What AMACO’s $1.5bn Kenya AI project can teach Africa about energy

HERCULES is AMACO’s proposed $1.5 billion project in Kenya. Its core idea is to place electricity generation alongside the facilities that use the power. That could reduce reliance on Kenya’s national grid as AI computing expands. The project combines three functions: electricity generation, cooling, and AI data-centre infrastructure.

For example, a HERCULES site could generate electricity on location, send it directly to computer servers, and use an integrated cooling system to remove heat. This arrangement connects supply and demand closely. It may reduce the need to transmit all the required electricity over long distances through the national network.

The initial design uses liquefied natural gas, or LNG. AMACO says the system could later accommodate renewable energy and hydrogen. Therefore, HERCULES is presented as an energy-and-computing platform that could change its fuel mix over time, rather than as a fixed LNG-only design.

An electricity grid moves power from generating plants through transmission and distribution lines to customers. If an AI data center generates electricity next door, it can consume some power directly. That reduces the amount it needs to purchase or receive from Kenya’s national grid. The article identifies this local generation as a way to reduce grid dependence.

For example, LNG equipment at a HERCULES site could produce electricity for nearby servers. The power would travel only a short distance inside the facility or campus. This can avoid some transmission constraints and limit exposure to interruptions elsewhere on the network. It does not necessarily eliminate the need for a grid connection.

The practical result depends on the local plant’s capacity, fuel supply, maintenance, and backup systems. A data center might still use the grid when local generation is unavailable or insufficient. Even so, producing power close to the computers could make their electricity supply more self-contained as AI demand grows.

AI data centers do not have one standard electricity requirement. Individual facilities may use several megawatts, while large AI campuses can require tens or hundreds of megawatts. The exact figure depends on the number of servers, the chips installed, and how continuously they operate. The article does not provide a specific HERCULES demand figure.

AI processors perform enormous numbers of calculations. Nearly all the electricity they use ultimately becomes heat. Cooling equipment moves that heat away from chips and buildings so the machines remain within safe operating temperatures. Air cooling uses fans and chillers; liquid cooling can move heat more efficiently in some high-density systems.

Cooling therefore adds to the facility’s total energy demand rather than being an optional extra. It also requires water or other cooling resources in some designs. This is why HERCULES combines power generation and cooling with AI infrastructure: the site must supply both computing electricity and dependable heat management.

Liquefied natural gas, or LNG, is mainly methane cooled to about minus 162 degrees Celsius. At that temperature, natural gas becomes a liquid and takes up far less space. It can then be transported in specialized ships, trucks, or storage tanks. Before use, it is usually warmed back into gas and burned or processed to produce electricity.

The article says HERCULES initially uses LNG. A likely reason is that gas-fired generators can provide steady, controllable power when computers and cooling systems need it. Unlike sunlight or wind, gas generation can operate on demand, subject to fuel availability and equipment limits. That makes it suitable for a data center requiring continuous power.

LNG is still a fossil fuel. Burning it releases carbon dioxide, and leaks of methane can add climate impact. The article does not state AMACO’s specific reason for selecting LNG. It does say the design could later accommodate renewable energy and hydrogen, suggesting the initial fuel choice may not define the system permanently.

If AI facilities generate electricity locally with LNG, Kenya could gain additional supply near a major electricity user. That may reduce pressure on the national grid and help the facility continue operating during some grid disruptions. However, national reliability would improve only if local plants provide genuine additional capacity rather than simply shifting fuel use away from the grid.

Costs could move in either direction. On-site generation might avoid some grid charges, transmission limits, or outage costs. LNG purchase, transport, storage, generators, maintenance, and cooling systems could also be expensive. The outcome would depend on fuel prices, plant efficiency, financing, and how much electricity the facility consumes. Local generation does not automatically mean cheaper power.

Emissions would still occur because LNG is fossil fuel. It can produce fewer emissions than more carbon-intensive fuels in some comparisons, but combustion releases carbon dioxide. Methane leakage during production and transport is another concern. Renewable energy or hydrogen could reduce those impacts if supplied reliably and at scale.

Renewable energy could replace or supplement LNG through solar, wind, hydropower, or other low-carbon sources connected to the HERCULES system. Hydrogen could also supply a generator or fuel cell, depending on the equipment. AMACO says the project can later accommodate renewable energy and hydrogen, but the article does not specify the technologies, timing, or quantities.

The key challenge is intermittency. Solar power falls at night and varies with clouds. Wind output changes with weather. AI servers and cooling systems, however, generally need electricity continuously. A renewable-heavy design would therefore need batteries, other storage, flexible backup generation, excess renewable capacity, or a connection to the national grid. Hydrogen could potentially provide stored energy when renewable output is low.

Using more renewables could reduce fossil-fuel consumption and operational emissions. Yet reliability and cost would depend on local resources, storage duration, equipment, and fuel availability. Hydrogen’s climate benefit also depends on how it is produced. HERCULES’s proposed flexibility could allow its energy mix to evolve as these technologies become more practical.

An electricity grid is a connected system of generators, high-voltage lines, local distribution networks, and customers. Generators produce electricity, transmission lines move it over long distances, and distribution lines deliver it to homes, businesses, and data centers. Grid operators coordinate these parts so electricity is available when users need it.

Supply and consumption must stay balanced almost instantly. If demand rises, generators or stored power must respond. If generation exceeds demand, system frequency and voltage can move outside safe ranges. For example, an AI data center starting many servers could create a large new load. Operators would need enough generation, reserves, or demand controls to cover it.

An imbalance can trigger protective equipment, outages, or wider instability. Batteries and other storage help, but they do not remove the need for careful coordination. Local HERCULES generation could make an AI facility less dependent on grid supply. It would still need its own controls and reliable equipment to match generation with the servers’ changing demand.