consistent basis for measuring the 12 dimensions of national economic contribution, with project-level reporting and annual publication. The scorecard should distinguish proposed, committed, under-construction, and operational facilities, while separating announced investment from realised expenditure and observed employment from modelled effects. Government should use the evidence to assess incentives, infrastructure allocation, and subsequent project performance. Boards should use it to test investment claims, identify exposure, and hold management accountable for the economic assumptions underpinning major decisions. Data gaps should be disclosed rather than filled with unsupported estimates. The scorecard must inform decisions, not merely describe activity. What is measured can be governed; what is concealed cannot be held accountable.

The eight decisions establish a common standard for investors, operators, public institutions, and corporate boards. They do not oppose foreign capital or seek to diminish the commercial case for data-centre investment. They seek to improve the terms on which that capital contributes to the domestic economy. South Africa must compete for investment without competing away its capacity to secure value from it.

The distinction will determine whether the country becomes a strategic participant in the AI economy or merely a location from which that economy operates. The boardroom test is therefore not how much capital arrives, but how much productive capability, domestic value, and enduring economic advantage it leaves behind.

Strategic Mandate / The Final ReckoningThe Dividend Decides: South Africa Will Be Judged by What It Keeps, Not What It Hosts

The country that attracts investment secures its physical footprint; the country that builds the capabilities around it secures a greater share of its economic value. South Africa has established a credible data-centre presence, but the evidence examined in this briefing exposes a more difficult imbalance: substantial capital deployment, a comparatively small permanent employment base, and insufficiently disclosed evidence of the wider domestic economic multiplier. The infrastructure is visible. The national dividend remains less certain. That is not an argument against investment. It is an indictment of measuring success by what arrives rather than by what the country retains.

Economic growth is not merely a statistic in a country facing structural unemployment; it is the foundation of fiscal resilience, social cohesion, and technological sovereignty. At an official unemployment rate of 33.6 per cent, South Africa cannot mistake a projected 100 permanent jobs for an employment strategy, any more than it can mistake a billion dollars of investment for an economic verdict. The country needs both capital and consequence: global investment on competitive terms, domestic participation on measurable terms. Attracting capital without securing wider economic value produces an impressive footprint with a potentially modest national return. Demanding participation without remaining commercially competitive risks losing the investment altogether. The strategic challenge is to secure both, without surrendering either.

Every board, cabinet, and investment committee confronting a major data-centre proposal must therefore ask a more demanding question: what, precisely, will South Africa retain when the capital has been deployed and the construction crews have left? Audit every investment announcement previously approved or publicly endorsed. Reconcile employment claims to actual payrolls, distinguish temporary construction activity from permanent positions, and separate observed employment from modelled multipliers. Appoint an executive accountable for domestic economic retention, with defined measures, reporting obligations, and delivery dates. Commission independent assessments of employment, procurement, electricity, water, and net fiscal returns before granting further material incentives. Do not leave disclosure requirements until commissioning; by then, much of the negotiating leverage has already been surrendered. The time to secure the dividend is before the investment is locked in, not after the infrastructure is built.

No material agreement should proceed without clearly defined commitments covering permanent employment, genuine domestic value added, skills development, and the project’s contribution to additional energy supply. These commitments must be measurable, independently verifiable, and enforceable where public concessions are involved. Where an outcome cannot reasonably be guaranteed, it must at least be disclosed, monitored, and reviewed. The objective is not to punish capital for being capital-intensive. It is to prevent capital intensity from becoming an excuse for weak domestic participation.

South Africa does not need to choose between being an attractive investment destination and being an economically demanding host. It must become both. The country must compete for the capital, negotiate for the capability, and account for the value. Otherwise, the AI infrastructure boom may leave South Africa with the physical assets of the digital economy while much of its productive advantage accumulates elsewhere.

Measure what stays. Govern what is granted. Secure what comes next.