AMACO Energy Group, a Greek multinational, is pitching an answer to one of the hardest questions facing Africa's artificial intelligence (AI) ambitions: where will all the electricity needed to run the servers come from?

The company's HERCULES project bundles electricity generation, cooling and AI data-centre infrastructure into a single design. The core idea is to reduce dependence on the national grid by generating power close to where it is consumed. The initial configuration runs on liquefied natural gas (LNG), although AMACO says the system can later accommodate renewable energy and hydrogen.

Theodore Theodoropoulos, AMACO's founder and chief executive, is in Nairobi to discuss the proposal with Kenyan officials and potential partners, according to an email sent to TechCabal.

"HERCULES has the potential to transform Mombasa Port and Kenya into a global AI Data Center hub by integrating advanced, independent smart-power generation directly with AI-DC infrastructure," AMACO said in the email. "Designed as a next-generation solution, it offers a highly mobile, scalable, adaptive and energy-efficient platform that can be deployed independently of the conventional electricity grid."

There is plenty left to prove. The project has not been built, and questions remain over financing, customers, approvals, fuel supply and the environmental case for using LNG. Africa has no shortage of ambitious infrastructure proposals that failed to progress beyond announcements.

But the thinking behind HERCULES matters more than whether a single $1.5 billion project succeeds. AI is pulling data centres and electricity systems closer together: global data-centre electricity consumption stood at roughly 415 terawatt-hours in 2024 and could reach 945 TWh by 2030, with AI the main driver of that increase. For countries trying to attract more compute-intensive infrastructure, the constraints are no longer just land and fibre, but also power, cooling and water.

What might Africa learn from AMACO's approach? Five lessons stand out.

Build the power with the data centre

The most useful idea in HERCULES is that anyone planning a very large AI data centre on the continent should work out where its electricity will come from at the same time.

This has not always been necessary. A conventional data centre might require 10MW-25MW. An AI-focused hyperscale facility can require 100MW or more — enough electricity to supply about 100,000 households — and the largest facilities now being planned globally are measured in gigawatts. At that scale, the power supply becomes part of the design decision.

AMACO is testing that logic at its proposed Mombasa data centre. HERCULES places generation alongside the facility and integrates the two systems directly, which the company says avoids reliance on national grids and allows capacity to expand as computing demand increases.

The rest of the continent could learn from this. A government trying to attract 500MW of new data-centre capacity should have an answer for where the electricity will come from; otherwise, the success of one infrastructure policy can create a problem for another.

Timelines matter too. The International Energy Agency (IEA) notes that a data centre can be operational within two or three years, while the electricity infrastructure needed to support one takes considerably longer to plan and build. The agency estimates that grid constraints could delay about 20% of planned data-centre projects globally unless they are addressed. The lesson: plan them together.

Do not assume the grid can absorb everything

AMACO's decision to make HERCULES capable of operating independently of national grids is particularly relevant to Africa.

Kenya's electricity demand reached a record peak of 2,439MW in December 2025, up 151MW from a year earlier. Thermal generation rose 24% in the first half of the 2025/26 financial year as demand increased, while imports from Ethiopia and Uganda rose almost 25%.

Now set a 100MW AI data centre against those numbers: one facility would represent potential demand equivalent to about 4% of Kenya's current national peak.

That does not mean Kenya lacks the electricity to host data centres. Its geothermal resources and high share of renewable generation are considerable advantages. It does suggest, however, that African countries should be wary of treating hyperscale facilities like ordinary commercial facilities.

AMACO's answer is dedicated generation. Other developers may choose different arrangements: governments could require large data-centre projects to finance additional generation or transmission infrastructure, and dedicated renewable plants paired with storage are another possibility. If AI infrastructure brings enormous new electricity demand, it should ideally bring new electricity supply with it.

Ask about the cooling system early

HERCULES does something else differently: it treats cooling as part of the power system.

That is becoming more important as AI chips pack more computing into each server rack. Higher density means more electricity and more heat. AMACO proposes recovering heat and cold within the same system used to generate electricity, and says its Mombasa project would avoid putting additional pressure on local water resources.

That is a consequential claim in Mombasa. The coastal city has long struggled with water shortages and relies heavily on supplies from outside the county. A large industrial user cannot assume water is simply available because it needs it.

Data centres use very different amounts of water depending on their cooling technology, location and design. Some newer facilities use closed-loop liquid cooling systems that substantially reduce direct water consumption; others can consume large quantities. This should be established before a project is approved.

How many litres of water will a facility consume for each kilowatt-hour of computing? Where will the water come from? What happens during shortages? Does the cooling technology require potable water? Can waste heat be put to another use? These questions are easier to answer before construction than after residents and data centres find themselves drawing from the same constrained supply.

Mombasa is part of the idea

The location AMACO has chosen also deserves attention. Most of Kenya's existing data-centre capacity is concentrated around Nairobi, which makes commercial sense: the capital contains the banks, telecom companies, technology businesses and government agencies that buy much of the capacity.

Mombasa offers a different proposition. Several submarine cables connecting Kenya to international internet networks land at the coast, and the city has a major port — which suits AMACO's LNG plans. A large data centre there would sit close to both its proposed fuel supply and international connectivity.

The lesson here has little to do with LNG. African governments tend to talk about technology clusters in terms of cities that already have technology companies, but large AI infrastructure makes energy geography more important. A country could instead ask where electricity and fibre are cheapest and most plentiful, then consider whether computing capacity should follow them.

Kenya's geothermal resources provide one possibility. Ethiopia's hydroelectric resources provide another. Nigeria and Mozambique have gas. Morocco has invested heavily in solar and wind. The location of Africa's next large data-centre clusters does not have to follow the map of its startup ecosystems.

Do not assume LNG is the lesson

HERCULES has a complication: its initial answer to AI's rapidly growing electricity consumption is another fossil fuel.

There is commercial logic to LNG. Gas generation is dispatchable, and data centres need electricity continuously. The IEA estimates that natural gas currently supplies about 26% of data-centre electricity globally and expects gas generation serving data centres to continue growing.

But AMACO is proposing this in Kenya, where the electricity mix makes the case less straightforward. The country has about 1GW of geothermal capacity, and geothermal shares one of the qualities that makes gas attractive to data-centre operators: it can produce electricity continuously. Kenya also gets most of its electricity from renewable sources. An LNG-based system therefore needs to demonstrate an advantage over the alternatives on cost and reliability sufficient to justify imported fuel and higher emissions.

The customer side matters as well. Large technology companies are trying to reduce the carbon footprint of their data centres at the same time as AI is increasing their electricity consumption, so the carbon intensity of a data centre could affect its attractiveness to the companies AMACO hopes will use it.

AMACO says HERCULES can incorporate renewable energy and eventually hydrogen — and that may prove more important than the LNG component itself. The architecture can travel without the fuel. Kenya could pair large computing facilities with geothermal. Ethiopia could explore hydro. Countries with abundant gas may decide that gas makes sense. Solar-rich markets may require a mixture of solar, storage and firm generation. AMACO's contribution is to put the energy choice at the centre of the data-centre decision.

Finally

None of this answers whether AMACO will build HERCULES in Mombasa. That depends on details that have not yet been made public: the $1.5 billion financing is one, and the size of the proposed data centre is another. Customers, electricity costs, LNG requirements, water consumption and the amount of additional generation available to Kenya will all determine whether the project makes commercial sense.

Those details should receive more attention than the headline investment figure. But AMACO is raising these questions at a useful time.