Bloomberg broke the story on May 10 that Microsoft and its UAE partner G42 have stalled the planned $1 billion Azure data center campus in Kenya after the Kenyan government declined to commit to a minimum annual capacity payment at the level the partners were demanding. The project was first announced in May 2024 and was structured around geothermal baseload power and an East African cloud region for Azure. The full-scale design would have drawn power equivalent to a mid-sized city.
John Tanui, principal secretary at Kenya's Ministry of Information, told reporters the talks have not collapsed and the project is "not failed or withdrawn." Negotiations continue on a smaller 60 MW project with local developer EcoCloud as a fallback. But the original scope is now uncertain, and the public framing around the project as a marquee Africa expansion has shifted to a much more careful set of qualifications about what actually gets built and when.
The reason this stall matters to the cooling industry is that the Kenya project was supposed to validate a specific architectural thesis. Hyperscalers have spent five years searching for sites that combine three rare attributes: stable baseload renewable power, naturally cool ambient conditions for free cooling, and a host government willing to accept a multi-decade industrial footprint. Kenya's Rift Valley geothermal corridor scored on all three. The site sat on the Kenya Electricity Generating Company's geothermal complex around Olkaria, where steam fields supply 24-hour baseload at competitive rates and elevation moderates ambient temperatures into a band where evaporative cooling is rarely required.
The thermal architecture that this combination unlocks is materially more efficient than what hyperscalers can build in equivalent regions. Free cooling availability at altitude means lower mechanical refrigeration load. Geothermal baseload eliminates the seasonal renewable intermittency that forces gas peaker hedges elsewhere. The cooling water profile drops because evaporative tower draw shrinks when ambient wet-bulb temperatures stay low year-round. The site was a thermal showcase waiting to happen.
The capacity payment guarantee that Microsoft and G42 were seeking is a standard hyperscaler procurement structure. The partner asks the host country or its utility to commit to paying for a baseline amount of compute capacity annually, regardless of whether Microsoft sells through that capacity to enterprise customers. The economics of building a $1 billion campus require the partner to pre-sell enough demand to underwrite the construction debt. If the host government will not be the underwriter, the project does not pencil at the announced scale.
Kenya, per Bloomberg, could not provide guarantees at the level requested. The reasons are not hard to read. A mid-sized-city power draw committed to a single industrial customer carries currency risk, sovereign exposure, and grid stability implications that smaller economies are unwilling to absorb without a backstop they do not have. The deal structure that works for Northern Virginia or rural Iowa does not transfer cleanly to East Africa.
The Kenya project was a leading indicator for the broader Africa hyperscale buildout. Cooling vendors had been actively building Africa pipelines on the assumption that the next wave of geothermal-powered, free-cooled hyperscale sites would land across the East African Rift. Several vendors had quietly moved engineering capacity into Nairobi to support exactly the type of facility Microsoft was scoping.
The stall does not kill that pipeline. It does push it out. A 60 MW EcoCloud-led fallback at a fraction of the original scope is a different opportunity than a flagship $1 billion Azure region. The thermal architecture for a 60 MW project skews toward more conventional CRAH and direct-to-chip hybrid designs. The free-cooling, geothermal-anchored, no-evaporative-tower campus that the original Kenya plan called for needs a partner willing to underwrite the capacity, and that partner is not currently visible in Nairobi.
The longer pattern is clear. Hyperscalers want exotic sites with rare combinations of attributes. Host governments want investment without sovereign exposure. The negotiation friction sits at exactly that interface, and cooling vendors with Africa-deployed expertise need to plan for project timelines that are longer and more uncertain than the announcement language suggests.