By August 2026, the proposed Olkaria data centre site is quiet: no cranes, no servers, just geothermal steam rising over a deal that has not been closed.
Two years ago, Kenya announced what was billed as the launchpad for its hyperscale future: a $1 billion data centre campus built by Microsoft and UAE-based AI company G42. The first phase would deliver 100 megawatts of capacity and anchor a new East Africa Cloud Region for Microsoft Azure.
By May 2026, the project had stalled.
Kenya’s Principal Secretary for ICT, John Tanui, told Business Daily that the project risked delay over the company’s request for guaranteed payments, saying there were “some guarantees that the government cannot give” because certain commitments are determined by market forces such as procurement and investment.
Tanui maintained that talks had not collapsed and were still ongoing, and that missing an initial deadline should not be read as a sign the project had failed.
This matters beyond Kenya. It is a live test of how African governments finance digital infrastructure and of what happens when they are asked to underwrite demand that has not yet materialised.
Analysts say the original proposition was strong on paper: a data centre campus at Olkaria, near Kenya’s geothermal fields, designed to run entirely on renewable geothermal power. Olkaria’s geothermal complex generates roughly 950 MW, against phase one of the project’s 100 MW requirement.
Industry reliability standards for data centres go beyond the power source itself. The Uptime Institute’s tier classification system, the international benchmark for data centre performance, requires redundant power paths and backup systems so operations continue even if one component fails, with many mission-critical facilities targeting 99.999% uptime.
Kenya’s case for Olkaria has centred on the source of the primary electricity; the reliability standards a data centre is held to cover the full system built around it.
Tanui’s comments confirm that guarantees were the sticking point and that some of what was being asked for fell outside what the government could commit to, given that certain outcomes are shaped by market forces rather than state control. He has not detailed the mechanics of what was requested.
Government-backed capacity guarantees are a recognised structure in infrastructure financing, most familiar from power purchase agreements, where a state utility commits to paying for generation capacity regardless of how much electricity it actually draws, an arrangement known as take-or-pay. Applied to a data centre, the same logic has government paying for a minimum level of capacity over a long contract term, whether or not it is fully used.
The IMF and World Bank classify this kind of arrangement as a contingent liability: a government obligation that depends on a specific outcome, such as usage falling short of an agreed minimum, rather than a scheduled fixed payment.
Their joint guidance on public debt management notes that such guarantees shift risk from the private investor to the state and that unfunded contingent liabilities have been a recurring factor in fiscal crises. World Bank Treasury guidance adds that foreign-currency-denominated obligations carry a further source of uncertainty, since the local-currency cost of repayment moves with the exchange rate.
Moody’s has flagged technological obsolescence as a growing risk across the data centre sector, noting that fast-moving advances in chip design, cooling and architecture increase the chance that a facility built for today’s workloads becomes mismatched to demand within years.
None of this confirms what Kenya was actually asked to guarantee. IMF and World Bank guidance on government guarantees describes this general profile of exposure, demand risk shifted onto the state and currency exposure on foreign-denominated terms, as typical of such arrangements.
Speaking at State House, Nairobi in April 2026, President William Ruto said the full project would have required around 1,000 MW against a national grid he put at close to 3,000 MW, meaning the facility “would have meant switching off half of Kenya.” He cited the episode as the moment he recognised the country’s power constraints, calling for national generation capacity to scale to 10,000 MW. Kenya’s interconnected installed capacity stood at 3,192 MW as of June 2025, according to EPRA’s Energy and Petroleum Statistics Report, close to the figure Ruto cited.
Kenya’s Special Technology Envoy, Ambassador Philip Thigo, MBS , clarified this, arguing the President’s point was about the scale of investment needed in generation capacity, not a signal that the project had been suspended.
That exchange reframes the scale problem. Kenya’s geothermal capacity is real, but a full build was never going to run on Olkaria’s output alone; it would draw on the national grid at a level that competes directly with other demand. Phase one, at 100 MW, is a materially smaller ask than the 1,000 MW full build, which is part of why the guarantee question not the power question alone, is what stalled talks.
Locating servers inside Kenya is not the same as controlling the digital infrastructure stack. A Kenya-based facility can still run on foreign-owned software, hardware, financing and cloud platforms, with pricing, access and security standards set elsewhere.
The relevant question for policymakers is not whether the servers sit inside Kenyan borders. It is whether the deal builds local technical capability, secures meaningful governance rights, and captures enough economic value to justify the public commitments attached to it. A data residency address is not, on its own, digital sovereignty.
Kenya is negotiating this in a market where South Africa already has an established data centre ecosystem and Nigeria offers scale and rising cloud demand. Other African governments are courting the same hyperscale investment. That competition creates an incentive for governments to offer larger incentives, subsidised power, land, tax terms, payment guarantees to avoid losing a project to a neighbour.
Tanui described talks as ongoing, not collapsed, with some guarantees falling outside what the government can offer. Several structures could address that gap without requiring government to match the original terms in full.
A phased build would start at or below the already-agreed 100 MW and expand against verified customer demand and grid capacity, rather than committing to the full 1,000 MW upfront. A portfolio of anchor customers, public agencies, banks, telcos, and universities could replace a single government guarantee, creating baseline demand without making the Treasury the sole buyer of capacity.
Partial guarantees from development finance institutions could share the exposure rather than placing it entirely on Kenya’s sovereign balance sheet, while adding independent scrutiny to the deal terms.
Renegotiating such a contract partway through its term could mean either continued payments at the agreed level or an exit cost. Data centre technology, including cooling systems, chip architecture and AI workload design, changes quickly enough that a facility sized to today’s projections can be mismatched to demand a decade out.
Experts agree that Africa needs more cloud capacity, stronger data governance and computing infrastructure that can support local businesses, government services and AI applications built for African markets.
Digital infrastructure carries the same fiscal weight as other capital-intensive projects. It draws on electricity, requires long-term capital commitments, and creates contractual obligations that outlast the news cycle around the announcement.”
What happens next depends on which path Kenya and its partners take: a guarantee structure the government can commit to at full scale, a smaller phased build that shares the risk, or continued delay while other markets build the capacity Kenya had hoped to host.