World Bank Proposes Pay-as-You-Go AI Computing for African Governments

Sub-Saharan Africa has 18% of the world's population but just 0.6% of global data centre capacity. Only 5% of the facilities already available are equipped to handle AI workloads. For governments that want to use artificial intelligence in public services, that leaves an expensive problem: the computing power has to come from somewhere.

The World Bank is proposing that they share it.

Under a new initiative announced on October 20, 2026, governments in developing economies would be able to rent AI computing power from shared infrastructure and pay for what they use, instead of building dedicated facilities of their own.

The programme, called the AI Infrastructure and Readiness Initiative, was unveiled in Seoul alongside the World Bank Group's first digital strategy. Africa and Latin America will be the initial focus, with Cassava Technologies and NVIDIA joining as knowledge partners.

At the centre of the plan are graphics processing units, or GPUs, the specialised chips used to train and run AI models. They are expensive to buy and require significant electricity and cooling when deployed at scale. Under the World Bank's proposed model, a ministry building an AI tool to help agricultural officers identify crop diseases, for example, could rent the computing capacity it needs instead of buying and maintaining its own GPU servers.

World bank's digital strategy places more emphasis on what it calls small AI: systems adapted to a particular problem and local setting. That could mean helping a health worker make sense of medical information or giving farmers a tool that recognises disease from a photograph.

These applications need less computing power than training frontier models from scratch, but they still need somewhere to run. In a region where AI-ready data centre capacity remains scarce, access to the hardware can become part of the cost of building the service itself.

What this could mean for Kenya

Kenya has not been named among the countries that will participate in the World Bank initiative. One of the Bank's partners, however, already operates data centre infrastructure in Nairobi.

Cassava Technologies owns Africa Data Centres, which runs a facility at Sameer Industrial Park, and the company has also been expanding its AI infrastructure business with NVIDIA. Earlier this year, Cassava said it planned to extend its NVIDIA-powered AI Factory to Kenya after an initial deployment in South Africa.

Whether that infrastructure will have anything to do with the World Bank programme is unclear. Cassava and NVIDIA have been named as knowledge partners, but the Bank has not said that either company will supply or operate the computing capacity.

Kenya also offers a useful example of why the economics matter.

In May 2024, Microsoft and Abu Dhabi-based G42 announced a $1 billion investment centred on a geothermal-powered data centre at Olkaria, with plans to support a Microsoft Azure cloud region serving East Africa. Two years later, the project had run into difficulties.

Reuters, citing Bloomberg, reported in May 2026 that negotiations had stalled after Microsoft and G42 sought guarantees that the Kenyan government would buy a certain amount of computing capacity every year. The government could not commit at the requested level. Kenyan officials maintained that the project had not been cancelled.

That disagreement exposes the problem shared computing is supposed to ease. Large data centres are expensive to build, so operators need enough customers using them to recover that investment. Governments, meanwhile, may not want to commit public money to computing capacity they cannot be certain they will need.

A pay-as-you-go model could spread that risk across several users. One government could use more capacity when it needs it and less when it does not, without carrying the cost of an entire facility.

But the infrastructure does not become cheap simply because it is shared. Someone still has to buy the GPUs, power them and keep the facilities running. To work, the whole plan relies on what governments will have to pay.