Green Data Centers in Africa: Inside the Dakhla 500MW Project

Dakhla: Where Geography Meets AI Infrastructure

Dakhla sits at 23 degrees north latitude on the Atlantic coast of Southern Morocco, a position that makes it one of the most naturally advantaged locations on Earth for green data center development. The region receives over 3,000 hours of sunshine per year — among the highest solar irradiance levels globally — and benefits from consistent Atlantic trade winds that average 7-9 meters per second across the Dakhla wind corridor. These are not marginal advantages: they represent a renewable energy resource that is both abundant and predictable, enabling power generation profiles that align closely with data center baseload requirements. The combination of solar peak during midday hours and wind peak during evening and overnight hours creates a complementary generation profile that covers the majority of the 24-hour cycle without storage.

The city's location also provides strategic connectivity. Dakhla lies on the route of the Africa Coast to Europe (ACE) submarine cable system and the South Atlantic Express, providing direct fiber links to Europe and West Africa. New cable projects currently in deployment will add additional capacity and redundancy, positioning Dakhla as a connectivity hub for AI workloads serving both African and European markets. Round-trip latency to Madrid is approximately 28 milliseconds via the Casablanca landing station — fast enough for latency-sensitive inference workloads and well within the requirements for training and batch processing.

The 500MW Capacity: What It Means at Scale

At 500 megawatts of IT load capacity, the Dakhla data center will be the largest AI compute facility on the African continent and one of the largest purpose-built AI training centers globally. To contextualize this scale: 500 MW of IT load can support approximately 500,000 NVIDIA H100-equivalent GPUs operating simultaneously, assuming a rack density of 100 kW per rack and standard H100 power profiles. This is sufficient compute to train multiple frontier-scale language models in parallel, or to serve billions of inference requests per day across the African and European markets. The facility is designed around high-density GPU computing from the ground up — not retrofitted from traditional enterprise data center designs — with rack densities optimized for AI accelerators rather than general-purpose servers.

The workload mix is designed to serve three categories. First, sovereign AI training: large-scale model training for African governments and enterprises that require data to remain within Moroccan jurisdiction. Second, European inference overflow: low-latency inference serving for European AI applications, leveraging Morocco's sub-30ms connectivity to major European peering points. Third, carbon-optimized training: workloads from global AI companies seeking to reduce their carbon footprint by training models on one of the cleanest grids available at scale. HarchOS's carbon-aware scheduling engine routes workloads across these categories in real time, maximizing GPU utilization while minimizing carbon intensity.

Renewable Energy Architecture

Dakhla's power architecture is built on three renewable pillars. Solar photovoltaic capacity is sourced from the Noor complex extension — a dedicated solar installation that adds to Morocco's existing concentrated solar power infrastructure at nearby Ouarzazate. At Dakhla's latitude, solar PV achieves capacity factors of 22-25%, significantly higher than the 15-18% typical in Northern Europe. Wind power comes from the Dakhla wind corridor, where Atlantic trade winds drive capacity factors of 35-45% — among the highest onshore wind potentials in the world. The third pillar, currently in feasibility study, is green hydrogen produced from surplus renewable generation during peak solar and wind hours, which would provide long-duration energy storage to cover the remaining gaps in the renewable generation profile.

The power purchase structure is designed to ensure that the facility's location-based carbon intensity remains at or below 50 gCO2/kWh — consistent with Harch Intelligence's current fleet average of 47 gCO2/kWh. This is achieved through a combination of on-site generation, dedicated PPAs with nearby renewable installations, and grid-supplied power from Morocco's 81.5% renewable national grid. Unlike data centers that claim low carbon intensity through purchased RECs from distant generators, Dakhla's low carbon intensity reflects the physical reality of the local generation mix.

Cooling Innovation in a Desert-Adjacent Climate

Data center cooling is one of the largest contributors to energy consumption and water use, and Dakhla's design addresses both challenges through innovative approaches. Despite its latitude, Dakhla benefits from moderate temperatures due to the Atlantic Ocean's thermal regulation — the city's average high temperature is 25 degrees Celsius, with nighttime temperatures regularly dropping below 18 degrees. This enables economizer-mode cooling for approximately 60-70% of the year, where ambient air — filtered for dust and humidity — directly cools the data center without mechanical refrigeration. During the remaining 30-40% of hours, indirect evaporative cooling supplements the economizer, reducing the cooling energy penalty to a fraction of what traditional compressor-based systems require.

