SINENHLANHLA ZULU | The deep physical opportunities of the data economy

A data centre is, at its simplest, a building filled with technological equipment designed to store, process and distribute data. It is the physical backbone of the digital economy. Every e-mail you send, every video you stream, every transaction you make online passes through a data centre somewhere.

But AI fundamentally changes the equation. A single ChatGPT query uses roughly 10 times the energy of a Google search. AI does not just add to digital growth; it multiplies the infrastructure requirement per unit of activity.

That’s because AI does not simply retrieve stored information the way a search engine does. It generates new content, processes vast data sets and performs real-time inferences — complex calculations that happen instantly every time someone interacts with an AI product such as ChatGPT or other large language models (LLMs).

This is technically different from older forms of computing. When you stream a video, a file is sent to your device. When you use AI, a massive computation happens live in a data centre, drawing significant electricity and generating heat that must be managed.

AI also creates data continuously, and that data needs to be stored somewhere. More importantly, it needs processing power at a scale and speed that older digital infrastructure was not designed for.

The result? Data centres are no longer just technology assets. They are physical infrastructure sitting at the intersection of real estate, communications and electricity — three of the oldest and most capital-intensive industries in the world. The biggest misconception is that data centres are a tech investment. They are not. They are infrastructure investments with tech tenants, and that distinction matters enormously for how we think about risk, return, duration and regulation.

The dominant narrative is that AI data centres threaten electricity grids by drawing too much power too fast. The missed opportunity is that a flexibly designed data centre can actually stabilise a grid

There is understandable scepticism around AI, given the scale of the disruption it is causing and is expected to cause. But the simplest way to test this belief is to look at capital allocation. Microsoft, Google, Meta and Amazon are collectively committing hundreds of billions, if not trillions, of dollars to AI-driven infrastructure over the next decade. Pension funds and sovereign wealth funds, the most cautious pools of capital in the world, are allocating materially to data centres.

The data centre growth story closely resembles the electrification of the early 20th century. Electricity was invented in the 1870s but did not meaningfully reshape the economy until the 1920s, nearly 50 years later. The reason electricity became so important is because it touched every single industry — from factories and homes to communications and transport. AI is following a similar trajectory. Today most industries rely on the cloud or the internet. AI is becoming a second foundational layer underpinning virtually every sector.

Utilities across the US, Europe and Asia are expanding electricity grids specifically to power AI-driven data centres. Chip manufacturers are building factories that cost tens of billions of dollars. Governments including those of the US, EU and China are embedding AI into national strategy and regulation.

Data centres are location-dependent. The right sites must meet a demanding set of conditions: affordable and reliable electricity (increasingly linked to renewable energy), high-capacity fibre connectivity, sufficient water access and planning permissions.

Energy demand dominates most conversations, but water is equally central. Older data centres relied heavily on evaporative cooling systems, in some cases consuming millions of litres of water per day. A modest 1MW data centre using traditional cooling can consume roughly 20Ml-30Ml of water per year. At the hyperscale end, public disclosures show that a single data centre campus in Iowa consumed approximately 3.8-billion litres of water in 2024 alone. Critically, studies estimate that around 70%–80% of cooling water drawn by traditional systems is lost to evaporation.

In water-scarce countries like South Africa, this deserves more attention than it currently receives. But the industry is evolving fast.

Modern facilities increasingly rely on closed-loop liquid cooling, air-based systems and site selection in cooler climates to reduce water intensity. Microsoft, for example, has committed to eliminating evaporative water use in its next-generation AI data centres, targeting a roughly 95% reduction in cooling-related water consumption — equivalent to several billion litres of water savings annually across its global fleet.

The dominant narrative is that AI data centres threaten electricity grids by drawing too much power too fast. The missed opportunity is that a flexibly designed data centre can actually stabilise a grid.

The single most important policy shift needed is reclassifying data centres as critical national infrastructure, not simply warehouses full of servers

Two under-appreciated concepts matter here: “behind-the-meter power” and “flex power”:

• “Behind-the-meter” means a data centre generates some of its own power on-site through solar, batteries, gas turbines or potentially in the future even small modular nuclear reactors. Rather than drawing everything from the public grid, the data centre generates its own power literally on site or behind the “meter” — the point where the grid connects to the building.
• “Flex power” or “flexible demand” means designing facilities that can ramp their electricity usage up or down in response to grid and electricity supply conditions. When national wind and solar power plants generate excess supply, data centres can absorb that surplus. When the grid is stressed, they can curtail the load. Utilities increasingly need this kind of productive flexible demand, and data centres capable of offering flexibility can become partners to grid operators, not problems.

This points to a missed strategic opportunity: AI infrastructure can actually accelerate the energy transition. It creates a large, credit-worthy customer for renewable energy at scale and in substitution for old tax and tariff incentives provided by state entities. We traditionally think in terms of flexible supply to better manage intermittency. However, the flexible demand from data centres is as useful, if not more so given its rarity.

Another misconception is that a mature AI ecosystem will require less infrastructure. In reality, as AI becomes woven into the fabric of daily life, demand will only accelerate, with more users, more queries and more real-time processing to support.

For investors, capital is moving towards what sits beneath it all: fibre, power, land and water. These are the deep physical opportunities of the data economy.

The single most important policy shift needed is reclassifying data centres as critical national infrastructure, not simply warehouses full of servers. That reclassification would unlock faster permitting, dedicated planning zones and long-term investment certainty.

From there, governments can predesignate infrastructure zones with power availability and environmental conditions already defined, giving investors certainty while protecting communities. Sustainability standards should be mandated upfront: closed-loop cooling, water efficiency thresholds and renewable energy thresholds.

We are in the midst of one of the most significant infrastructure build-outs of our generation. The AI revolution is already being poured into concrete, steel and transmission lines.

For Africa the question is simple and urgent: who will own the ground on which the next century of African economic activity runs? The only real choice is whether we build this infrastructure ourselves, or whether it is quietly built far away from us and the fruits of such investment are imported at a premium.

• Zulu is an investment principal at African Infrastructure Investment Managers