A Nairobi startup CTO compared three cloud pricing quotes in late 2024. AWS ap-southeast-1 in Singapore. AWS eu-west-1 in Ireland. AWS af-south-1 in Cape Town. The Cape Town region ran measurably more expensive per compute-hour than either alternative. Three subsea cables had been active in African waters for over a year.
The cables did not move the price.
The reason is that bandwidth was not the bottleneck. Data centres are.
Subsea cable capacity delivers bandwidth to a landing station. It does not build a data centre on the other end. Equiano, PEACE, and 2Africa together deliver more bandwidth than all previous cables combined. Their full impact is mapped in Africa's digital infrastructure shift. What they did not deliver was server capacity to process that bandwidth at scale.
The data centre pipeline is moving. Most of the capacity is not yet online. The gap between cable capacity and local facility supply is still measured in years, not months.
AWS Africa (Cape Town) launched in 2022 as the continent's first hyperscale cloud region. It remains the only one. A single provider with a single African region does not produce competitive pricing. The Cape Town premium is structural, not accidental.
That structure is changing from two directions. New data centre capacity is coming online in Lagos and Nairobi - markets with no hyperscale region today. When a hyperscale provider builds into a Lagos or Nairobi facility, Cape Town loses its monopoly position. Competition compresses pricing.
The new cables delivered on latency. Lagos and Nairobi users experience measurably faster connections to European services than in 2022. They did not deliver on cost - because cost depends on supply, not on wire.
For founders making infrastructure decisions in 2026, the timeline is specific. The current premium reflects a supply gap with a defined close date. A founder building on AWS Cape Town today carries a premium that will not exist in 24 to 36 months.
That is not a reason to wait. It is a reason to architect for portability. That same pricing structure is what makes LLM inference disproportionately expensive for African AI teams today. The workloads most sensitive to cloud cost gain the most from local latency improvements. High-compute training, large storage, and video processing all benefit from 20-millisecond round-trip reductions. Building them on African cloud today pays for latency. Building them outside Africa avoids the cost premium but absorbs that latency. Neither is obviously wrong.
The data centre gap made the cloud expensive. The data centre pipeline will close it. The cable is already there.