For high-density GPU clusters, where rack-level heat dissipation can exceed 100 kW, Dakhla deploys direct liquid cooling using cold plates mounted on GPU processors. Liquid cooling is approximately 3,000 times more effective than air at transferring heat, enabling higher rack densities with lower fan energy consumption. The warm water returned from the cold plates is routed through dry coolers on the facility roof, where Dakhla's consistent winds provide free heat rejection without evaporative water loss. This hybrid cooling architecture — free air economization for general compute, liquid cooling for GPU clusters, and dry coolers for heat rejection — minimizes both energy and water consumption.

Water Usage Effectiveness in a Water-Scarce Region

Water usage effectiveness (WUE), measured in liters of water per kilowatt-hour of IT load, is a critical metric for data centers in arid regions. Dakhla receives less than 40 mm of rainfall per year, making water conservation an existential design requirement. The facility targets a WUE of less than 0.5 L/kWh — approximately 80% lower than the industry average of 1.8 L/kWh — by eliminating evaporative cooling towers from the primary cooling loop. Traditional data centers use cooling towers that evaporate water to reject heat, consuming millions of liters per year. Dakhla's dry cooler and liquid cooling architecture avoids this entirely, using water only for humidity control and domestic purposes. This design decision adds capital cost — dry coolers require more surface area than cooling towers — but eliminates the operational and environmental risk of large-scale water consumption in a water-stressed region.

Construction Timeline and Phased Deployment

The Dakhla project is structured in three phases. Phase 1 delivers 100 MW of IT load capacity with supporting renewable energy infrastructure, establishing operational capability and connectivity. Phase 2 expands to 300 MW with additional GPU halls and enhanced renewable generation. Phase 3 reaches the full 500 MW design capacity with completed submarine cable connections and full green hydrogen integration. This phased approach allows the facility to begin generating revenue and serving customers while construction continues, and provides checkpoints for incorporating technology improvements — such as next-generation GPU architectures and cooling systems — as they become available during the multi-year build-out.

Global Comparison: Dakhla in Context

The Dakhla 500MW project joins a small number of purpose-built AI training facilities at this scale worldwide. Microsoft's data center campus in Mount Pleasant, Wisconsin, built for OpenAI, operates at a similar scale but runs on a grid with approximately 400 gCO2/kWh location-based carbon intensity — roughly eight times Dakhla's. Amazon's data center clusters in Ashburn, Virginia, the largest concentration of data center capacity on Earth, operate on a mixed grid at approximately 380 gCO2/kWh. The comparable European facilities — such as the Mega Data Center campus in Hamina, Finland, running on Nordic hydro and nuclear at approximately 50 gCO2/kWh — achieve similar carbon intensity but lack Dakhla's solar and wind growth potential. Dakhla's unique proposition is the combination of frontier-scale capacity, sub-50 gCO2/kWh carbon intensity, direct European connectivity, and room for expansion in a region with unlimited renewable energy headroom.

Economic Impact and Skills Development

The Dakhla project generates economic value that extends far beyond the data center campus. Construction phases create an estimated 2,500 direct jobs in building trades, electrical engineering, and project management. Once operational, the facility employs approximately 400 permanent staff in data center operations, network engineering, and AI infrastructure management. Harch Corp's training programs, developed in partnership with Moroccan engineering schools, prepare local graduates for these roles — building a talent pipeline that supports not just Dakhla but Morocco's broader AI infrastructure ambitions. The local supply chain benefits from recurring demand for electrical maintenance, security, catering, and transportation services, creating a multiplier effect that extends the economic impact across the Dakhla-Oued Ed-Dahab region.

Why Africa Needs This Now

AI adoption across Africa is accelerating at a pace that outstrips the continent's existing infrastructure. Mobile money platforms process hundreds of millions of transactions daily and require real-time fraud detection. Agricultural AI models serving smallholder farmers across West and East Africa need low-latency inference capacity. Natural language processing for Amazigh, Wolof, Swahili, and hundreds of other African languages requires training compute that does not exist on the continent today. Every one of these workloads currently traverses a submarine cable to a European or American data center, adding latency, cost, and data sovereignty risk. The Dakhla 500MW project — integrated with HarchOS's carbon-aware scheduling and connected through Morocco's submarine cable hub — is the infrastructure that brings this compute home. It is not just a data center. It is the foundation of Africa's sovereign AI capability, built where the sun and wind make it possible to run it cleanly, connected where the cables make it possible to serve it fast, and scaled where the demand makes it necessary to build it now.