Artificial intelligence is usually discussed as software, but its limits are increasingly physical. Every model depends on processors housed in data centres, continuous electricity, cooling systems, fibre connections and engineers who can keep the equipment running.
Solomon Asamoah sees this as a defining infrastructure question for Ghana and Africa: how can the continent expand compute without allowing a new class of industrial demand to compete carelessly with homes, hospitals and businesses for constrained power and water?
The answer cannot be to avoid data centres. Depending entirely on distant infrastructure can make services slower, expose businesses and governments to foreign-currency costs, complicate decisions about where sensitive data is stored and leave local firms at the edge of the AI economy.
Yet approving capacity wherever a developer proposes it would also be costly. Computing capacity is valuable only when the supporting system is reliable, affordable and accepted by the public.
The constraint is already visible
The International Energy Agency reported in 2025 that electricity consumption by data centres worldwide was on course to more than double by 2030, reaching about 945 terawatt-hours.
The agency’s 2026 update found that data-centre electricity demand had risen 17 per cent in 2025, with AI-focused facilities growing faster still.
Most of that expansion remains concentrated in advanced economies and China, but the engineering lesson travels: large, clustered loads can arrive faster than grids, transformers and generation projects can be built.
Africa begins from a different baseline. The World Bank reported in June 2026 that more than 560 million people in sub-Saharan Africa still lacked electricity. An IMF regional assessment put electricity access at about 53 per cent and internet access at 38 per cent.
In that context, a data-centre connection is not merely another commercial load. It can affect the timing and cost of wider electrification, especially if network investment is not planned alongside it.
Water adds a second constraint. Servers produce heat, and many cooling designs consume water through evaporation. Usage varies enormously with climate, equipment, workload and cooling technology, so a single headline figure can mislead.
The relevant question for regulators is local: how much potable or non-potable water will a proposed facility use during the hottest and driest periods, where will it come from, and who bears the risk when supply tightens?
Start with resource budgets, not ribbon-cuttings
A credible approval process should require developers to show how much power they will use each hour, how backup generation will work, how much water they expect to need, which cooling system they will use and how the site may expand.
Those plans should be tested against drought, heat and grid outages. Renewable-energy promises must match the hours when the facility actually uses electricity, not just a yearly total.
For Solomon Asamoah, this is familiar infrastructure discipline applied to a new asset class. A project’s private return cannot be separated from the public systems it uses.
If a facility requires a substation, transmission reinforcement or dedicated generation, the financing plan should identify who pays, who owns the new assets and whether other customers benefit.
Concessions on land, tax or power tariffs should be valued transparently against jobs, local procurement, network investment and the strategic value of the capacity created.
The objective is not to burden every project with impossible conditions. It is to distinguish between investment that strengthens the system and investment that quietly transfers costs to it.
Clear rules can make Ghana more attractive because serious developers can price requirements early, while utilities and communities can see what has been promised.
Design for African conditions
Cooling should be a design decision, not an afterthought. Air-cooled and closed-loop systems can reduce water withdrawals, though they may use more electricity or carry higher capital costs in particular climates.
Reclaimed wastewater can be appropriate where treatment and pipelines are dependable. Direct-to-chip liquid cooling can remove heat efficiently from high-density equipment, but it demands specialist operation and careful supply chains.
The best option will vary by location; the governing principle is to optimise power, water, resilience and cost together.
Location is equally important. Building close to strong fibre routes but far from a robust grid solves only half the problem. Coastal sites may offer cable access but face corrosion, flood or heat risks.
Inland locations may have land and renewable resources but require new connections. Planners need maps that combine grid capacity, possible power sources, water stress, climate hazards, transport access and the speed of digital connections.
Publishing non-sensitive versions of those maps would improve competition and steer projects towards places where they add value.
Efficiency standards can keep the market honest. Facilities should disclose how much total power and water they use compared with the computing service they provide, as well as how often the service is available.
These figures should follow common definitions and be explained clearly rather than reduced to marketing claims. An efficient building running on an unreliable or heavily polluting power supply may still create serious costs for the wider system.
Not every workload must live in the same place
The choice is not between a giant data-centre campus in every country and permanent dependence on overseas cloud services. A mix of local and regional capacity is more realistic. Sensitive public records and services that need a very fast response may justify hosting within the country.
Routine business applications can use regional cloud capacity. Universities and start-ups can share powerful specialist computers. Smaller local facilities can support services that need a quick response without carrying the cost of a giant centre.
Regional power and digital markets make this model stronger. The West African Power Pool has connected 15 countries with more than 4,000 kilometres of high-voltage transmission lines, according to the World Bank, and cross-border trade now accounts for about 8 per cent of regional electricity.
Digital infrastructure can follow the same logic: countries pool scale, locate major loads where resources are strongest and negotiate common standards for security, data movement and service continuity.
Buying together can also improve access to scarce computing capacity. The IMF has identified shared data centres, pooled access to advanced computer chips and joint cloud purchasing as options for sub-Saharan Africa.
Combining demand from universities, public agencies and smaller businesses can create a reliable first customer while preventing one institution from carrying the full cost.
Finance the surrounding system
Data-centre investment is often described as a property or technology deal. In Africa it should be treated as a group of connected infrastructure projects. Financing may need to cover power generation, batteries, grid improvements, fibre, water treatment and skills, as well as the building and servers.
Each part has different risks and a different useful life, so forcing everything into one financing package can raise costs unnecessarily.
Solomon Asamoah has spent more than three decades structuring investment across developed and emerging markets. His senior roles have included the International Finance Corporation, the Development Bank of Southern Africa, the Africa Finance Corporation and the African Development Bank.
As the first substantive chief executive of the Ghana Infrastructure Investment Fund, he oversaw investment across multiple infrastructure sectors and worked to bring external capital alongside public funds.
That experience points towards using different sources of finance with clear rules. Pension funds and other long-term investors can support stable network assets. Commercial banks can finance facilities with dependable customer contracts.
Climate finance may support efficient cooling, renewable power and batteries where the environmental benefit can be measured. Public support should depend on agreed delivery stages and clear evidence that the promised benefits have been achieved.
Build capacity without creating scarcity
Ghana and its neighbours have a genuine chance to host more of the infrastructure on which their digital economies depend. They also have the chance to avoid mistakes now confronting busier markets, where long waits for grid connections, local opposition and water concerns have slowed projects.
The region can set the terms earlier through clear reporting on resource use, tests of the wider public benefit, efficient design and regional planning.
The strongest AI infrastructure policy will not count data centres alone. It will count reliable power added, water conserved, engineers trained, local services improved and customers connected beyond the facility fence.
Solomon Asamoah’s central argument is that Africa need not choose between joining the AI economy and meeting basic infrastructure needs.
With careful preparation and financing, demand from data centres can help expand the wider system, but only if power and water are treated as core investment duties rather than hidden inputs.
